{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "411c59b3-f177-4a10-8925-d931ce572eaa",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from diffusers import StableDiffusionAdapterPipeline, T2IAdapter, DDIMScheduler, AutoencoderKL\n",
    "from PIL import Image\n",
    "\n",
    "from ip_adapter import IPAdapter"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6b6dc69c-192d-4d74-8b1e-f0d9ccfbdb49",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "\n",
    "current_dir = os.getcwd()\n",
    "print(current_dir)\n",
    "\n",
    "base_model_path = f\"{current_dir}/pretrained_models/sd1.5/Realistic_Vision_v4.0_noVAE\"\n",
    "vae_model_path = f\"{current_dir}/pretrained_models/sd1.5/sd-vae-ft-mse\"\n",
    "image_encoder_path = f\"{current_dir}/pretrained_models/models/image_encoder/\"\n",
    "ip_ckpt = f\"{current_dir}/pretrained_models/models/ip-adapter_sd15.safetensors\"\n",
    "device = \"cuda\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "63ec542f-8474-4f38-9457-073425578073",
   "metadata": {},
   "outputs": [],
   "source": [
    "def image_grid(imgs, rows, cols):\n",
    "    assert len(imgs) == rows*cols\n",
    "\n",
    "    w, h = imgs[0].size\n",
    "    grid = Image.new('RGB', size=(cols*w, rows*h))\n",
    "    grid_w, grid_h = grid.size\n",
    "    \n",
    "    for i, img in enumerate(imgs):\n",
    "        grid.paste(img, box=(i%cols*w, i//cols*h))\n",
    "    return grid\n",
    "\n",
    "noise_scheduler = DDIMScheduler(\n",
    "    num_train_timesteps=1000,\n",
    "    beta_start=0.00085,\n",
    "    beta_end=0.012,\n",
    "    beta_schedule=\"scaled_linear\",\n",
    "    clip_sample=False,\n",
    "    set_alpha_to_one=False,\n",
    "    steps_offset=1,\n",
    ")\n",
    "vae = AutoencoderKL.from_pretrained(vae_model_path).to(dtype=torch.float16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3849f9d0-5f68-4a49-9190-69dd50720cae",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load t2i-adapter\n",
    "adapter_model_path = f\"{current_dir}/pretrained_models/sd1.5/diffusers/t2iadapter_depth_sd15v2/\"\n",
    "adapter = T2IAdapter.from_pretrained(adapter_model_path, torch_dtype=torch.float16)\n",
    "# load SD pipeline\n",
    "pipe = StableDiffusionAdapterPipeline.from_pretrained(\n",
    "    base_model_path,\n",
    "    adapter=adapter,\n",
    "    torch_dtype=torch.float16,\n",
    "    scheduler=noise_scheduler,\n",
    "    vae=vae,\n",
    "    feature_extractor=None,\n",
    "    safety_checker=None\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ec09e937-3904-4d8e-a559-9066502ded36",
   "metadata": {},
   "outputs": [],
   "source": [
    "# read image prompt\n",
    "image = Image.open(\"assets/images/river.png\")\n",
    "depth_map = Image.open(\"assets/structure_controls/depth2.png\")\n",
    "image_grid([image.resize((256, 256)), depth_map.resize((256, 256))], 1, 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "81b1ab06-d3ed-4a7e-a356-9ddf1a2eecd6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load ip-adapter\n",
    "ip_model = IPAdapter(pipe, image_encoder_path, ip_ckpt, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b77f52de-a9e4-44e1-aeec-8165414f1273",
   "metadata": {},
   "outputs": [],
   "source": [
    "# generate image\n",
    "num_samples = 4\n",
    "depth_map = [depth_map] * num_samples # a bug of diffuser, we have to set the number by hard code\n",
    "images = ip_model.generate(pil_image=image, image=depth_map, num_samples=num_samples, num_inference_steps=50, seed=42)\n",
    "grid = image_grid(images, 1, num_samples)\n",
    "grid"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}