Add GLIGEN Text Image implementation (#4777)

* Add GLIGEN Text Image implementation

* add style transfer from image

* fix check_repository_consistency

* add convert script GLIGEN model to Diffusers

* rename attention type

* fix style code

* remove PositionNetTextImage

* Revert "fix check_repository_consistency"

This reverts commit 15f098c96e00bb9e67b831161615b30a2d28d815.

* change attention type name

* update docs for GLIGEN

* change examples with hf-document-image

* fix style

* add CLIPImageProjection for GLIGEN

* Add new encode_prompt, load project matrix in pipe init

* move CLIPImageProjection to stable_diffusion

* add comment

docs/source/en/api/pipelines/stable_diffusion/gligen.md

@@ -12,7 +12,7 @@ specific language governing permissions and limitations under the License.
 
 # GLIGEN (Grounded Language-to-Image Generation)
 
-The GLIGEN model was created by researchers and engineers from [University of Wisconsin-Madison, Columbia University, and Microsoft](https://github.com/gligen/GLIGEN). The [`StableDiffusionGLIGENPipeline`] can generate photorealistic images conditioned on grounding inputs. Along with text and bounding boxes, if input images are given, this pipeline can insert objects described by text at the region defined by bounding boxes. Otherwise, it'll generate an image described by the caption/prompt and insert objects described by text at the region defined by bounding boxes. It's trained on COCO2014D and COCO2014CD datasets, and the model uses a frozen CLIP ViT-L/14 text encoder to condition itself on grounding inputs.
+The GLIGEN model was created by researchers and engineers from [University of Wisconsin-Madison, Columbia University, and Microsoft](https://github.com/gligen/GLIGEN). The [`StableDiffusionGLIGENPipeline`] and [`StableDiffusionGLIGENTextImagePipeline`] can generate photorealistic images conditioned on grounding inputs. Along with text and bounding boxes with [`StableDiffusionGLIGENPipeline`], if input images are given, [`StableDiffusionGLIGENTextImagePipeline`] can insert objects described by text at the region defined by bounding boxes. Otherwise, it'll generate an image described by the caption/prompt and insert objects described by text at the region defined by bounding boxes. It's trained on COCO2014D and COCO2014CD datasets, and the model uses a frozen CLIP ViT-L/14 text encoder to condition itself on grounding inputs.
 
 The abstract from the [paper](https://huggingface.co/papers/2301.07093) is:
 
@@ -26,7 +26,7 @@ If you want to use one of the official checkpoints for a task, explore the [glig
 
 </Tip>
 
-This pipeline was contributed by [Nikhil Gajendrakumar](https://github.com/nikhil-masterful).
+[`StableDiffusionGLIGENPipeline`] was contributed by [Nikhil Gajendrakumar](https://github.com/nikhil-masterful) and [`StableDiffusionGLIGENTextImagePipeline`] was contributed by [Nguyễn Công Tú Anh](https://github.com/tuanh123789).
 
 ## StableDiffusionGLIGENPipeline
 
@@ -41,6 +41,19 @@ This pipeline was contributed by [Nikhil Gajendrakumar](https://github.com/nikhi
 	- prepare_latents
 	- enable_fuser
 
+## StableDiffusionGLIGENTextImagePipeline
+
+[[autodoc]] StableDiffusionGLIGENTextImagePipeline
+	- all
+	- __call__
+	- enable_vae_slicing
+	- disable_vae_slicing
+	- enable_vae_tiling
+	- disable_vae_tiling
+	- enable_model_cpu_offload
+	- prepare_latents
+	- enable_fuser
+
 ## StableDiffusionPipelineOutput
 
 [[autodoc]] pipelines.stable_diffusion.StableDiffusionPipelineOutput

scripts/convert_gligen_to_diffusers.py

+import argparse
+import re
+
+import torch
+from transformers import (
+    CLIPProcessor,
+    CLIPTextModel,
+    CLIPTokenizer,
+    CLIPVisionModelWithProjection,
+)
+
+from diffusers import (
+    AutoencoderKL,
+    DDIMScheduler,
+    StableDiffusionGLIGENPipeline,
+    StableDiffusionGLIGENTextImagePipeline,
+    UNet2DConditionModel,
+)
+from diffusers.pipelines.stable_diffusion.convert_from_ckpt import (
+    assign_to_checkpoint,
+    conv_attn_to_linear,
+    protected,
+    renew_attention_paths,
+    renew_resnet_paths,
+    renew_vae_attention_paths,
+    renew_vae_resnet_paths,
+    shave_segments,
+    textenc_conversion_map,
+    textenc_pattern,
+)
+from diffusers.utils import is_omegaconf_available
+from diffusers.utils.import_utils import BACKENDS_MAPPING
+
+
+def convert_open_clip_checkpoint(checkpoint):
+    checkpoint = checkpoint["text_encoder"]
+    text_model = CLIPTextModel.from_pretrained("openai/clip-vit-large-patch14")
+
+    keys = list(checkpoint.keys())
+
+    text_model_dict = {}
+
+    if "cond_stage_model.model.text_projection" in checkpoint:
+        d_model = int(checkpoint["cond_stage_model.model.text_projection"].shape[0])
+    else:
+        d_model = 1024
+
+    for key in keys:
+        if "resblocks.23" in key:  # Diffusers drops the final layer and only uses the penultimate layer
+            continue
+        if key in textenc_conversion_map:
+            text_model_dict[textenc_conversion_map[key]] = checkpoint[key]
+        # if key.startswith("cond_stage_model.model.transformer."):
+        new_key = key[len("transformer.") :]
+        if new_key.endswith(".in_proj_weight"):
+            new_key = new_key[: -len(".in_proj_weight")]
+            new_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], new_key)
+            text_model_dict[new_key + ".q_proj.weight"] = checkpoint[key][:d_model, :]
+            text_model_dict[new_key + ".k_proj.weight"] = checkpoint[key][d_model : d_model * 2, :]
+            text_model_dict[new_key + ".v_proj.weight"] = checkpoint[key][d_model * 2 :, :]
+        elif new_key.endswith(".in_proj_bias"):
+            new_key = new_key[: -len(".in_proj_bias")]
+            new_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], new_key)
+            text_model_dict[new_key + ".q_proj.bias"] = checkpoint[key][:d_model]
+            text_model_dict[new_key + ".k_proj.bias"] = checkpoint[key][d_model : d_model * 2]
+            text_model_dict[new_key + ".v_proj.bias"] = checkpoint[key][d_model * 2 :]
+        else:
+            if key != "transformer.text_model.embeddings.position_ids":
+                new_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], new_key)
+
+                text_model_dict[new_key] = checkpoint[key]
+
+            if key == "transformer.text_model.embeddings.token_embedding.weight":
+                text_model_dict["text_model.embeddings.token_embedding.weight"] = checkpoint[key]
+
+    text_model_dict.pop("text_model.embeddings.transformer.text_model.embeddings.token_embedding.weight")
+
+    text_model.load_state_dict(text_model_dict)
+
+    return text_model
+
+
+def convert_gligen_vae_checkpoint(checkpoint, config):
+    checkpoint = checkpoint["autoencoder"]
+    vae_state_dict = {}
+    vae_key = "first_stage_model."
+    keys = list(checkpoint.keys())
+    for key in keys:
+        vae_state_dict[key.replace(vae_key, "")] = checkpoint.get(key)
+
+    new_checkpoint = {}
+
+    new_checkpoint["encoder.conv_in.weight"] = vae_state_dict["encoder.conv_in.weight"]
+    new_checkpoint["encoder.conv_in.bias"] = vae_state_dict["encoder.conv_in.bias"]
+    new_checkpoint["encoder.conv_out.weight"] = vae_state_dict["encoder.conv_out.weight"]
+    new_checkpoint["encoder.conv_out.bias"] = vae_state_dict["encoder.conv_out.bias"]
+    new_checkpoint["encoder.conv_norm_out.weight"] = vae_state_dict["encoder.norm_out.weight"]
+    new_checkpoint["encoder.conv_norm_out.bias"] = vae_state_dict["encoder.norm_out.bias"]
+
+    new_checkpoint["decoder.conv_in.weight"] = vae_state_dict["decoder.conv_in.weight"]
+    new_checkpoint["decoder.conv_in.bias"] = vae_state_dict["decoder.conv_in.bias"]
+    new_checkpoint["decoder.conv_out.weight"] = vae_state_dict["decoder.conv_out.weight"]
+    new_checkpoint["decoder.conv_out.bias"] = vae_state_dict["decoder.conv_out.bias"]
+    new_checkpoint["decoder.conv_norm_out.weight"] = vae_state_dict["decoder.norm_out.weight"]
+    new_checkpoint["decoder.conv_norm_out.bias"] = vae_state_dict["decoder.norm_out.bias"]
+
+    new_checkpoint["quant_conv.weight"] = vae_state_dict["quant_conv.weight"]
+    new_checkpoint["quant_conv.bias"] = vae_state_dict["quant_conv.bias"]
+    new_checkpoint["post_quant_conv.weight"] = vae_state_dict["post_quant_conv.weight"]
+    new_checkpoint["post_quant_conv.bias"] = vae_state_dict["post_quant_conv.bias"]
+
+    # Retrieves the keys for the encoder down blocks only
+    num_down_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "encoder.down" in layer})
+    down_blocks = {
+        layer_id: [key for key in vae_state_dict if f"down.{layer_id}" in key] for layer_id in range(num_down_blocks)
+    }
+
+    # Retrieves the keys for the decoder up blocks only
+    num_up_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "decoder.up" in layer})
+    up_blocks = {
+        layer_id: [key for key in vae_state_dict if f"up.{layer_id}" in key] for layer_id in range(num_up_blocks)
+    }
+
+    for i in range(num_down_blocks):
+        resnets = [key for key in down_blocks[i] if f"down.{i}" in key and f"down.{i}.downsample" not in key]
+
+        if f"encoder.down.{i}.downsample.conv.weight" in vae_state_dict:
+            new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.weight"] = vae_state_dict.pop(
+                f"encoder.down.{i}.downsample.conv.weight"
+            )
+            new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.bias"] = vae_state_dict.pop(
+                f"encoder.down.{i}.downsample.conv.bias"
+            )
+
+        paths = renew_vae_resnet_paths(resnets)
+        meta_path = {"old": f"down.{i}.block", "new": f"down_blocks.{i}.resnets"}
+        assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+
+    mid_resnets = [key for key in vae_state_dict if "encoder.mid.block" in key]
+    num_mid_res_blocks = 2
+    for i in range(1, num_mid_res_blocks + 1):
+        resnets = [key for key in mid_resnets if f"encoder.mid.block_{i}" in key]
+
+        paths = renew_vae_resnet_paths(resnets)
+        meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"}
+        assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+
+    mid_attentions = [key for key in vae_state_dict if "encoder.mid.attn" in key]
+    paths = renew_vae_attention_paths(mid_attentions)
+    meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"}
+    assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+    conv_attn_to_linear(new_checkpoint)
+
+    for i in range(num_up_blocks):
+        block_id = num_up_blocks - 1 - i
+        resnets = [
+            key for key in up_blocks[block_id] if f"up.{block_id}" in key and f"up.{block_id}.upsample" not in key
+        ]
+
+        if f"decoder.up.{block_id}.upsample.conv.weight" in vae_state_dict:
+            new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.weight"] = vae_state_dict[
+                f"decoder.up.{block_id}.upsample.conv.weight"
+            ]
+            new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.bias"] = vae_state_dict[
+                f"decoder.up.{block_id}.upsample.conv.bias"
+            ]
+
+        paths = renew_vae_resnet_paths(resnets)
+        meta_path = {"old": f"up.{block_id}.block", "new": f"up_blocks.{i}.resnets"}
+        assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+
+    mid_resnets = [key for key in vae_state_dict if "decoder.mid.block" in key]
+    num_mid_res_blocks = 2
+    for i in range(1, num_mid_res_blocks + 1):
+        resnets = [key for key in mid_resnets if f"decoder.mid.block_{i}" in key]
+
+        paths = renew_vae_resnet_paths(resnets)
+        meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"}
+        assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+
+    mid_attentions = [key for key in vae_state_dict if "decoder.mid.attn" in key]
+    paths = renew_vae_attention_paths(mid_attentions)
+    meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"}
+    assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config)
+    conv_attn_to_linear(new_checkpoint)
+
+    for key in new_checkpoint.keys():
+        if "encoder.mid_block.attentions.0" in key or "decoder.mid_block.attentions.0" in key:
+            if "query" in key:
+                new_checkpoint[key.replace("query", "to_q")] = new_checkpoint.pop(key)
+            if "value" in key:
+                new_checkpoint[key.replace("value", "to_v")] = new_checkpoint.pop(key)
+            if "key" in key:
+                new_checkpoint[key.replace("key", "to_k")] = new_checkpoint.pop(key)
+            if "proj_attn" in key:
+                new_checkpoint[key.replace("proj_attn", "to_out.0")] = new_checkpoint.pop(key)
+
+    return new_checkpoint
+
+
+def convert_gligen_unet_checkpoint(checkpoint, config, path=None, extract_ema=False):
+    unet_state_dict = {}
+    checkpoint = checkpoint["model"]
+    keys = list(checkpoint.keys())
+
+    unet_key = "model.diffusion_model."
+
+    if sum(k.startswith("model_ema") for k in keys) > 100 and extract_ema:
+        print(f"Checkpoint {path} has bot EMA and non-EMA weights.")
+        print(
+            "In this conversion only the EMA weights are extracted. If you want to instead extract the non-EMA"
+            " weights (useful to continue fine-tuning), please make sure to remove the `--extract_ema` flag."
+        )
+        for key in keys:
+            if key.startswith("model.diffusion_model"):
+                flat_ema_key = "model_ema." + "".join(key.split(".")[1:])
+                unet_state_dict[key.replace(unet_key, "")] = checkpoint.pop(flat_ema_key)
+    else:
+        if sum(k.startswith("model_ema") for k in keys) > 100:
+            print(
+                "In this conversion only the non-EMA weights are extracted. If you want to instead extract the EMA"
+                " weights (usually better for inference), please make sure to add the `--extract_ema` flag."
+            )
+    for key in keys:
+        unet_state_dict[key.replace(unet_key, "")] = checkpoint.pop(key)
+
+    new_checkpoint = {}
+
+    new_checkpoint["time_embedding.linear_1.weight"] = unet_state_dict["time_embed.0.weight"]
+    new_checkpoint["time_embedding.linear_1.bias"] = unet_state_dict["time_embed.0.bias"]
+    new_checkpoint["time_embedding.linear_2.weight"] = unet_state_dict["time_embed.2.weight"]
+    new_checkpoint["time_embedding.linear_2.bias"] = unet_state_dict["time_embed.2.bias"]
+
+    new_checkpoint["conv_in.weight"] = unet_state_dict["input_blocks.0.0.weight"]
+    new_checkpoint["conv_in.bias"] = unet_state_dict["input_blocks.0.0.bias"]
+
+    new_checkpoint["conv_norm_out.weight"] = unet_state_dict["out.0.weight"]
+    new_checkpoint["conv_norm_out.bias"] = unet_state_dict["out.0.bias"]
+    new_checkpoint["conv_out.weight"] = unet_state_dict["out.2.weight"]
+    new_checkpoint["conv_out.bias"] = unet_state_dict["out.2.bias"]
+
+    # Retrieves the keys for the input blocks only
+    num_input_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "input_blocks" in layer})
+    input_blocks = {
+        layer_id: [key for key in unet_state_dict if f"input_blocks.{layer_id}" in key]
+        for layer_id in range(num_input_blocks)
+    }
+
+    # Retrieves the keys for the middle blocks only
+    num_middle_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "middle_block" in layer})
+    middle_blocks = {
+        layer_id: [key for key in unet_state_dict if f"middle_block.{layer_id}" in key]
+        for layer_id in range(num_middle_blocks)
+    }
+
+    # Retrieves the keys for the output blocks only
+    num_output_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "output_blocks" in layer})
+    output_blocks = {
+        layer_id: [key for key in unet_state_dict if f"output_blocks.{layer_id}" in key]
+        for layer_id in range(num_output_blocks)
+    }
+
+    for i in range(1, num_input_blocks):
+        block_id = (i - 1) // (config["layers_per_block"] + 1)
+        layer_in_block_id = (i - 1) % (config["layers_per_block"] + 1)
+
+        resnets = [
+            key for key in input_blocks[i] if f"input_blocks.{i}.0" in key and f"input_blocks.{i}.0.op" not in key
+        ]
+        attentions = [key for key in input_blocks[i] if f"input_blocks.{i}.1" in key]
+
+        if f"input_blocks.{i}.0.op.weight" in unet_state_dict:
+            new_checkpoint[f"down_blocks.{block_id}.downsamplers.0.conv.weight"] = unet_state_dict.pop(
+                f"input_blocks.{i}.0.op.weight"
+            )
+            new_checkpoint[f"down_blocks.{block_id}.downsamplers.0.conv.bias"] = unet_state_dict.pop(
+                f"input_blocks.{i}.0.op.bias"
+            )
+
+        paths = renew_resnet_paths(resnets)
+        meta_path = {"old": f"input_blocks.{i}.0", "new": f"down_blocks.{block_id}.resnets.{layer_in_block_id}"}
+        assign_to_checkpoint(
+            paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config
+        )
+
+        if len(attentions):
+            paths = renew_attention_paths(attentions)
+            meta_path = {"old": f"input_blocks.{i}.1", "new": f"down_blocks.{block_id}.attentions.{layer_in_block_id}"}
+            assign_to_checkpoint(
+                paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config
+            )
+
+    resnet_0 = middle_blocks[0]
+    attentions = middle_blocks[1]
+    resnet_1 = middle_blocks[2]
+
+    resnet_0_paths = renew_resnet_paths(resnet_0)
+    assign_to_checkpoint(resnet_0_paths, new_checkpoint, unet_state_dict, config=config)
+
+    resnet_1_paths = renew_resnet_paths(resnet_1)
+    assign_to_checkpoint(resnet_1_paths, new_checkpoint, unet_state_dict, config=config)
+
+    attentions_paths = renew_attention_paths(attentions)
+    meta_path = {"old": "middle_block.1", "new": "mid_block.attentions.0"}
+    assign_to_checkpoint(
+        attentions_paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config
+    )
+
+    for i in range(num_output_blocks):
+        block_id = i // (config["layers_per_block"] + 1)
+        layer_in_block_id = i % (config["layers_per_block"] + 1)
+        output_block_layers = [shave_segments(name, 2) for name in output_blocks[i]]
+        output_block_list = {}
+
+        for layer in output_block_layers:
+            layer_id, layer_name = layer.split(".")[0], shave_segments(layer, 1)
+            if layer_id in output_block_list:
+                output_block_list[layer_id].append(layer_name)
+            else:
+                output_block_list[layer_id] = [layer_name]
+
+        if len(output_block_list) > 1:
+            resnets = [key for key in output_blocks[i] if f"output_blocks.{i}.0" in key]
+            attentions = [key for key in output_blocks[i] if f"output_blocks.{i}.1" in key]
+
+            resnet_0_paths = renew_resnet_paths(resnets)
+            paths = renew_resnet_paths(resnets)
+
+            meta_path = {"old": f"output_blocks.{i}.0", "new": f"up_blocks.{block_id}.resnets.{layer_in_block_id}"}
+            assign_to_checkpoint(
+                paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config
+            )
+
+            output_block_list = {k: sorted(v) for k, v in output_block_list.items()}
+            if ["conv.bias", "conv.weight"] in output_block_list.values():
+                index = list(output_block_list.values()).index(["conv.bias", "conv.weight"])
+                new_checkpoint[f"up_blocks.{block_id}.upsamplers.0.conv.weight"] = unet_state_dict[
+                    f"output_blocks.{i}.{index}.conv.weight"
+                ]
+                new_checkpoint[f"up_blocks.{block_id}.upsamplers.0.conv.bias"] = unet_state_dict[
+                    f"output_blocks.{i}.{index}.conv.bias"
+                ]
+
+                # Clear attentions as they have been attributed above.
+                if len(attentions) == 2:
+                    attentions = []
+
+            if len(attentions):
+                paths = renew_attention_paths(attentions)
+                meta_path = {
+                    "old": f"output_blocks.{i}.1",
+                    "new": f"up_blocks.{block_id}.attentions.{layer_in_block_id}",
+                }
+                assign_to_checkpoint(
+                    paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config
+                )
+        else:
+            resnet_0_paths = renew_resnet_paths(output_block_layers, n_shave_prefix_segments=1)
+            for path in resnet_0_paths:
+                old_path = ".".join(["output_blocks", str(i), path["old"]])
+                new_path = ".".join(["up_blocks", str(block_id), "resnets", str(layer_in_block_id), path["new"]])
+
+                new_checkpoint[new_path] = unet_state_dict[old_path]
+
+    for key in keys:
+        if "position_net" in key:
+            new_checkpoint[key] = unet_state_dict[key]
+
+    return new_checkpoint
+
+
+def create_vae_config(original_config, image_size: int):
+    vae_params = original_config.autoencoder.params.ddconfig
+    _ = original_config.autoencoder.params.embed_dim
+
+    block_out_channels = [vae_params.ch * mult for mult in vae_params.ch_mult]
+    down_block_types = ["DownEncoderBlock2D"] * len(block_out_channels)
+    up_block_types = ["UpDecoderBlock2D"] * len(block_out_channels)
+
+    config = {
+        "sample_size": image_size,
+        "in_channels": vae_params.in_channels,
+        "out_channels": vae_params.out_ch,
+        "down_block_types": tuple(down_block_types),
+        "up_block_types": tuple(up_block_types),
+        "block_out_channels": tuple(block_out_channels),
+        "latent_channels": vae_params.z_channels,
+        "layers_per_block": vae_params.num_res_blocks,
+    }
+
+    return config
+
+
+def create_unet_config(original_config, image_size: int, attention_type):
+    unet_params = original_config.model.params
+    vae_params = original_config.autoencoder.params.ddconfig
+
+    block_out_channels = [unet_params.model_channels * mult for mult in unet_params.channel_mult]
+
+    down_block_types = []
+    resolution = 1
+    for i in range(len(block_out_channels)):
+        block_type = "CrossAttnDownBlock2D" if resolution in unet_params.attention_resolutions else "DownBlock2D"
+        down_block_types.append(block_type)
+        if i != len(block_out_channels) - 1:
+            resolution *= 2
+
+    up_block_types = []
+    for i in range(len(block_out_channels)):
+        block_type = "CrossAttnUpBlock2D" if resolution in unet_params.attention_resolutions else "UpBlock2D"
+        up_block_types.append(block_type)
+        resolution //= 2
+
+    vae_scale_factor = 2 ** (len(vae_params.ch_mult) - 1)
+
+    head_dim = unet_params.num_heads if "num_heads" in unet_params else None
+    use_linear_projection = (
+        unet_params.use_linear_in_transformer if "use_linear_in_transformer" in unet_params else False
+    )
+    if use_linear_projection:
+        if head_dim is None:
+            head_dim = [5, 10, 20, 20]
+
+    config = {
+        "sample_size": image_size // vae_scale_factor,
+        "in_channels": unet_params.in_channels,
+        "down_block_types": tuple(down_block_types),
+        "block_out_channels": tuple(block_out_channels),
+        "layers_per_block": unet_params.num_res_blocks,
+        "cross_attention_dim": unet_params.context_dim,
+        "attention_head_dim": head_dim,
+        "use_linear_projection": use_linear_projection,
+        "attention_type": attention_type,
+    }
+
+    return config
+
+
+def convert_gligen_to_diffusers(
+    checkpoint_path: str,
+    original_config_file: str,
+    attention_type: str,
+    image_size: int = 512,
+    extract_ema: bool = False,
+    num_in_channels: int = None,
+    device: str = None,
+):
+    if not is_omegaconf_available():
+        raise ValueError(BACKENDS_MAPPING["omegaconf"][1])
+
+    from omegaconf import OmegaConf
+
+    if device is None:
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        checkpoint = torch.load(checkpoint_path, map_location=device)
+    else:
+        checkpoint = torch.load(checkpoint_path, map_location=device)
+
+    if "global_step" in checkpoint:
+        checkpoint["global_step"]
+    else:
+        print("global_step key not found in model")
+
+    original_config = OmegaConf.load(original_config_file)
+
+    if num_in_channels is not None:
+        original_config["model"]["params"]["in_channels"] = num_in_channels
+
+    num_train_timesteps = original_config.diffusion.params.timesteps
+    beta_start = original_config.diffusion.params.linear_start
+    beta_end = original_config.diffusion.params.linear_end
+
+    scheduler = DDIMScheduler(
+        beta_end=beta_end,
+        beta_schedule="scaled_linear",
+        beta_start=beta_start,
+        num_train_timesteps=num_train_timesteps,
+        steps_offset=1,
+        clip_sample=False,
+        set_alpha_to_one=False,
+        prediction_type="epsilon",
+    )
+
+    # Convert the UNet2DConditionalModel model
+    unet_config = create_unet_config(original_config, image_size, attention_type)
+    unet = UNet2DConditionModel(**unet_config)
+
+    converted_unet_checkpoint = convert_gligen_unet_checkpoint(
+        checkpoint, unet_config, path=checkpoint_path, extract_ema=extract_ema
+    )
+
+    unet.load_state_dict(converted_unet_checkpoint)
+
+    # Convert the VAE model
+    vae_config = create_vae_config(original_config, image_size)
+    converted_vae_checkpoint = convert_gligen_vae_checkpoint(checkpoint, vae_config)
+
+    vae = AutoencoderKL(**vae_config)
+    vae.load_state_dict(converted_vae_checkpoint)
+
+    # Convert the text model
+    text_encoder = convert_open_clip_checkpoint(checkpoint)
+    tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14")
+
+    if attention_type == "gated-text-image":
+        image_encoder = CLIPVisionModelWithProjection.from_pretrained("openai/clip-vit-large-patch14")
+        processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
+
+        pipe = StableDiffusionGLIGENTextImagePipeline(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            image_encoder=image_encoder,
+            processor=processor,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=None,
+            feature_extractor=None,
+        )
+    elif attention_type == "gated":
+        pipe = StableDiffusionGLIGENPipeline(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=None,
+            feature_extractor=None,
+        )
+
+    return pipe
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert."
+    )
+    parser.add_argument(
+        "--original_config_file",
+        default=None,
+        type=str,
+        required=True,
+        help="The YAML config file corresponding to the gligen architecture.",
+    )
+    parser.add_argument(
+        "--num_in_channels",
+        default=None,
+        type=int,
+        help="The number of input channels. If `None` number of input channels will be automatically inferred.",
+    )
+    parser.add_argument(
+        "--extract_ema",
+        action="store_true",
+        help=(
+            "Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights"
+            " or not. Defaults to `False`. Add `--extract_ema` to extract the EMA weights. EMA weights usually yield"
+            " higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning."
+        ),
+    )
+    parser.add_argument(
+        "--attention_type",
+        default=None,
+        type=str,
+        required=True,
+        help="Type of attention ex: gated or gated-text-image",
+    )
+    parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output model.")
+    parser.add_argument("--device", type=str, help="Device to use.")
+    parser.add_argument("--half", action="store_true", help="Save weights in half precision.")
+
+    args = parser.parse_args()
+
+    pipe = convert_gligen_to_diffusers(
+        checkpoint_path=args.checkpoint_path,
+        original_config_file=args.original_config_file,
+        attention_type=args.attention_type,
+        extract_ema=args.extract_ema,
+        num_in_channels=args.num_in_channels,
+        device=args.device,
+    )
+
+    if args.half:
+        pipe.to(torch_dtype=torch.float16)
+
+    pipe.save_pretrained(args.dump_path)

src/diffusers/__init__.py

@@ -66,6 +66,7 @@ else:
         AutoPipelineForImage2Image,
         AutoPipelineForInpainting,
         AutoPipelineForText2Image,
+        CLIPImageProjection,
         ConsistencyModelPipeline,
         DanceDiffusionPipeline,
         DDIMPipeline,
@@ -176,6 +177,7 @@ else:
         StableDiffusionDepth2ImgPipeline,
         StableDiffusionDiffEditPipeline,
         StableDiffusionGLIGENPipeline,
+        StableDiffusionGLIGENTextImagePipeline,
         StableDiffusionImageVariationPipeline,
         StableDiffusionImg2ImgPipeline,
         StableDiffusionInpaintPipeline,

src/diffusers/models/attention.py

@@ -154,7 +154,7 @@ class BasicTransformerBlock(nn.Module):
         self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn, final_dropout=final_dropout)
 
         # 4. Fuser
-        if attention_type == "gated":
+        if attention_type == "gated" or attention_type == "gated-text-image":
             self.fuser = GatedSelfAttentionDense(dim, cross_attention_dim, num_attention_heads, attention_head_dim)
 
         # let chunk size default to None

src/diffusers/models/embeddings.py

@@ -563,7 +563,7 @@ class FourierEmbedder(nn.Module):
 
 
 class PositionNet(nn.Module):
-    def __init__(self, positive_len, out_dim, fourier_freqs=8):
+    def __init__(self, positive_len, out_dim, feature_type="text-only", fourier_freqs=8):
         super().__init__()
         self.positive_len = positive_len
         self.out_dim = out_dim
@@ -573,30 +573,83 @@ class PositionNet(nn.Module):
 
         if isinstance(out_dim, tuple):
             out_dim = out_dim[0]
-        self.linears = nn.Sequential(
-            nn.Linear(self.positive_len + self.position_dim, 512),
-            nn.SiLU(),
-            nn.Linear(512, 512),
-            nn.SiLU(),
-            nn.Linear(512, out_dim),
-        )
 
-        self.null_positive_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+        if feature_type == "text-only":
+            self.linears = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.null_positive_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+
+        elif feature_type == "text-image":
+            self.linears_text = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.linears_image = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.null_text_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+            self.null_image_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+
         self.null_position_feature = torch.nn.Parameter(torch.zeros([self.position_dim]))
 
-    def forward(self, boxes, masks, positive_embeddings):
+    def forward(
+        self,
+        boxes,
+        masks,
+        positive_embeddings=None,
+        phrases_masks=None,
+        image_masks=None,
+        phrases_embeddings=None,
+        image_embeddings=None,
+    ):
         masks = masks.unsqueeze(-1)
 
         # embedding position (it may includes padding as placeholder)
         xyxy_embedding = self.fourier_embedder(boxes)  # B*N*4 -> B*N*C
 
         # learnable null embedding
-        positive_null = self.null_positive_feature.view(1, 1, -1)
         xyxy_null = self.null_position_feature.view(1, 1, -1)
 
         # replace padding with learnable null embedding
-        positive_embeddings = positive_embeddings * masks + (1 - masks) * positive_null
         xyxy_embedding = xyxy_embedding * masks + (1 - masks) * xyxy_null
 
-        objs = self.linears(torch.cat([positive_embeddings, xyxy_embedding], dim=-1))
+        # positionet with text only information
+        if positive_embeddings is not None:
+            # learnable null embedding
+            positive_null = self.null_positive_feature.view(1, 1, -1)
+
+            # replace padding with learnable null embedding
+            positive_embeddings = positive_embeddings * masks + (1 - masks) * positive_null
+
+            objs = self.linears(torch.cat([positive_embeddings, xyxy_embedding], dim=-1))
+
+        # positionet with text and image infomation
+        else:
+            phrases_masks = phrases_masks.unsqueeze(-1)
+            image_masks = image_masks.unsqueeze(-1)
+
+            # learnable null embedding
+            text_null = self.null_text_feature.view(1, 1, -1)
+            image_null = self.null_image_feature.view(1, 1, -1)
+
+            # replace padding with learnable null embedding
+            phrases_embeddings = phrases_embeddings * phrases_masks + (1 - phrases_masks) * text_null
+            image_embeddings = image_embeddings * image_masks + (1 - image_masks) * image_null
+
+            objs_text = self.linears_text(torch.cat([phrases_embeddings, xyxy_embedding], dim=-1))
+            objs_image = self.linears_image(torch.cat([image_embeddings, xyxy_embedding], dim=-1))
+            objs = torch.cat([objs_text, objs_image], dim=1)
+
         return objs

src/diffusers/models/unet_2d_condition.py

@@ -565,13 +565,17 @@ class UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin)
             block_out_channels[0], out_channels, kernel_size=conv_out_kernel, padding=conv_out_padding
         )
 
-        if attention_type == "gated":
+        if attention_type in ["gated", "gated-text-image"]:
             positive_len = 768
             if isinstance(cross_attention_dim, int):
                 positive_len = cross_attention_dim
             elif isinstance(cross_attention_dim, tuple) or isinstance(cross_attention_dim, list):
                 positive_len = cross_attention_dim[0]
-            self.position_net = PositionNet(positive_len=positive_len, out_dim=cross_attention_dim)
+
+            feature_type = "text-only" if attention_type == "gated" else "text-image"
+            self.position_net = PositionNet(
+                positive_len=positive_len, out_dim=cross_attention_dim, feature_type=feature_type
+            )
 
     @property
     def attn_processors(self) -> Dict[str, AttentionProcessor]:

src/diffusers/pipelines/__init__.py

@@ -94,6 +94,7 @@ else:
         StableDiffusionDepth2ImgPipeline,
         StableDiffusionDiffEditPipeline,
         StableDiffusionGLIGENPipeline,
+        StableDiffusionGLIGENTextImagePipeline,
         StableDiffusionImageVariationPipeline,
         StableDiffusionImg2ImgPipeline,
         StableDiffusionInpaintPipeline,
@@ -111,6 +112,7 @@ else:
         StableUnCLIPImg2ImgPipeline,
         StableUnCLIPPipeline,
     )
+    from .stable_diffusion.clip_image_project_model import CLIPImageProjection
     from .stable_diffusion_safe import StableDiffusionPipelineSafe
     from .stable_diffusion_xl import (
         StableDiffusionXLImg2ImgPipeline,

src/diffusers/pipelines/stable_diffusion/__init__.py

@@ -42,10 +42,12 @@ try:
 except OptionalDependencyNotAvailable:
     from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
 else:
+    from .clip_image_project_model import CLIPImageProjection
     from .pipeline_cycle_diffusion import CycleDiffusionPipeline
     from .pipeline_stable_diffusion import StableDiffusionPipeline
     from .pipeline_stable_diffusion_attend_and_excite import StableDiffusionAttendAndExcitePipeline
     from .pipeline_stable_diffusion_gligen import StableDiffusionGLIGENPipeline
+    from .pipeline_stable_diffusion_gligen_text_image import StableDiffusionGLIGENTextImagePipeline
     from .pipeline_stable_diffusion_img2img import StableDiffusionImg2ImgPipeline
     from .pipeline_stable_diffusion_inpaint import StableDiffusionInpaintPipeline
     from .pipeline_stable_diffusion_inpaint_legacy import StableDiffusionInpaintPipelineLegacy

src/diffusers/pipelines/stable_diffusion/clip_image_project_model.py

+# Copyright 2023 The GLIGEN Authors and HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from torch import nn
+
+from ...configuration_utils import ConfigMixin, register_to_config
+from ...models.modeling_utils import ModelMixin
+
+
+class CLIPImageProjection(ModelMixin, ConfigMixin):
+    @register_to_config
+    def __init__(self, hidden_size: int = 768):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.project = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
+
+    def forward(self, x):
+        return self.project(x)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py

+# Copyright 2023 The GLIGEN Authors and HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import warnings
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import PIL
+import torch
+from transformers import (
+    CLIPFeatureExtractor,
+    CLIPProcessor,
+    CLIPTextModel,
+    CLIPTokenizer,
+    CLIPVisionModelWithProjection,
+)
+
+from ...image_processor import VaeImageProcessor
+from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, UNet2DConditionModel
+from ...models.attention import GatedSelfAttentionDense
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+    is_accelerate_available,
+    is_accelerate_version,
+    logging,
+    randn_tensor,
+    replace_example_docstring,
+)
+from ..pipeline_utils import DiffusionPipeline
+from . import StableDiffusionPipelineOutput
+from .clip_image_project_model import CLIPImageProjection
+from .safety_checker import StableDiffusionSafetyChecker
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import StableDiffusionGLIGENTextImagePipeline
+        >>> from diffusers.utils import load_image
+
+        >>> # Insert objects described by image at the region defined by bounding boxes
+        >>> pipe = StableDiffusionGLIGENTextImagePipeline.from_pretrained(
+        ...     "anhnct/Gligen_Inpainting_Text_Image", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> input_image = load_image(
+        ...     "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/gligen/livingroom_modern.png"
+        ... )
+        >>> prompt = "a backpack"
+        >>> boxes = [[0.2676, 0.4088, 0.4773, 0.7183]]
+        >>> phrases = None
+        >>> gligen_image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/gligen/backpack.jpeg"
+        ... )
+
+        >>> images = pipe(
+        ...     prompt=prompt,
+        ...     gligen_phrases=phrases,
+        ...     gligen_inpaint_image=input_image,
+        ...     gligen_boxes=boxes,
+        ...     gligen_images=[gligen_image],
+        ...     gligen_scheduled_sampling_beta=1,
+        ...     output_type="pil",
+        ...     num_inference_steps=50,
+        ... ).images
+
+        >>> images[0].save("./gligen-inpainting-text-image-box.jpg")
+
+        >>> # Generate an image described by the prompt and
+        >>> # insert objects described by text and image at the region defined by bounding boxes
+        >>> pipe = StableDiffusionGLIGENTextImagePipeline.from_pretrained(
+        ...     "anhnct/Gligen_Text_Image", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> prompt = "a flower sitting on the beach"
+        >>> boxes = [[0.0, 0.09, 0.53, 0.76]]
+        >>> phrases = ["flower"]
+        >>> gligen_image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/gligen/pexels-pixabay-60597.jpg"
+        ... )
+
+        >>> images = pipe(
+        ...     prompt=prompt,
+        ...     gligen_phrases=phrases,
+        ...     gligen_images=[gligen_image],
+        ...     gligen_boxes=boxes,
+        ...     gligen_scheduled_sampling_beta=1,
+        ...     output_type="pil",
+        ...     num_inference_steps=50,
+        ... ).images
+
+        >>> images[0].save("./gligen-generation-text-image-box.jpg")
+
+        >>> # Generate an image described by the prompt and
+        >>> # transfer style described by image at the region defined by bounding boxes
+        >>> pipe = StableDiffusionGLIGENTextImagePipeline.from_pretrained(
+        ...     "anhnct/Gligen_Text_Image", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> prompt = "a dragon flying on the sky"
+        >>> boxes = [[0.4, 0.2, 1.0, 0.8], [0.0, 1.0, 0.0, 1.0]]  # Set `[0.0, 1.0, 0.0, 1.0]` for the style
+
+        >>> gligen_image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/landscape.png"
+        ... )
+
+        >>> gligen_placeholder = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/landscape.png"
+        ... )
+
+        >>> images = pipe(
+        ...     prompt=prompt,
+        ...     gligen_phrases=[
+        ...         "dragon",
+        ...         "placeholder",
+        ...     ],  # Can use any text instead of `placeholder` token, because we will use mask here
+        ...     gligen_images=[
+        ...         gligen_placeholder,
+        ...         gligen_image,
+        ...     ],  # Can use any image in gligen_placeholder, because we will use mask here
+        ...     input_phrases_mask=[1, 0],  # Set 0 for the placeholder token
+        ...     input_images_mask=[0, 1],  # Set 0 for the placeholder image
+        ...     gligen_boxes=boxes,
+        ...     gligen_scheduled_sampling_beta=1,
+        ...     output_type="pil",
+        ...     num_inference_steps=50,
+        ... ).images
+
+        >>> images[0].save("./gligen-generation-text-image-box-style-transfer.jpg")
+        ```
+"""
+
+
+class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline):
+    r"""
+    Pipeline for text-to-image generation using Stable Diffusion with Grounded-Language-to-Image Generation (GLIGEN).
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.).
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+        text_encoder ([`~transformers.CLIPTextModel`]):
+            Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+        tokenizer ([`~transformers.CLIPTokenizer`]):
+            A `CLIPTokenizer` to tokenize text.
+        processor ([`~transformers.CLIPProcessor`]):
+            A `CLIPProcessor` to procces reference image.
+        image_encoder ([`~transformers.CLIPVisionModelWithProjection`]):
+            Frozen image-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+        image_project ([`CLIPImageProjection`]):
+            A `CLIPImageProjection` to project image embedding into phrases embedding space.
+        unet ([`UNet2DConditionModel`]):
+            A `UNet2DConditionModel` to denoise the encoded image latents.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offensive or harmful.
+            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
+            about a model's potential harms.
+        feature_extractor ([`~transformers.CLIPImageProcessor`]):
+            A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+    """
+    _optional_components = ["safety_checker", "feature_extractor"]
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        processor: CLIPProcessor,
+        image_encoder: CLIPVisionModelWithProjection,
+        image_project: CLIPImageProjection,
+        unet: UNet2DConditionModel,
+        scheduler: KarrasDiffusionSchedulers,
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+        requires_safety_checker: bool = True,
+    ):
+        super().__init__()
+
+        if safety_checker is None and requires_safety_checker:
+            logger.warning(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
+        if safety_checker is not None and feature_extractor is None:
+            raise ValueError(
+                "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+                " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+            )
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            image_encoder=image_encoder,
+            processor=processor,
+            image_project=image_project,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+        self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    def enable_model_cpu_offload(self, gpu_id=0):
+        r"""
+        Offload all models to CPU to reduce memory usage with a low impact on performance. Moves one whole model at a
+        time to the GPU when its `forward` method is called, and the model remains in GPU until the next model runs.
+        Memory savings are lower than using `enable_sequential_cpu_offload`, but performance is much better due to the
+        iterative execution of the `unet`.
+        """
+        if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
+            from accelerate import cpu_offload_with_hook
+        else:
+            raise ImportError("`enable_model_offload` requires `accelerate v0.17.0` or higher.")
+
+        device = torch.device(f"cuda:{gpu_id}")
+
+        if self.device.type != "cpu":
+            self.to("cpu", silence_dtype_warnings=True)
+            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
+
+        hook = None
+        for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
+            _, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)
+
+        if self.safety_checker is not None:
+            _, hook = cpu_offload_with_hook(self.safety_checker, device, prev_module_hook=hook)
+
+        # We'll offload the last model manually.
+        self.final_offload_hook = hook
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt,
+        device,
+        num_images_per_prompt,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        lora_scale: Optional[float] = None,
+    ):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            do_classifier_free_guidance (`bool`):
+                whether to use classifier free guidance or not
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+                less than `1`).
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+        """
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        if prompt_embeds is None:
+            # textual inversion: procecss multi-vector tokens if necessary
+            if isinstance(self, TextualInversionLoaderMixin):
+                prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+            text_inputs = self.tokenizer(
+                prompt,
+                padding="max_length",
+                max_length=self.tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+                text_input_ids, untruncated_ids
+            ):
+                removed_text = self.tokenizer.batch_decode(
+                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                )
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = text_inputs.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            prompt_embeds = self.text_encoder(
+                text_input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            prompt_embeds = prompt_embeds[0]
+
+        if self.text_encoder is not None:
+            prompt_embeds_dtype = self.text_encoder.dtype
+        elif self.unet is not None:
+            prompt_embeds_dtype = self.unet.dtype
+        else:
+            prompt_embeds_dtype = prompt_embeds.dtype
+
+        prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+        bs_embed, seq_len, _ = prompt_embeds.shape
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance and negative_prompt_embeds is None:
+            uncond_tokens: List[str]
+            if negative_prompt is None:
+                uncond_tokens = [""] * batch_size
+            elif prompt is not None and type(prompt) is not type(negative_prompt):
+                raise TypeError(
+                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                    f" {type(prompt)}."
+                )
+            elif isinstance(negative_prompt, str):
+                uncond_tokens = [negative_prompt]
+            elif batch_size != len(negative_prompt):
+                raise ValueError(
+                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                    " the batch size of `prompt`."
+                )
+            else:
+                uncond_tokens = negative_prompt
+
+            # textual inversion: procecss multi-vector tokens if necessary
+            if isinstance(self, TextualInversionLoaderMixin):
+                uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)
+
+            max_length = prompt_embeds.shape[1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = uncond_input.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            negative_prompt_embeds = self.text_encoder(
+                uncond_input.input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            negative_prompt_embeds = negative_prompt_embeds[0]
+
+        if do_classifier_free_guidance:
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = negative_prompt_embeds.shape[1]
+
+            negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+        return prompt_embeds, negative_prompt_embeds
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is None:
+            has_nsfw_concept = None
+        else:
+            if torch.is_tensor(image):
+                feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+            else:
+                feature_extractor_input = self.image_processor.numpy_to_pil(image)
+            safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.check_inputs
+    def check_inputs(
+        self,
+        prompt,
+        height,
+        width,
+        callback_steps,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+    ):
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
+    def enable_fuser(self, enabled=True):
+        for module in self.unet.modules():
+            if type(module) is GatedSelfAttentionDense:
+                module.enabled = enabled
+
+    def draw_inpaint_mask_from_boxes(self, boxes, size):
+        """
+        Create an inpainting mask based on given boxes. This function generates an inpainting mask using the provided
+        boxes to mark regions that need to be inpainted.
+        """
+        inpaint_mask = torch.ones(size[0], size[1])
+        for box in boxes:
+            x0, x1 = box[0] * size[0], box[2] * size[0]
+            y0, y1 = box[1] * size[1], box[3] * size[1]
+            inpaint_mask[int(y0) : int(y1), int(x0) : int(x1)] = 0
+        return inpaint_mask
+
+    def crop(self, im, new_width, new_height):
+        """
+        Crop the input image to the specified dimensions.
+        """
+        width, height = im.size
+        left = (width - new_width) / 2
+        top = (height - new_height) / 2
+        right = (width + new_width) / 2
+        bottom = (height + new_height) / 2
+        return im.crop((left, top, right, bottom))
+
+    def target_size_center_crop(self, im, new_hw):
+        """
+        Crop and resize the image to the target size while keeping the center.
+        """
+        width, height = im.size
+        if width != height:
+            im = self.crop(im, min(height, width), min(height, width))
+        return im.resize((new_hw, new_hw), PIL.Image.LANCZOS)
+
+    def complete_mask(self, has_mask, max_objs, device):
+        """
+        Based on the input mask corresponding value `0 or 1` for each phrases and image, mask the features
+        corresponding to phrases and images.
+        """
+        mask = torch.ones(1, max_objs).type(self.text_encoder.dtype).to(device)
+        if has_mask is None:
+            return mask
+
+        if isinstance(has_mask, int):
+            return mask * has_mask
+        else:
+            for idx, value in enumerate(has_mask):
+                mask[0, idx] = value
+            return mask
+
+    def get_clip_feature(self, input, normalize_constant, device, is_image=False):
+        """
+        Get image and phrases embedding by using CLIP pretrain model. The image embedding is transformed into the
+        phrases embedding space through a projection.
+        """
+        if is_image:
+            if input is None:
+                return None
+            inputs = self.processor(images=[input], return_tensors="pt").to(device)
+            outputs = self.image_encoder(**inputs)
+            feature = outputs.image_embeds
+            feature = self.image_project(feature).squeeze(0)
+            feature = (feature / feature.norm()) * normalize_constant
+            feature = feature.unsqueeze(0)
+        else:
+            if input is None:
+                return None
+            inputs = self.tokenizer(input, return_tensors="pt", padding=True).to(device)
+            outputs = self.text_encoder(**inputs)
+            feature = outputs.pooler_output
+        return feature
+
+    def get_cross_attention_kwargs_with_grounded(
+        self,
+        hidden_size,
+        gligen_phrases,
+        gligen_images,
+        gligen_boxes,
+        input_phrases_mask,
+        input_images_mask,
+        repeat_batch,
+        normalize_constant,
+        max_objs,
+        device,
+    ):
+        """
+        Prepare the cross-attention kwargs containing information about the grounded input (boxes, mask, image
+        embedding, phrases embedding).
+        """
+        phrases, images = gligen_phrases, gligen_images
+        images = [None] * len(phrases) if images is None else images
+        phrases = [None] * len(images) if phrases is None else phrases
+
+        boxes = torch.zeros(max_objs, 4, device=device, dtype=self.text_encoder.dtype)
+        masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        phrases_masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        image_masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        phrases_embeddings = torch.zeros(max_objs, hidden_size, device=device, dtype=self.text_encoder.dtype)
+        image_embeddings = torch.zeros(max_objs, hidden_size, device=device, dtype=self.text_encoder.dtype)
+
+        text_features = []
+        image_features = []
+        for phrase, image in zip(phrases, images):
+            text_features.append(self.get_clip_feature(phrase, normalize_constant, device, is_image=False))
+            image_features.append(self.get_clip_feature(image, normalize_constant, device, is_image=True))
+
+        for idx, (box, text_feature, image_feature) in enumerate(zip(gligen_boxes, text_features, image_features)):
+            boxes[idx] = torch.tensor(box)
+            masks[idx] = 1
+            if text_feature is not None:
+                phrases_embeddings[idx] = text_feature
+                phrases_masks[idx] = 1
+            if image_feature is not None:
+                image_embeddings[idx] = image_feature
+                image_masks[idx] = 1
+
+        input_phrases_mask = self.complete_mask(input_phrases_mask, max_objs, device)
+        phrases_masks = phrases_masks.unsqueeze(0).repeat(repeat_batch, 1) * input_phrases_mask
+        input_images_mask = self.complete_mask(input_images_mask, max_objs, device)
+        image_masks = image_masks.unsqueeze(0).repeat(repeat_batch, 1) * input_images_mask
+        boxes = boxes.unsqueeze(0).repeat(repeat_batch, 1, 1)
+        masks = masks.unsqueeze(0).repeat(repeat_batch, 1)
+        phrases_embeddings = phrases_embeddings.unsqueeze(0).repeat(repeat_batch, 1, 1)
+        image_embeddings = image_embeddings.unsqueeze(0).repeat(repeat_batch, 1, 1)
+
+        out = {
+            "boxes": boxes,
+            "masks": masks,
+            "phrases_masks": phrases_masks,
+            "image_masks": image_masks,
+            "phrases_embeddings": phrases_embeddings,
+            "image_embeddings": image_embeddings,
+        }
+
+        return out
+
+    def get_cross_attention_kwargs_without_grounded(self, hidden_size, repeat_batch, max_objs, device):
+        """
+        Prepare the cross-attention kwargs without information about the grounded input (boxes, mask, image embedding,
+        phrases embedding) (All are zero tensor).
+        """
+        boxes = torch.zeros(max_objs, 4, device=device, dtype=self.text_encoder.dtype)
+        masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        phrases_masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        image_masks = torch.zeros(max_objs, device=device, dtype=self.text_encoder.dtype)
+        phrases_embeddings = torch.zeros(max_objs, hidden_size, device=device, dtype=self.text_encoder.dtype)
+        image_embeddings = torch.zeros(max_objs, hidden_size, device=device, dtype=self.text_encoder.dtype)
+
+        out = {
+            "boxes": boxes.unsqueeze(0).repeat(repeat_batch, 1, 1),
+            "masks": masks.unsqueeze(0).repeat(repeat_batch, 1),
+            "phrases_masks": phrases_masks.unsqueeze(0).repeat(repeat_batch, 1),
+            "image_masks": image_masks.unsqueeze(0).repeat(repeat_batch, 1),
+            "phrases_embeddings": phrases_embeddings.unsqueeze(0).repeat(repeat_batch, 1, 1),
+            "image_embeddings": image_embeddings.unsqueeze(0).repeat(repeat_batch, 1, 1),
+        }
+
+        return out
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 7.5,
+        gligen_scheduled_sampling_beta: float = 0.3,
+        gligen_phrases: List[str] = None,
+        gligen_images: List[PIL.Image.Image] = None,
+        input_phrases_mask: Union[int, List[int]] = None,
+        input_images_mask: Union[int, List[int]] = None,
+        gligen_boxes: List[List[float]] = None,
+        gligen_inpaint_image: Optional[PIL.Image.Image] = None,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        gligen_normalize_constant: float = 28.7,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                A higher guidance scale value encourages the model to generate images closely linked to the text
+                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+            gligen_phrases (`List[str]`):
+                The phrases to guide what to include in each of the regions defined by the corresponding
+                `gligen_boxes`. There should only be one phrase per bounding box.
+            gligen_images (`List[PIL.Image.Image]`):
+                The images to guide what to include in each of the regions defined by the corresponding `gligen_boxes`.
+                There should only be one image per bounding box
+            input_phrases_mask (`int` or `List[int]`):
+                pre phrases mask input defined by the correspongding `input_phrases_mask`
+            input_images_mask (`int` or `List[int]`):
+                pre images mask input defined by the correspongding `input_images_mask`
+            gligen_boxes (`List[List[float]]`):
+                The bounding boxes that identify rectangular regions of the image that are going to be filled with the
+                content described by the corresponding `gligen_phrases`. Each rectangular box is defined as a
+                `List[float]` of 4 elements `[xmin, ymin, xmax, ymax]` where each value is between [0,1].
+            gligen_inpaint_image (`PIL.Image.Image`, *optional*):
+                The input image, if provided, is inpainted with objects described by the `gligen_boxes` and
+                `gligen_phrases`. Otherwise, it is treated as a generation task on a blank input image.
+            gligen_scheduled_sampling_beta (`float`, defaults to 0.3):
+                Scheduled Sampling factor from [GLIGEN: Open-Set Grounded Text-to-Image
+                Generation](https://arxiv.org/pdf/2301.07093.pdf). Scheduled Sampling factor is only varied for
+                scheduled sampling during inference for improved quality and controllability.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that calls every `callback_steps` steps during inference. The function is called with the
+                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function is called. If not specified, the callback is called at
+                every step.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            gligen_normalize_constant (`float`, *optional*, defaults to 28.7):
+                The normalize value of the image embedding.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+                otherwise a `tuple` is returned where the first element is a list with the generated images and the
+                second element is a list of `bool`s indicating whether the corresponding generated image contains
+                "not-safe-for-work" (nsfw) content.
+        """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * self.vae_scale_factor
+        width = width or self.unet.config.sample_size * self.vae_scale_factor
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            height,
+            width,
+            callback_steps,
+            negative_prompt,
+            prompt_embeds,
+            negative_prompt_embeds,
+        )
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+            prompt,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+        )
+
+        if do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 5.1 Prepare GLIGEN variables
+        max_objs = 30
+        if len(gligen_boxes) > max_objs:
+            warnings.warn(
+                f"More that {max_objs} objects found. Only first {max_objs} objects will be processed.",
+                FutureWarning,
+            )
+            gligen_phrases = gligen_phrases[:max_objs]
+            gligen_boxes = gligen_boxes[:max_objs]
+            gligen_images = gligen_images[:max_objs]
+
+        repeat_batch = batch_size * num_images_per_prompt
+
+        if do_classifier_free_guidance:
+            repeat_batch = repeat_batch * 2
+
+        if cross_attention_kwargs is None:
+            cross_attention_kwargs = {}
+
+        hidden_size = prompt_embeds.shape[2]
+
+        cross_attention_kwargs["gligen"] = self.get_cross_attention_kwargs_with_grounded(
+            hidden_size=hidden_size,
+            gligen_phrases=gligen_phrases,
+            gligen_images=gligen_images,
+            gligen_boxes=gligen_boxes,
+            input_phrases_mask=input_phrases_mask,
+            input_images_mask=input_images_mask,
+            repeat_batch=repeat_batch,
+            normalize_constant=gligen_normalize_constant,
+            max_objs=max_objs,
+            device=device,
+        )
+
+        cross_attention_kwargs_without_grounded = {}
+        cross_attention_kwargs_without_grounded["gligen"] = self.get_cross_attention_kwargs_without_grounded(
+            hidden_size=hidden_size, repeat_batch=repeat_batch, max_objs=max_objs, device=device
+        )
+
+        # Prepare latent variables for GLIGEN inpainting
+        if gligen_inpaint_image is not None:
+            # if the given input image is not of the same size as expected by VAE
+            # center crop and resize the input image to expected shape
+            if gligen_inpaint_image.size != (self.vae.sample_size, self.vae.sample_size):
+                gligen_inpaint_image = self.target_size_center_crop(gligen_inpaint_image, self.vae.sample_size)
+            # Convert a single image into a batch of images with a batch size of 1
+            # The resulting shape becomes (1, C, H, W), where C is the number of channels,
+            # and H and W are the height and width of the image.
+            # scales the pixel values to a range [-1, 1]
+            gligen_inpaint_image = self.image_processor.preprocess(gligen_inpaint_image)
+            gligen_inpaint_image = gligen_inpaint_image.to(dtype=self.vae.dtype, device=self.vae.device)
+            # Run AutoEncoder to get corresponding latents
+            gligen_inpaint_latent = self.vae.encode(gligen_inpaint_image).latent_dist.sample()
+            gligen_inpaint_latent = self.vae.config.scaling_factor * gligen_inpaint_latent
+            # Generate an inpainting mask
+            # pixel value = 0, where the object is present (defined by bounding boxes above)
+            #               1, everywhere else
+            gligen_inpaint_mask = self.draw_inpaint_mask_from_boxes(gligen_boxes, gligen_inpaint_latent.shape[2:])
+            gligen_inpaint_mask = gligen_inpaint_mask.to(
+                dtype=gligen_inpaint_latent.dtype, device=gligen_inpaint_latent.device
+            )
+            gligen_inpaint_mask = gligen_inpaint_mask[None, None]
+            gligen_inpaint_mask_addition = torch.cat(
+                (gligen_inpaint_latent * gligen_inpaint_mask, gligen_inpaint_mask), dim=1
+            )
+            # Convert a single mask into a batch of masks with a batch size of 1
+            gligen_inpaint_mask_addition = gligen_inpaint_mask_addition.expand(repeat_batch, -1, -1, -1).clone()
+
+        int(gligen_scheduled_sampling_beta * len(timesteps))
+        self.enable_fuser(True)
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if latents.shape[1] != 4:
+                    latents = torch.randn_like(latents[:, :4])
+
+                if gligen_inpaint_image is not None:
+                    gligen_inpaint_latent_with_noise = (
+                        self.scheduler.add_noise(gligen_inpaint_latent, torch.randn_like(gligen_inpaint_latent), t)
+                        .expand(latents.shape[0], -1, -1, -1)
+                        .clone()
+                    )
+                    latents = gligen_inpaint_latent_with_noise * gligen_inpaint_mask + latents * (
+                        1 - gligen_inpaint_mask
+                    )
+
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+                if gligen_inpaint_image is not None:
+                    latent_model_input = torch.cat((latent_model_input, gligen_inpaint_mask_addition), dim=1)
+
+                # predict the noise residual with grounded information
+                noise_pred_with_grounding = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    cross_attention_kwargs=cross_attention_kwargs,
+                ).sample
+
+                # predict the noise residual without grounded information
+                noise_pred_without_grounding = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    cross_attention_kwargs=cross_attention_kwargs_without_grounded,
+                ).sample
+
+                # perform guidance
+                if do_classifier_free_guidance:
+                    # Using noise_pred_text from noise residual with grounded information and noise_pred_uncond from noise residual without grounded information
+                    _, noise_pred_text = noise_pred_with_grounding.chunk(2)
+                    noise_pred_uncond, _ = noise_pred_without_grounding.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+                else:
+                    noise_pred = noise_pred_with_grounding
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        callback(i, t, latents)
+
+        if not output_type == "latent":
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+        else:
+            image = latents
+            has_nsfw_concept = None
+
+        if has_nsfw_concept is None:
+            do_denormalize = [True] * image.shape[0]
+        else:
+            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+        # Offload last model to CPU
+        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+            self.final_offload_hook.offload()
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

src/diffusers/pipelines/versatile_diffusion/modeling_text_unet.py

@@ -170,7 +170,7 @@ class FourierEmbedder(nn.Module):
 
 
 class PositionNet(nn.Module):
-    def __init__(self, positive_len, out_dim, fourier_freqs=8):
+    def __init__(self, positive_len, out_dim, feature_type, fourier_freqs=8):
         super().__init__()
         self.positive_len = positive_len
         self.out_dim = out_dim
@@ -180,32 +180,72 @@ class PositionNet(nn.Module):
 
         if isinstance(out_dim, tuple):
             out_dim = out_dim[0]
-        self.linears = nn.Sequential(
-            nn.Linear(self.positive_len + self.position_dim, 512),
-            nn.SiLU(),
-            nn.Linear(512, 512),
-            nn.SiLU(),
-            nn.Linear(512, out_dim),
-        )
 
-        self.null_positive_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+        if feature_type == "text-only":
+            self.linears = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.null_positive_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+
+        elif feature_type == "text-image":
+            self.linears_text = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.linears_image = nn.Sequential(
+                nn.Linear(self.positive_len + self.position_dim, 512),
+                nn.SiLU(),
+                nn.Linear(512, 512),
+                nn.SiLU(),
+                nn.Linear(512, out_dim),
+            )
+            self.null_text_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+            self.null_image_feature = torch.nn.Parameter(torch.zeros([self.positive_len]))
+
         self.null_position_feature = torch.nn.Parameter(torch.zeros([self.position_dim]))
 
-    def forward(self, boxes, masks, positive_embeddings):
+    def forward(
+        self,
+        boxes,
+        masks,
+        positive_embeddings=None,
+        phrases_masks=None,
+        image_masks=None,
+        phrases_embeddings=None,
+        image_embeddings=None,
+    ):
         masks = masks.unsqueeze(-1)
 
-        # embedding position (it may includes padding as placeholder)
-        xyxy_embedding = self.fourier_embedder(boxes)  # B*N*4 -> B*N*C
-
-        # learnable null embedding
-        positive_null = self.null_positive_feature.view(1, 1, -1)
+        xyxy_embedding = self.fourier_embedder(boxes)
         xyxy_null = self.null_position_feature.view(1, 1, -1)
-
-        # replace padding with learnable null embedding
-        positive_embeddings = positive_embeddings * masks + (1 - masks) * positive_null
         xyxy_embedding = xyxy_embedding * masks + (1 - masks) * xyxy_null
 
-        objs = self.linears(torch.cat([positive_embeddings, xyxy_embedding], dim=-1))
+        if positive_embeddings:
+            positive_null = self.null_positive_feature.view(1, 1, -1)
+            positive_embeddings = positive_embeddings * masks + (1 - masks) * positive_null
+
+            objs = self.linears(torch.cat([positive_embeddings, xyxy_embedding], dim=-1))
+        else:
+            phrases_masks = phrases_masks.unsqueeze(-1)
+            image_masks = image_masks.unsqueeze(-1)
+
+            text_null = self.null_text_feature.view(1, 1, -1)
+            image_null = self.null_image_feature.view(1, 1, -1)
+
+            phrases_embeddings = phrases_embeddings * phrases_masks + (1 - phrases_masks) * text_null
+            image_embeddings = image_embeddings * image_masks + (1 - image_masks) * image_null
+
+            objs_text = self.linears_text(torch.cat([phrases_embeddings, xyxy_embedding], dim=-1))
+            objs_image = self.linears_image(torch.cat([image_embeddings, xyxy_embedding], dim=-1))
+            objs = torch.cat([objs_text, objs_image], dim=1)
+
         return objs
 
 
@@ -730,13 +770,17 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             block_out_channels[0], out_channels, kernel_size=conv_out_kernel, padding=conv_out_padding
         )
 
-        if attention_type == "gated":
+        if attention_type in ["gated", "gated-text-image"]:
             positive_len = 768
             if isinstance(cross_attention_dim, int):
                 positive_len = cross_attention_dim
             elif isinstance(cross_attention_dim, tuple) or isinstance(cross_attention_dim, list):
                 positive_len = cross_attention_dim[0]
-            self.position_net = PositionNet(positive_len=positive_len, out_dim=cross_attention_dim)
+
+            feature_type = "text-only" if attention_type == "gated" else "text-image"
+            self.position_net = PositionNet(
+                positive_len=positive_len, out_dim=cross_attention_dim, feature_type=feature_type
+            )
 
     @property
     def attn_processors(self) -> Dict[str, AttentionProcessor]:

src/diffusers/utils/dummy_pt_objects.py

@@ -315,6 +315,21 @@ class AutoPipelineForText2Image(metaclass=DummyObject):
         requires_backends(cls, ["torch"])
 
 
+class CLIPImageProjection(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
 class ConsistencyModelPipeline(metaclass=DummyObject):
     _backends = ["torch"]
 

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -677,6 +677,21 @@ class StableDiffusionGLIGENPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class StableDiffusionGLIGENTextImagePipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class StableDiffusionImageVariationPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

tests/pipelines/stable_diffusion/test_stable_diffusion_gligen_text_image.py

+# coding=utf-8
+# Copyright 2023 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import numpy as np
+import torch
+from transformers import (
+    CLIPProcessor,
+    CLIPTextConfig,
+    CLIPTextModel,
+    CLIPTokenizer,
+    CLIPVisionConfig,
+    CLIPVisionModelWithProjection,
+)
+
+from diffusers import (
+    AutoencoderKL,
+    CLIPImageProjection,
+    DDIMScheduler,
+    StableDiffusionGLIGENTextImagePipeline,
+    UNet2DConditionModel,
+)
+from diffusers.utils import load_image
+from diffusers.utils.testing_utils import enable_full_determinism
+
+from ..pipeline_params import (
+    TEXT_TO_IMAGE_BATCH_PARAMS,
+    TEXT_TO_IMAGE_IMAGE_PARAMS,
+    TEXT_TO_IMAGE_PARAMS,
+)
+from ..test_pipelines_common import PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin
+
+
+enable_full_determinism()
+
+
+class GligenTextImagePipelineFastTests(
+    PipelineLatentTesterMixin, PipelineKarrasSchedulerTesterMixin, PipelineTesterMixin, unittest.TestCase
+):
+    pipeline_class = StableDiffusionGLIGENTextImagePipeline
+    params = TEXT_TO_IMAGE_PARAMS | {"gligen_phrases", "gligen_images", "gligen_boxes"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        unet = UNet2DConditionModel(
+            block_out_channels=(32, 64),
+            layers_per_block=2,
+            sample_size=32,
+            in_channels=4,
+            out_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
+            cross_attention_dim=32,
+            attention_type="gated-text-image",
+        )
+        # unet.position_net = PositionNet(32,32)
+        scheduler = DDIMScheduler(
+            beta_start=0.00085,
+            beta_end=0.012,
+            beta_schedule="scaled_linear",
+            clip_sample=False,
+            set_alpha_to_one=False,
+        )
+        torch.manual_seed(0)
+        vae = AutoencoderKL(
+            block_out_channels=[32, 64],
+            in_channels=3,
+            out_channels=3,
+            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+            latent_channels=4,
+            sample_size=128,
+        )
+        torch.manual_seed(0)
+        text_encoder_config = CLIPTextConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=32,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=5,
+            pad_token_id=1,
+            vocab_size=1000,
+        )
+        text_encoder = CLIPTextModel(text_encoder_config)
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        image_encoder_config = CLIPVisionConfig(
+            hidden_size=32,
+            projection_dim=32,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=5,
+        )
+        image_encoder = CLIPVisionModelWithProjection(image_encoder_config)
+        processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
+
+        image_project = CLIPImageProjection(hidden_size=32)
+
+        components = {
+            "unet": unet,
+            "scheduler": scheduler,
+            "vae": vae,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "safety_checker": None,
+            "feature_extractor": None,
+            "image_encoder": image_encoder,
+            "image_project": image_project,
+            "processor": processor,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+
+        gligen_images = load_image(
+            "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/gligen/livingroom_modern.png"
+        )
+        inputs = {
+            "prompt": "A modern livingroom",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "gligen_phrases": ["a birthday cake"],
+            "gligen_images": [gligen_images],
+            "gligen_boxes": [[0.2676, 0.6088, 0.4773, 0.7183]],
+            "output_type": "np",
+        }
+        return inputs
+
+    def test_gligen(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        components = self.get_dummy_components()
+        sd_pipe = StableDiffusionGLIGENTextImagePipeline(**components)
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = sd_pipe(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        assert image.shape == (1, 64, 64, 3)
+        expected_slice = np.array([0.5069, 0.5561, 0.4577, 0.4792, 0.5203, 0.4089, 0.5039, 0.4919, 0.4499])
+
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_attention_slicing_forward_pass(self):
+        super().test_attention_slicing_forward_pass(expected_max_diff=3e-3)
+
+    def test_inference_batch_single_identical(self):
+        super().test_inference_batch_single_identical(batch_size=3, expected_max_diff=3e-3)