Skip to content

GitLab

Unverified Commit ad8097b9 authored by Muyang Li's avatar Muyang Li Committed by GitHub
Browse files

Release v0.2.0 

Ready to release v0.2.0
parents 804a6d30 998192ca
Hide whitespace changes
Inline Side-by-side
Showing with 196 additions and 1771 deletions
.github/workflows/lint.yaml 0 → 100644
View file @ ad8097b9
name: Lint
on:
push:
branches:
- main
pull_request:
branches:
- main
jobs:
lint:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Set up Python
uses: actions/setup-python@v5
with:
python-version: '3.10'
- name: Install dependencies
run: pip install ruff
- name: Run ruff check
run: ruff check nunchaku examples tests --output-format github
.pre-commit-config.yaml 0 → 100644
View file @ ad8097b9
repos:
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.3.2
hooks:
- id: ruff
args: ["--output-format", "github"]
Dockerfile 0 → 100644
View file @ ad8097b9
# Use an NVIDIA base image with CUDA support
ARG CUDA_IMAGE="12.8.1-devel-ubuntu24.04"
FROM nvidia/cuda:${CUDA_IMAGE}
ENV DEBIAN_FRONTEND=noninteractive
ARG PYTHON_VERSION=3.11
ARG TORCH_VERSION=2.6
ARG TORCHVISION_VERSION=0.21
ARG TORCHAUDIO_VERSION=2.6
ARG CUDA_SHORT_VERSION=12.8
# Set working directory
WORKDIR /
RUN echo PYTHON_VERSION=${PYTHON_VERSION} \
&& echo CUDA_SHORT_VERSION=${CUDA_SHORT_VERSION} \
&& echo TORCH_VERSION=${TORCH_VERSION} \
&& echo TORCHVISION_VERSION=${TORCHVISION_VERSION} \
&& echo TORCHAUDIO_VERSION=${TORCHAUDIO_VERSION}
# Setup timezone and install system dependencies
RUN 'tzdata tzdata/Areas select America' | debconf-set-selections \
&& echo 'tzdata tzdata/Zones/America select New_York' | debconf-set-selections \
&& apt update -y \
&& apt install software-properties-common -y \
&& add-apt-repository ppa:deadsnakes/ppa -y \
&& apt update
RUN apt install python${PYTHON_VERSION} python${PYTHON_VERSION}-dev g++-11 gcc-11 -y \
&& update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
&& update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} && apt install python${PYTHON_VERSION}-distutils -y \
&& update-alternatives --install /usr/bin/python python /usr/bin/python${PYTHON_VERSION} 1 \
&& update-alternatives --set python /usr/bin/python${PYTHON_VERSION}
RUN apt install curl git sudo libibverbs-dev -y \
&& apt install -y rdma-core infiniband-diags openssh-server perftest ibverbs-providers libibumad3 libibverbs1 libnl-3-200 libnl-route-3-200 librdmacm1 \
&& curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && python3 get-pip.py \
&& python3 --version \
&& python3 -m pip --version \
&& rm -rf /var/lib/apt/lists/* \
&& update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-11 1 && update-alternatives --set gcc /usr/bin/gcc-11 \
&& update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-11 1 && update-alternatives --set g++ /usr/bin/g++-11 \
&& apt clean
# Install building dependencies
RUN pip install torch==${TORCH_VERSION} torchvision==${TORCHVISION_VERSION} torchaudio==${TORCHAUDIO_VERSION} --index-url https://download.pytorch.org/whl/cu${CUDA_SHORT_VERSION}
RUN pip install ninja wheel diffusers transformers accelerate sentencepiece protobuf huggingface_hub comfy-cli
# Start building
RUN git clone https://github.com/mit-han-lab/nunchaku.git \
&& cd nunchaku \
&& git submodule init \
&& git submodule update \
&& NUNCHAKU_INSTALL_MODE=ALL python setup.py develop
RUN cd .. && git clone https://github.com/comfyanonymous/ComfyUI \
&& cd ComfyUI/custom_nodes && git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager \
&& git clone https://github.com/mit-han-lab/ComfyUI-nunchaku.git nunchaku_nodes \
&& cd .. && mkdir -p user/default/workflows/ && cp custom_nodes/nunchaku_nodes/workflows/* user/default/workflows/
README.md
View file @ ad8097b9
......@@ -5,27 +5,28 @@
<a href="http://arxiv.org/abs/2411.05007"><b>Paper</b></a> | <a href="https://hanlab.mit.edu/projects/svdquant"><b>Website</b></a> | <a href="https://hanlab.mit.edu/blog/svdquant"><b>Blog</b></a> | <a href="https://svdquant.mit.edu"><b>Demo</b></a> | <a href="https://huggingface.co/collections/mit-han-lab/svdquant-67493c2c2e62a1fc6e93f45c"><b>HuggingFace</b></a> | <a href="https://modelscope.cn/collections/svdquant-468e8f780c2641"><b>ModelScope</b></a> | <a href="https://github.com/mit-han-lab/ComfyUI-nunchaku"><b>ComfyUI</b></a>
</h3>
**Nunchaku** is a high-performance inference engine optimized for 4-bit diffusion models, as introduced in our paper [SVDQuant](http://arxiv.org/abs/2411.05007). For the underlying quantization library, check out [DeepCompressor](https://github.com/mit-han-lab/deepcompressor).
**Nunchaku** is a high-performance inference engine optimized for 4-bit neural networks, as introduced in our paper [SVDQuant](http://arxiv.org/abs/2411.05007). For the underlying quantization library, check out [DeepCompressor](https://github.com/mit-han-lab/deepcompressor).
Join our user groups on [**Slack**](https://join.slack.com/t/nunchaku/shared_invite/zt-3170agzoz-NgZzWaTrEj~n2KEV3Hpl5Q) and [**WeChat**](./assets/wechat.jpg) to engage in discussions with the community! More details can be found [here](https://github.com/mit-han-lab/nunchaku/issues/149). If you have any questions, run into issues, or are interested in contributing, don’t hesitate to reach out!
## News
- **[2025-04-05]** 🚀 **Nunchaku v0.2.0 released!** This release brings [**multi-LoRA**](examples/flux.1-dev-multiple-lora.py) and [**ControlNet**](examples/flux.1-dev-controlnet-union-pro.py) support with even faster performance powered by [**FP16 attention**](#fp16-attention) and [**First-Block Cache**](#first-block-cache). We've also added compatibility for [**20-series GPUs**](examples/flux.1-dev-turing.py) — Nunchaku is now more accessible than ever!
- **[2025-03-17]** 🚀 Released NVFP4 4-bit [Shuttle-Jaguar](https://huggingface.co/mit-han-lab/svdq-int4-shuttle-jaguar) and FLUX.1-tools and also upgraded the INT4 FLUX.1-tool models. Download and update your models from our [HuggingFace](https://huggingface.co/collections/mit-han-lab/svdquant-67493c2c2e62a1fc6e93f45c) or [ModelScope](https://modelscope.cn/collections/svdquant-468e8f780c2641) collections!
- **[2025-03-13]** 📦 Separate the ComfyUI node into a [standalone repository](https://github.com/mit-han-lab/ComfyUI-nunchaku) for easier installation and release node v0.1.6! Plus, [4-bit Shuttle-Jaguar](https://huggingface.co/mit-han-lab/svdq-int4-shuttle-jaguar) is now fully supported!
- **[2025-03-07]** 🚀 **Nunchaku v0.1.4 Released!** We've supported [4-bit text encoder and per-layer CPU offloading](#Low-Memory-Inference), reducing FLUX's minimum memory requirement to just **4 GiB** while maintaining a **2–3× speedup**. This update also fixes various issues related to resolution, LoRA, pin memory, and runtime stability. Check out the release notes for full details!
- **[2025-02-20]** 🚀 We release the [pre-built wheels](https://huggingface.co/mit-han-lab/nunchaku) to simplify installation! Check [here](#Installation) for the guidance!
- **[2025-02-20]** 🚀 **Support NVFP4 precision on NVIDIA RTX 5090!** NVFP4 delivers superior image quality compared to INT4, offering **~3× speedup** on the RTX 5090 over BF16. Learn more in our [blog](https://hanlab.mit.edu/blog/svdquant-nvfp4), checkout [`examples`](./examples) for usage and try [our demo](https://svdquant.mit.edu/flux1-schnell/) online!
- **[2025-02-18]** 🔥 [**Customized LoRA conversion**](#Customized-LoRA) and [**model quantization**](#Customized-Model-Quantization) instructions are now available! **[ComfyUI](./comfyui)** workflows now support **customized LoRA**, along with **FLUX.1-Tools**!
- **[2025-02-14]** 🔥 **[LoRA conversion script](nunchaku/convert_lora.py)** is now available! [ComfyUI FLUX.1-tools workflows](./comfyui) is released!
- **[2025-02-11]** 🎉 **[SVDQuant](http://arxiv.org/abs/2411.05007) has been selected as a ICLR 2025 Spotlight! FLUX.1-tools Gradio demos are now available!** Check [here](#gradio-demos) for the usage details! Our new [depth-to-image demo](https://svdquant.mit.edu/flux1-depth-dev/) is also online—try it out!
<details>
<summary>More</summary>
- **[2025-02-04]** **🚀 4-bit [FLUX.1-tools](https://blackforestlabs.ai/flux-1-tools/) is here!** Enjoy a **2-3× speedup** over the original models. Check out the [examples](./examples) for usage. **ComfyUI integration is coming soon!**
- **[2025-01-23]** 🚀 **4-bit [SANA](https://nvlabs.github.io/Sana/) support is here!** Experience a 2-3× speedup compared to the 16-bit model. Check out the [usage example](./examples/int4-sana_1600m_pag.py) and the [deployment guide](app/sana/t2i) for more details. Explore our live demo at [svdquant.mit.edu](https://svdquant.mit.edu)!
- **[2025-01-23]** 🚀 **4-bit [SANA](https://nvlabs.github.io/Sana/) support is here!** Experience a 2-3× speedup compared to the 16-bit model. Check out the [usage example](./examples/sana_1600m_pag.py) and the [deployment guide](app/sana/t2i) for more details. Explore our live demo at [svdquant.mit.edu](https://svdquant.mit.edu)!
- **[2025-01-22]** 🎉 [**SVDQuant**](http://arxiv.org/abs/2411.05007) has been accepted to **ICLR 2025**!
- **[2024-12-08]** Support [ComfyUI](https://github.com/comfyanonymous/ComfyUI). Please check [mit-han-lab/ComfyUI-nunchaku](https://github.com/mit-han-lab/ComfyUI-nunchaku) for the usage.
- **[2024-11-07]** 🔥 Our latest **W4A4** Diffusion model quantization work [**SVDQuant**](https://hanlab.mit.edu/projects/svdquant) is publicly released! Check [**DeepCompressor**](https://github.com/mit-han-lab/deepcompressor) for the quantization library.
......@@ -64,184 +65,197 @@ SVDQuant is a post-training quantization technique for 4-bit weights and activat
### Wheels
**Note:** For native Windows users, we have released a preliminary wheel to ease the installation. See [here](https://github.com/mit-han-lab/nunchaku/issues/169) for more details!
#### For Windows WSL Users
To install and use WSL (Windows Subsystem for Linux), follow the instructions [here](https://learn.microsoft.com/en-us/windows/wsl/install). You can also install WSL directly by running the following commands in PowerShell:
```shell
wsl --install # install the latest WSL
wsl # launch WSL
```
#### Prerequisites for all users
#### Prerequisites
Before installation, ensure you have [PyTorch>=2.5](https://pytorch.org/) installed. For example, you can use the following command to install PyTorch 2.6:
```shell
pip install torch==2.6 torchvision==0.21 torchaudio==2.6
```
#### Installing nunchaku
Once PyTorch is installed, you can directly install `nunchaku` from our [Hugging Face repository](https://huggingface.co/mit-han-lab/nunchaku/tree/main). Be sure to select the appropriate wheel for your Python and PyTorch version. For example, for Python 3.11 and PyTorch 2.6:
#### Install nunchaku
Once PyTorch is installed, you can directly install `nunchaku` from our wheel repositories on [Hugging Face](https://huggingface.co/mit-han-lab/nunchaku/tree/main) or [ModelScope](https://modelscope.cn/models/Lmxyy1999/nunchaku) or [GitHub release](https://github.com/mit-han-lab/nunchaku/releases). Be sure to select the appropriate wheel for your Python and PyTorch version. For example, for Python 3.11 and PyTorch 2.6:
```shell
pip install https://huggingface.co/mit-han-lab/nunchaku/resolve/main/nunchaku-0.1.4+torch2.6-cp311-cp311-linux_x86_64.whl
pip install https://huggingface.co/mit-han-lab/nunchaku/resolve/main/nunchaku-0.2.0+torch2.6-cp311-cp311-linux_x86_64.whl
```
**Note**: NVFP4 wheels are not currently available because PyTorch has not officially supported CUDA 11.8. To use NVFP4, you will need **Blackwell GPUs (e.g., 50-series GPUs)** and must **build from source**.
**Note**: If you're using a Blackwell GPU (e.g., 50-series GPUs), install a wheel with PyTorch 2.7. Additionally, use **FP4 models** instead of INT4 models.
### Build from Source
**Note**:
* Ensure your CUDA version is **≥ 12.2 on Linux** and **≥ 12.6 on Windows**.
* Make sure your CUDA version is **at least 12.2 on Linux** and **at least 12.6 on Windows**. If you're using a Blackwell GPU (e.g., 50-series GPUs), CUDA **12.8 or higher is required**.
* For Windows users, please refer to [this issue](https://github.com/mit-han-lab/nunchaku/issues/6) for the instruction. Please upgrade your MSVC compiler to the latest version.
* We currently support only NVIDIA GPUs with architectures sm_86 (Ampere: RTX 3090, A6000), sm_89 (Ada: RTX 4090), and sm_80 (A100). See [this issue](https://github.com/mit-han-lab/nunchaku/issues/1) for more details.
* We currently support only NVIDIA GPUs with architectures sm_75 (Turing: RTX 2080), sm_86 (Ampere: RTX 3090, A6000), sm_89 (Ada: RTX 4090), and sm_80 (A100). See [this issue](https://github.com/mit-han-lab/nunchaku/issues/1) for more details.
1. Install dependencies:
```shell
conda create -n nunchaku python=3.11
conda activate nunchaku
pip install torch torchvision torchaudio
pip install ninja wheel diffusers transformers accelerate sentencepiece protobuf huggingface_hub
pip install peft opencv-python gradio spaces GPUtil # For gradio demos
```
To enable NVFP4 on Blackwell GPUs (e.g., 50-series GPUs), please install nightly PyTorch with CUDA 12.8. The installation command can be:
```shell
conda create -n nunchaku python=3.11
conda activate nunchaku
pip install torch torchvision torchaudio
pip install ninja wheel diffusers transformers accelerate sentencepiece protobuf huggingface_hub
# For gradio demos
pip install peft opencv-python gradio spaces GPUtil
```
```shell
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
```
To enable NVFP4 on Blackwell GPUs (e.g., 50-series GPUs), please install nightly PyTorch with CUDA 12.8. The installation command can be:
```shell
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
```
2. Install `nunchaku` package:
Make sure you have `gcc/g++>=11`. If you don't, you can install it via Conda:
Make sure you have `gcc/g++>=11`. If you don't, you can install it via Conda on Linux:
```shell
conda install -c conda-forge gxx=11 gcc=11
```
Then build the package from source:
For Windows users, you can download and install the lastest [Visual Studio](https://visualstudio.microsoft.com/thank-you-downloading-visual-studio/?sku=Community&channel=Release&version=VS2022&source=VSLandingPage&cid=2030&passive=false).
Then build the package from source with
```shell
git clone https://github.com/mit-han-lab/nunchaku.git
cd nunchaku
git submodule init
git submodule update
pip install -e . --no-build-isolation
python setup.py develop
```
If you are building wheels for distribution, use:
```shell
NUNCHAKU_INSTALL_MODE=ALL NUNCHAKU_BUILD_WHEELS=1 python -m build --wheel --no-isolation
```
Make sure to set the environment variable `NUNCHAKU_INSTALL_MODE` to `ALL`. Otherwise, the generated wheels will only work on GPUs with the same architecture as the build machine.
### Docker (Coming soon)
**[Optional]** You can verify your installation by running: `python -m nunchaku.test`. This command will download and run our 4-bit FLUX.1-schnell model.
## Usage Example
In [examples](examples), we provide minimal scripts for running INT4 [FLUX.1](https://github.com/black-forest-labs/flux) and [SANA](https://github.com/NVlabs/Sana) models with Nunchaku. For example, the [script](examples/int4-flux.1-dev.py) for [FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev) is as follows:
In [examples](examples), we provide minimal scripts for running INT4 [FLUX.1](https://github.com/black-forest-labs/flux) and [SANA](https://github.com/NVlabs/Sana) models with Nunchaku. It shares the same APIs as [diffusers](https://github.com/huggingface/diffusers) and can be used in a similar way. For example, the [script](examples/flux.1-dev.py) for [FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev) is as follows:
```python
import torch
from diffusers import FluxPipeline
from nunchaku import NunchakuFluxTransformer2dModel
from nunchaku.utils import get_precision
transformer = NunchakuFluxTransformer2dModel.from_pretrained("mit-han-lab/svdq-int4-flux.1-dev")
precision = get_precision() # auto-detect your precision is 'int4' or 'fp4' based on your GPU
transformer = NunchakuFluxTransformer2dModel.from_pretrained(f"mit-han-lab/svdq-{precision}-flux.1-dev")
pipeline = FluxPipeline.from_pretrained(
"black-forest-labs/FLUX.1-dev", transformer=transformer, torch_dtype=torch.bfloat16
).to("cuda")
image = pipeline("A cat holding a sign that says hello world", num_inference_steps=50, guidance_scale=3.5).images[0]
image.save("flux.1-dev.png")
image.save(f"flux.1-dev-{precision}.png")
```
Specifically, `nunchaku` shares the same APIs as [diffusers](https://github.com/huggingface/diffusers) and can be used in a similar way.
**Note**: If you're using a **Turing GPU (e.g., NVIDIA 20-series)**, make sure to set `torch_dtype=torch.float16` and use our `nunchaku-fp16` attention module as below. A complete example is available in [`examples/flux.1-dev-turing.py`](examples/flux.1-dev-turing.py).
### Low Memory Inference
### FP16 Attention
To further reduce GPU memory usage, you can use our 4-bit T5 encoder along with CPU offloading, requiring a minimum of just 4GiB of memory. The usage is also simple in the diffusers' way. For example, the [script](examples/int4-flux.1-dev-qencoder.py) for FLUX.1-dev is as follows:
In addition to FlashAttention-2, Nunchaku introduces a custom FP16 attention implementation that achieves up to **1.2× faster performance** on NVIDIA 30-, 40-, and even 50-series GPUs—without loss in precision. To enable it, simply use:
```python
import torch
from diffusers import FluxPipeline
from nunchaku import NunchakuFluxTransformer2dModel, NunchakuT5EncoderModel
transformer = NunchakuFluxTransformer2dModel.from_pretrained(
"mit-han-lab/svdq-int4-flux.1-dev", offload=True
) # set offload to False if you want to disable offloading
text_encoder_2 = NunchakuT5EncoderModel.from_pretrained("mit-han-lab/svdq-flux.1-t5")
pipeline = FluxPipeline.from_pretrained(
"black-forest-labs/FLUX.1-dev", text_encoder_2=text_encoder_2, transformer=transformer, torch_dtype=torch.bfloat16
).to("cuda")
pipeline.enable_sequential_cpu_offload() # remove this line if you want to disable the CPU offloading
image = pipeline("A cat holding a sign that says hello world", num_inference_steps=50, guidance_scale=3.5).images[0]
image.save("flux.1-dev.png")
transformer.set_attention_impl("nunchaku-fp16")
```
## Customized LoRA
See [`examples/flux.1-dev-fp16attn.py`](examples/flux.1-dev-fp16attn.py) for a complete example.
![lora](./assets/lora.jpg)
### First-Block Cache
[SVDQuant](http://arxiv.org/abs/2411.05007) seamlessly integrates with off-the-shelf LoRAs without requiring requantization. To convert your LoRA safetensors to our format, use the following command:
Nunchaku supports [First-Block Cache](https://github.com/chengzeyi/ParaAttention?tab=readme-ov-file#first-block-cache-our-dynamic-caching) to accelerate long-step denoising. Enable it easily with:
```shell
python -m nunchaku.lora.flux.convert \
--quant-path mit-han-lab/svdq-int4-flux.1-dev/transformer_blocks.safetensors \
--lora-path aleksa-codes/flux-ghibsky-illustration/lora.safetensors \
--output-root ./nunchaku_loras \
--lora-name svdq-int4-flux.1-dev-ghibsky
```python
apply_cache_on_pipe(pipeline, residual_diff_threshold=0.12)
```
Argument Details:
You can tune the `residual_diff_threshold` to balance speed and quality: larger values yield faster inference at the cost of some quality. A recommended value is `0.12`, which provides up to **2× speedup** for 50-step denoising and **1.4× speedup** for 30-step denoising. See the full example in [`examples/flux.1-dev-cache.py`](examples/flux.1-dev-cache.py).
- `--quant-path`: The path to the quantized base model. It can be a local path or a remote Hugging Face model. For example, you can use [`mit-han-lab/svdq-int4-flux.1-dev/transformer_blocks.safetensors`](https://huggingface.co/mit-han-lab/svdq-int4-flux.1-dev/blob/main/transformer_blocks.safetensors) for FLUX.1-dev.
### CPU Offloading
- `--lora-path`: The path to your LoRA safetensors, which can also be a local or remote Hugging Face model.
To minimize GPU memory usage, Nunchaku supports CPU offloading—requiring as little as **4 GiB** of GPU memory. You can enable it by setting `offload=True` when initializing `NunchakuFluxTransformer2dModel`, and then calling:
- `--lora-format`: Specifies the LoRA format. Supported formats include:
- `auto`: The default option. Automatically detects the appropriate LoRA format.
- `diffusers` (e.g., [aleksa-codes/flux-ghibsky-illustration](https://huggingface.co/aleksa-codes/flux-ghibsky-illustration))
- `comfyui` (e.g., [Shakker-Labs/FLUX.1-dev-LoRA-Children-Simple-Sketch](https://huggingface.co/Shakker-Labs/FLUX.1-dev-LoRA-Children-Simple-Sketch))
- `xlab` (e.g., [XLabs-AI/flux-RealismLora](https://huggingface.co/XLabs-AI/flux-RealismLora))
- `--output-root`: Specifies the output directory for the converted LoRA.
```python
pipeline.enable_sequential_cpu_offload()
```
For a complete example, refer to [`examples/flux.1-dev-offload.py`](examples/flux.1-dev-offload.py).
## Customized LoRA
- `--lora-name`: Sets the name of the converted LoRA file (without `.safetensors` extension).
![lora](./assets/lora.jpg)
After converting your LoRA, you can use your converted weight with:
[SVDQuant](http://arxiv.org/abs/2411.05007) seamlessly integrates with off-the-shelf LoRAs without requiring requantization. You can simply use your LoRA with:
```python
transformer.update_lora_params(path_to_your_converted_lora)
transformer.update_lora_params(path_to_your_lora)
transformer.set_lora_strength(lora_strength)
```
`path_to_your_lora` can also be a remote HuggingFace path. In [examples/int4-flux.1-dev-lora.py](examples/int4-flux.1-dev-lora.py), we provide a minimal example script for running [Ghibsky](https://huggingface.co/aleksa-codes/flux-ghibsky-illustration) LoRA with SVDQuant's INT4 FLUX.1-dev:
`path_to_your_lora` can also be a remote HuggingFace path. In [`examples/flux.1-dev-lora.py`](examples/flux.1-dev-lora.py), we provide a minimal example script for running [Ghibsky](https://huggingface.co/aleksa-codes/flux-ghibsky-illustration) LoRA with SVDQuant's 4-bit FLUX.1-dev:
```python
import torch
from diffusers import FluxPipeline
from nunchaku import NunchakuFluxTransformer2dModel
from nunchaku.utils import get_precision
transformer = NunchakuFluxTransformer2dModel.from_pretrained("mit-han-lab/svdq-int4-flux.1-dev")
precision = get_precision() # auto-detect your precision is 'int4' or 'fp4' based on your GPU
transformer = NunchakuFluxTransformer2dModel.from_pretrained(f"mit-han-lab/svdq-{precision}-flux.1-dev")
pipeline = FluxPipeline.from_pretrained(
"black-forest-labs/FLUX.1-dev", transformer=transformer, torch_dtype=torch.bfloat16
).to("cuda")
### LoRA Related Code ###
transformer.update_lora_params(
"mit-han-lab/svdquant-lora-collection/svdq-int4-flux.1-dev-ghibsky.safetensors"
) # Path to your converted LoRA safetensors, can also be a remote HuggingFace path
"aleksa-codes/flux-ghibsky-illustration/lora.safetensors"
) # Path to your LoRA safetensors, can also be a remote HuggingFace path
transformer.set_lora_strength(1) # Your LoRA strength here
### End of LoRA Related Code ###
image = pipeline(
"GHIBSKY style, cozy mountain cabin covered in snow, with smoke curling from the chimney and a warm, inviting light spilling through the windows",
"GHIBSKY style, cozy mountain cabin covered in snow, with smoke curling from the chimney and a warm, inviting light spilling through the windows", # noqa: E501
num_inference_steps=25,
guidance_scale=3.5,
).images[0]
image.save("flux.1-dev-ghibsky.png")
image.save(f"flux.1-dev-ghibsky-{precision}.png")
```
To compose multiple LoRAs, you can use `nunchaku.lora.flux.compose.compose_lora` to compose them. The usage is
```python
composed_lora = compose_lora(
[
("PATH_OR_STATE_DICT_OF_LORA1", lora_strength1),
("PATH_OR_STATE_DICT_OF_LORA2", lora_strength2),
# Add more LoRAs as needed
]
) # set your lora strengths here when using composed lora
transformer.update_lora_params(composed_lora)
```
**For ComfyUI users, we have implemented a node to convert the LoRA weights on the fly. All you need to do is specify the correct LoRA format. Please refer to [mit-han-lab/ComfyUI-nunchaku](https://github.com/mit-han-lab/ComfyUI-nunchaku) for more details.**
You can specify individual strengths for each LoRA in the list. For a complete example, refer to [`examples/flux.1-dev-multiple-lora.py`](examples/flux.1-dev-multiple-lora.py).
**For ComfyUI users, you can directly use our LoRA loader. The converted LoRA is deprecated. Please refer to [mit-han-lab/ComfyUI-nunchaku](https://github.com/mit-han-lab/ComfyUI-nunchaku) for more details.**
## ControlNets
Nunchaku supports both the [FLUX.1-tools](https://blackforestlabs.ai/flux-1-tools/) and the [FLUX.1-dev-ControlNet-Union-Pro](https://huggingface.co/Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro) models. Example scripts can be found in the [`examples`](examples) directory.
![control](./assets/control.jpg)
## ComfyUI
......@@ -260,7 +274,7 @@ Please refer to [mit-han-lab/ComfyUI-nunchaku](https://github.com/mit-han-lab/Co
## Customized Model Quantization
Please refer to [mit-han-lab/deepcompressor](https://github.com/mit-han-lab/deepcompressor/tree/main/examples/diffusion).
Please refer to [mit-han-lab/deepcompressor](https://github.com/mit-han-lab/deepcompressor/tree/main/examples/diffusion). A simpler workflow is coming soon.
## Benchmark
......@@ -303,4 +317,4 @@ We thank MIT-IBM Watson AI Lab, MIT and Amazon Science Hub, MIT AI Hardware Prog
We use [img2img-turbo](https://github.com/GaParmar/img2img-turbo) to train the sketch-to-image LoRA. Our text-to-image and sketch-to-image UI is built upon [playground-v.25](https://huggingface.co/spaces/playgroundai/playground-v2.5/blob/main/app.py) and [img2img-turbo](https://github.com/GaParmar/img2img-turbo/blob/main/gradio_sketch2image.py), respectively. Our safety checker is borrowed from [hart](https://github.com/mit-han-lab/hart).
Nunchaku is also inspired by many open-source libraries, including (but not limited to) [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM), [vLLM](https://github.com/vllm-project/vllm), [QServe](https://github.com/mit-han-lab/qserve), [AWQ](https://github.com/mit-han-lab/llm-awq), [FlashAttention-2](https://github.com/Dao-AILab/flash-attention), and [Atom](https://github.com/efeslab/Atom).
Nunchaku is also inspired by many open-source libraries, including (but not limited to) [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM), [vLLM](https://github.com/vllm-project/vllm), [QServe](https://github.com/mit-han-lab/qserve), [AWQ](https://github.com/mit-han-lab/llm-awq), [FlashAttention-2](https://github.com/Dao-AILab/flash-attention), and [Atom](https://github.com/efeslab/Atom).
\ No newline at end of file
app/flux.1/t2i/data/DCI/DCI.py
View file @ ad8097b9
......@@ -17,8 +17,8 @@ _CITATION = """\
"""
_DESCRIPTION = """\
The Densely Captioned Images dataset, or DCI, consists of 7805 images from SA-1B,
each with a complete description aiming to capture the full visual detail of what is present in the image.
The Densely Captioned Images dataset, or DCI, consists of 7805 images from SA-1B,
each with a complete description aiming to capture the full visual detail of what is present in the image.
Much of the description is directly aligned to submasks of the image.
"""
......
app/flux.1/t2i/data/MJHQ/MJHQ.py
View file @ ad8097b9
......@@ -7,7 +7,7 @@ from PIL import Image
_CITATION = """\
@misc{li2024playground,
title={Playground v2.5: Three Insights towards Enhancing Aesthetic Quality in Text-to-Image Generation},
title={Playground v2.5: Three Insights towards Enhancing Aesthetic Quality in Text-to-Image Generation},
author={Daiqing Li and Aleks Kamko and Ehsan Akhgari and Ali Sabet and Linmiao Xu and Suhail Doshi},
year={2024},
eprint={2402.17245},
......@@ -17,7 +17,7 @@ _CITATION = """\
"""
_DESCRIPTION = """\
We introduce a new benchmark, MJHQ-30K, for automatic evaluation of a model’s aesthetic quality.
We introduce a new benchmark, MJHQ-30K, for automatic evaluation of a model’s aesthetic quality.
The benchmark computes FID on a high-quality dataset to gauge aesthetic quality.
"""
......
comfyui/LICENCE.txt deleted 100644 → 0
View file @ 804a6d30
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [2024] [MIT HAN Lab]
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.
\ No newline at end of file
comfyui/README.md deleted 100644 → 0
View file @ 804a6d30
# SVDQuant ComfyUI Node
**Note**: This node is **deprecated**! Please check **[ComfyUI-nunchaku/](https://github.com/mit-han-lab/ComfyUI-nunchaku/)** for the latest version.
![comfyui](../assets/comfyui.jpg)
## Installation
Please first install `nunchaku` following the instructions in [README.md](https://github.com/mit-han-lab/nunchaku?tab=readme-ov-file#installation). Then just install `image_gen_aux` with
```shell
pip install git+https://github.com/asomoza/image_gen_aux.git
```
### ComfyUI-CLI
```shell
pip install comfy-cli # install the comfyui-cli
comfy node registry-install svdquant
```
### ComfyUI-Manager (Experimental)
1. Install [ComfyUI-Manager](https://github.com/ltdrdata/ComfyUI-Manager) with the following commands then restart ComfyUI:
```shell
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager
```
2. Open the Manager, search `svdquant` in the Custom Nodes Manager and then install it.
### Manual Installation
1. Install dependencies needed to run custom ComfyUI nodes:
```shell
pip install git+https://github.com/asomoza/image_gen_aux.git
```
2. Set up the dependencies for [ComfyUI](https://github.com/comfyanonymous/ComfyUI/tree/master) with the following commands:
```shell
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
```
3. Navigate to the root directory of ComfyUI and link (or copy) the [`nunchaku/comfyui`](./) folder to `custom_nodes/svdquant`. For example:
```shell
# Clone repositories (skip if already cloned)
git clone https://github.com/comfyanonymous/ComfyUI.git
git clone https://github.com/mit-han-lab/nunchaku.git
cd ComfyUI
# Add SVDQuant nodes
cd custom_nodes
ln -s ../../nunchaku/comfyui svdquant
```
## Usage
1. **Set Up ComfyUI and SVDQuant**:
* SVDQuant workflows can be found at [`workflows`](./workflows). You can place them in `user/default/workflows` in ComfyUI root directory to load them. For example:
```shell
cd ComfyUI
# Copy workflow configurations
mkdir -p user/default/workflows
cp ../nunchaku/comfyui/workflows/* user/default/workflows/
```
* Install missing nodes (e.g., comfyui-inpainteasy) following [this tutorial](https://github.com/ltdrdata/ComfyUI-Manager?tab=readme-ov-file#support-of-missing-nodes-installation).
2. **Download Required Models**: Follow [this tutorial](https://comfyanonymous.github.io/ComfyUI_examples/flux/) and download the required models into the appropriate directories using the commands below:
```shell
huggingface-cli download comfyanonymous/flux_text_encoders clip_l.safetensors --local-dir models/text_encoders
huggingface-cli download comfyanonymous/flux_text_encoders t5xxl_fp16.safetensors --local-dir models/text_encoders
huggingface-cli download black-forest-labs/FLUX.1-schnell ae.safetensors --local-dir models/vae
```
3. **Run ComfyUI**: From ComfyUI’s root directory, execute the following command to start the application:
```shell
python main.py
```
4. **Select the SVDQuant Workflow**: Choose one of the SVDQuant workflows (workflows that start with `svdq-`) to get started. For the FLUX.1-Fill workflow, you can use the built-in MaskEditor tool to add mask on top of an image.
## SVDQuant Nodes
* **SVDQuant Flux DiT Loader**: A node for loading the FLUX diffusion model.
* `model_path`: Specifies the model location. If set to the folder starting with `mit-han-lab`, the model will be automatically downloaded from our Hugging Face repository. Alternatively, you can manually download the model directory by running the following command example:
```shell
huggingface-cli download mit-han-lab/svdq-int4-flux.1-dev --local-dir models/diffusion_models/svdq-int4-flux.1-dev
```
After downloading, specify the corresponding folder name as the `model_path`.
* `cpu_offload`: Enables CPU offloading for the transformer model. While this may reduce GPU memory usage, it can slow down inference. Memory usage will be further optimized in node v0.1.6.
* `device_id`: Indicates the GPU ID for running the model.
* **SVDQuant FLUX LoRA Loader**: A node for loading LoRA modules for SVDQuant FLUX models.
* Place your LoRA checkpoints in the `models/loras` directory. These will appear as selectable options under `lora_name`. Meanwhile, the [example Ghibsky LoRA](https://huggingface.co/aleksa-codes/flux-ghibsky-illustration) is included and will automatically download from our Hugging Face repository when used.
* `lora_format` specifies the LoRA format. Supported formats include:
* `auto`: Automatically detects the appropriate LoRA format.
* `diffusers` (e.g., [aleksa-codes/flux-ghibsky-illustration](https://huggingface.co/aleksa-codes/flux-ghibsky-illustration))
* `comfyui` (e.g., [Shakker-Labs/FLUX.1-dev-LoRA-Children-Simple-Sketch](https://huggingface.co/Shakker-Labs/FLUX.1-dev-LoRA-Children-Simple-Sketch))
* `xlab` (e.g., [XLabs-AI/flux-RealismLora](https://huggingface.co/XLabs-AI/flux-RealismLora))
* `svdquant` (e.g., [mit-han-lab/svdquant-lora-collection](https://huggingface.co/mit-han-lab/svdquant-lora-collection)).
* `base_model_name` specifies the path to the quantized base model. If `lora_format` is already set to `svdquant`, this option has no use. You can set it to the same value as `model_path` in the above **SVDQuant Flux DiT Loader**.
* **Note**: Currently, **only one LoRA** can be loaded at a time.
* **SVDQuant Text Encoder Loader**: A node for loading the text encoders.
* For FLUX, use the following files:
- `text_encoder1`: `t5xxl_fp16.safetensors`
- `text_encoder2`: `clip_l.safetensors`
* `t5_min_length`: Sets the minimum sequence length for T5 text embeddings. The default in `DualCLIPLoader` is hardcoded to 256, but for better image quality in SVDQuant, use 512 here.
* `t5_precision`: Specifies the precision of the T5 text encoder. Choose `INT4` to use the INT4 text encoder, which reduces GPU memory usage by approximately 15GB. Please install [`deepcompressor`](https://github.com/mit-han-lab/deepcompressor) when using it:
```shell
git clone https://github.com/mit-han-lab/deepcompressor
cd deepcompressor
pip install poetry
poetry install
```
* `int4_model`: Specifies the INT4 model location. This option is only used when `t5_precision` is set to `INT4`. By default, the path is `mit-han-lab/svdq-flux.1-t5`, and the model will automatically download from our Hugging Face repository. Alternatively, you can manually download the model directory by running the following command:
```shell
huggingface-cli download mit-han-lab/svdq-flux.1-t5 --local-dir models/text_encoders/svdq-flux.1-t5
```
After downloading, specify the corresponding folder name as the `int4_model`.
* **FLUX.1 Depth Preprocessor**: A node for loading the depth estimation model and output the depth map. `model_path` specifies the model location. If set to [`LiheYoung/depth-anything-large-hf`](https://huggingface.co/LiheYoung/depth-anything-large-hf), the model will be automatically downloaded from the Hugging Face repository. Alternatively, you can manually download the repository at `models/checkpoints` by running the following command example:
```shell
huggingface-cli download LiheYoung/depth-anything-large-hf --local-dir models/checkpoints/depth-anything-large-hf
```
comfyui/__init__.py deleted 100644 → 0
View file @ 804a6d30
# only import if running as a custom node
from .nodes.lora import SVDQuantFluxLoraLoader
from .nodes.models import SVDQuantFluxDiTLoader, SVDQuantTextEncoderLoader
from .nodes.preprocessors import FluxDepthPreprocessor
NODE_CLASS_MAPPINGS = {
"SVDQuantFluxDiTLoader": SVDQuantFluxDiTLoader,
"SVDQuantTextEncoderLoader": SVDQuantTextEncoderLoader,
"SVDQuantFluxLoraLoader": SVDQuantFluxLoraLoader,
"SVDQuantDepthPreprocessor": FluxDepthPreprocessor,
}
NODE_DISPLAY_NAME_MAPPINGS = {k: v.TITLE for k, v in NODE_CLASS_MAPPINGS.items()}
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
comfyui/nodes/lora/__init__.py deleted 100644 → 0
View file @ 804a6d30
from .flux import SVDQuantFluxLoraLoader
comfyui/nodes/lora/flux.py deleted 100644 → 0
View file @ 804a6d30
import os
import tempfile
import folder_paths
from safetensors.torch import save_file
from nunchaku.lora.flux import comfyui2diffusers, convert_to_nunchaku_flux_lowrank_dict, detect_format, xlab2diffusers
class SVDQuantFluxLoraLoader:
def __init__(self):
self.cur_lora_name = "None"
@classmethod
def INPUT_TYPES(s):
lora_name_list = [
"None",
*folder_paths.get_filename_list("loras"),
"aleksa-codes/flux-ghibsky-illustration/lora.safetensors",
]
base_model_paths = [
"mit-han-lab/svdq-int4-flux.1-dev",
"mit-han-lab/svdq-int4-flux.1-schnell",
"mit-han-lab/svdq-fp4-flux.1-dev",
"mit-han-lab/svdq-fp4-flux.1-schnell",
"mit-han-lab/svdq-int4-flux.1-canny-dev",
"mit-han-lab/svdq-int4-flux.1-depth-dev",
"mit-han-lab/svdq-int4-flux.1-fill-dev",
]
prefix = os.path.join(folder_paths.models_dir, "diffusion_models")
local_base_model_folders = os.listdir(prefix)
local_base_model_folders = sorted(
[
folder
for folder in local_base_model_folders
if not folder.startswith(".") and os.path.isdir(os.path.join(prefix, folder))
]
)
base_model_paths = local_base_model_folders + base_model_paths
return {
"required": {
"model": ("MODEL", {"tooltip": "The diffusion model the LoRA will be applied to."}),
"lora_name": (lora_name_list, {"tooltip": "The name of the LoRA."}),
"lora_format": (
["auto", "comfyui", "diffusers", "svdquant", "xlab"],
{"tooltip": "The format of the LoRA."},
),
"base_model_name": (
base_model_paths,
{
"tooltip": "If the lora format is SVDQuant, this field has no use. Otherwise, the base model's state dictionary is required for converting the LoRA weights to SVDQuant."
},
),
"lora_strength": (
"FLOAT",
{
"default": 1.0,
"min": -100.0,
"max": 100.0,
"step": 0.01,
"tooltip": "How strongly to modify the diffusion model. This value can be negative.",
},
),
}
}
RETURN_TYPES = ("MODEL",)
OUTPUT_TOOLTIPS = ("The modified diffusion model.",)
FUNCTION = "load_lora"
TITLE = "SVDQuant FLUX.1 LoRA Loader"
CATEGORY = "SVDQuant"
DESCRIPTION = (
"LoRAs are used to modify the diffusion model, "
"altering the way in which latents are denoised such as applying styles. "
"Currently, only one LoRA nodes can be applied."
)
def load_lora(self, model, lora_name: str, lora_format: str, base_model_name: str, lora_strength: float):
if self.cur_lora_name == lora_name:
if self.cur_lora_name == "None":
pass # Do nothing since the lora is None
else:
model.model.diffusion_model.model.set_lora_strength(lora_strength)
else:
if lora_name == "None":
model.model.diffusion_model.model.set_lora_strength(0)
else:
try:
lora_path = folder_paths.get_full_path_or_raise("loras", lora_name)
except FileNotFoundError:
lora_path = lora_name
if lora_format == "auto":
lora_format = detect_format(lora_path)
if lora_format != "svdquant":
if lora_format == "comfyui":
input_lora = comfyui2diffusers(lora_path)
elif lora_format == "xlab":
input_lora = xlab2diffusers(lora_path)
elif lora_format == "diffusers":
input_lora = lora_path
else:
raise ValueError(f"Invalid LoRA format {lora_format}.")
prefix = os.path.join(folder_paths.models_dir, "diffusion_models")
base_model_path = os.path.join(prefix, base_model_name, "transformer_blocks.safetensors")
if not os.path.exists(base_model_path):
# download from huggingface
base_model_path = os.path.join(base_model_name, "transformer_blocks.safetensors")
state_dict = convert_to_nunchaku_flux_lowrank_dict(base_model_path, input_lora)
with tempfile.NamedTemporaryFile(suffix=".safetensors", delete=True) as tmp_file:
save_file(state_dict, tmp_file.name)
model.model.diffusion_model.model.update_lora_params(tmp_file.name)
else:
model.model.diffusion_model.model.update_lora_params(lora_path)
model.model.diffusion_model.model.set_lora_strength(lora_strength)
self.cur_lora_name = lora_name
return (model,)
comfyui/nodes/models/__init__.py deleted 100644 → 0
View file @ 804a6d30
from .flux import SVDQuantFluxDiTLoader
from .text_encoder import SVDQuantTextEncoderLoader
comfyui/nodes/models/flux.py deleted 100644 → 0
View file @ 804a6d30
import os
import comfy.model_patcher
import folder_paths
import torch
from comfy.ldm.common_dit import pad_to_patch_size
from comfy.supported_models import Flux, FluxSchnell
from diffusers import FluxTransformer2DModel
from einops import rearrange, repeat
from torch import nn
from nunchaku import NunchakuFluxTransformer2dModel
class ComfyUIFluxForwardWrapper(nn.Module):
def __init__(self, model: NunchakuFluxTransformer2dModel, config):
super(ComfyUIFluxForwardWrapper, self).__init__()
self.model = model
self.dtype = next(model.parameters()).dtype
self.config = config
def forward(
self,
x,
timestep,
context,
y,
guidance,
control=None,
transformer_options={},
**kwargs,
):
assert control is None # for now
bs, c, h, w = x.shape
patch_size = self.config["patch_size"]
x = pad_to_patch_size(x, (patch_size, patch_size))
img = rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch_size, pw=patch_size)
h_len = (h + (patch_size // 2)) // patch_size
w_len = (w + (patch_size // 2)) // patch_size
img_ids = torch.zeros((h_len, w_len, 3), device=x.device, dtype=x.dtype)
img_ids[:, :, 1] = img_ids[:, :, 1] + torch.linspace(
0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype
).unsqueeze(1)
img_ids[:, :, 2] = img_ids[:, :, 2] + torch.linspace(
0, w_len - 1, steps=w_len, device=x.device, dtype=x.dtype
).unsqueeze(0)
img_ids = repeat(img_ids, "h w c -> b (h w) c", b=bs)
txt_ids = torch.zeros((bs, context.shape[1], 3), device=x.device, dtype=x.dtype)
out = self.model(
hidden_states=img,
encoder_hidden_states=context,
pooled_projections=y,
timestep=timestep,
img_ids=img_ids,
txt_ids=txt_ids,
guidance=guidance if self.config["guidance_embed"] else None,
).sample
out = rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)", h=h_len, w=w_len, ph=2, pw=2)[:, :, :h, :w]
return out
class SVDQuantFluxDiTLoader:
@classmethod
def INPUT_TYPES(s):
model_paths = [
"mit-han-lab/svdq-int4-flux.1-schnell",
"mit-han-lab/svdq-int4-flux.1-dev",
"mit-han-lab/svdq-fp4-flux.1-schnell",
"mit-han-lab/svdq-fp4-flux.1-dev",
"mit-han-lab/svdq-int4-flux.1-canny-dev",
"mit-han-lab/svdq-int4-flux.1-depth-dev",
"mit-han-lab/svdq-int4-flux.1-fill-dev",
]
prefixes = folder_paths.folder_names_and_paths["diffusion_models"][0]
local_folders = set()
for prefix in prefixes:
if os.path.exists(prefix) and os.path.isdir(prefix):
local_folders_ = os.listdir(prefix)
local_folders_ = [
folder
for folder in local_folders_
if not folder.startswith(".") and os.path.isdir(os.path.join(prefix, folder))
]
local_folders.update(local_folders_)
local_folders = sorted(list(local_folders))
model_paths = local_folders + model_paths
ngpus = torch.cuda.device_count()
return {
"required": {
"model_path": (
model_paths,
{"tooltip": "The SVDQuant quantized FLUX.1 models. It can be a huggingface path or a local path."},
),
"cpu_offload": (
["auto", "enable", "disable"],
{
"default": "auto",
"tooltip": "Whether to enable CPU offload for the transformer model. 'auto' will enable it if the GPU memory is less than 14G.",
},
),
"device_id": (
"INT",
{
"default": 0,
"min": 0,
"max": ngpus - 1,
"step": 1,
"display": "number",
"lazy": True,
"tooltip": "The GPU device ID to use for the model.",
},
),
}
}
RETURN_TYPES = ("MODEL",)
FUNCTION = "load_model"
CATEGORY = "SVDQuant"
TITLE = "SVDQuant Flux DiT Loader"
def load_model(self, model_path: str, cpu_offload: str, device_id: int, **kwargs) -> tuple[FluxTransformer2DModel]:
device = f"cuda:{device_id}"
prefixes = folder_paths.folder_names_and_paths["diffusion_models"][0]
for prefix in prefixes:
if os.path.exists(os.path.join(prefix, model_path)):
model_path = os.path.join(prefix, model_path)
break
# 验证 device_id 是否有效
if device_id >= torch.cuda.device_count():
raise ValueError(f"Invalid device_id: {device_id}. Only {torch.cuda.device_count()} GPUs available.")
# 获取 ComfyUI 指定 CUDA 设备的显存信息
gpu_properties = torch.cuda.get_device_properties(device_id)
gpu_memory = gpu_properties.total_memory / (1024 ** 2) # 转换为 MB
gpu_name = gpu_properties.name
print(f"GPU {device_id} ({gpu_name}) 显存: {gpu_memory} MB")
# 确定 CPU offload 是否启用
if cpu_offload == "auto":
if gpu_memory < 14336: # 14GB 阈值
cpu_offload_enabled = True
print("因显存小于14GB，启用 CPU offload")
else:
cpu_offload_enabled = False
print("显存大于14GB，不启用 CPU offload")
elif cpu_offload == "enable":
cpu_offload_enabled = True
print("用户启用 CPU offload")
else:
cpu_offload_enabled = False
print("用户禁用 CPU offload")
# 清理 GPU 缓存
# torch.cuda.empty_cache()
transformer = NunchakuFluxTransformer2dModel.from_pretrained(model_path, offload=cpu_offload_enabled)
transformer = transformer.to(device)
dit_config = {
"image_model": "flux",
"patch_size": 2,
"out_channels": 16,
"vec_in_dim": 768,
"context_in_dim": 4096,
"hidden_size": 3072,
"mlp_ratio": 4.0,
"num_heads": 24,
"depth": 19,
"depth_single_blocks": 38,
"axes_dim": [16, 56, 56],
"theta": 10000,
"qkv_bias": True,
"guidance_embed": True,
"disable_unet_model_creation": True,
}
if "schnell" in model_path:
dit_config["guidance_embed"] = False
dit_config["in_channels"] = 16
model_config = FluxSchnell(dit_config)
elif "canny" in model_path or "depth" in model_path:
dit_config["in_channels"] = 32
model_config = Flux(dit_config)
elif "fill" in model_path:
dit_config["in_channels"] = 64
model_config = Flux(dit_config)
else:
dit_config["in_channels"] = 16
model_config = Flux(dit_config)
model_config.set_inference_dtype(torch.bfloat16, None)
model_config.custom_operations = None
model = model_config.get_model({})
model.diffusion_model = ComfyUIFluxForwardWrapper(transformer, config=dit_config)
model = comfy.model_patcher.ModelPatcher(model, device, device_id)
return (model,)
comfyui/nodes/models/text_encoder.py deleted 100644 → 0
View file @ 804a6d30
import os
import types
import comfy.sd
import folder_paths
import torch
from torch import nn
from transformers import T5EncoderModel
from nunchaku import NunchakuT5EncoderModel
def svdquant_t5_forward(
self: T5EncoderModel,
input_ids: torch.LongTensor,
attention_mask,
embeds=None,
intermediate_output=None,
final_layer_norm_intermediate=True,
dtype: str | torch.dtype = torch.bfloat16,
**kwargs,
):
assert attention_mask is None
assert intermediate_output is None
assert final_layer_norm_intermediate
outputs = self.encoder(input_ids=input_ids, inputs_embeds=embeds, attention_mask=attention_mask)
hidden_states = outputs["last_hidden_state"]
hidden_states = hidden_states.to(dtype=dtype)
return hidden_states, None
class WrappedEmbedding(nn.Module):
def __init__(self, embedding: nn.Embedding):
super().__init__()
self.embedding = embedding
def forward(self, input: torch.Tensor, out_dtype: torch.dtype | None = None):
return self.embedding(input)
@property
def weight(self):
return self.embedding.weight
class SVDQuantTextEncoderLoader:
@classmethod
def INPUT_TYPES(s):
model_paths = ["mit-han-lab/svdq-flux.1-t5"]
prefixes = folder_paths.folder_names_and_paths["text_encoders"][0]
local_folders = set()
for prefix in prefixes:
if os.path.exists(prefix) and os.path.isdir(prefix):
local_folders_ = os.listdir(prefix)
local_folders_ = [
folder
for folder in local_folders_
if not folder.startswith(".") and os.path.isdir(os.path.join(prefix, folder))
]
local_folders.update(local_folders_)
local_folders = sorted(list(local_folders))
model_paths.extend(local_folders)
return {
"required": {
"model_type": (["flux"],),
"text_encoder1": (folder_paths.get_filename_list("text_encoders"),),
"text_encoder2": (folder_paths.get_filename_list("text_encoders"),),
"t5_min_length": (
"INT",
{"default": 512, "min": 256, "max": 1024, "step": 128, "display": "number", "lazy": True},
),
"t5_precision": (["BF16", "INT4"],),
"int4_model": (model_paths, {"tooltip": "The name of the INT4 model."}),
}
}
RETURN_TYPES = ("CLIP",)
FUNCTION = "load_text_encoder"
CATEGORY = "SVDQuant"
TITLE = "SVDQuant Text Encoder Loader"
def load_text_encoder(
self,
model_type: str,
text_encoder1: str,
text_encoder2: str,
t5_min_length: int,
t5_precision: str,
int4_model: str,
):
text_encoder_path1 = folder_paths.get_full_path_or_raise("text_encoders", text_encoder1)
text_encoder_path2 = folder_paths.get_full_path_or_raise("text_encoders", text_encoder2)
if model_type == "flux":
clip_type = comfy.sd.CLIPType.FLUX
else:
raise ValueError(f"Unknown type {model_type}")
clip = comfy.sd.load_clip(
ckpt_paths=[text_encoder_path1, text_encoder_path2],
embedding_directory=folder_paths.get_folder_paths("embeddings"),
clip_type=clip_type,
)
if model_type == "flux":
clip.tokenizer.t5xxl.min_length = t5_min_length
if t5_precision == "INT4":
transformer = clip.cond_stage_model.t5xxl.transformer
param = next(transformer.parameters())
dtype = param.dtype
device = param.device
prefixes = folder_paths.folder_names_and_paths["text_encoders"][0]
model_path = None
for prefix in prefixes:
if os.path.exists(os.path.join(prefix, int4_model)):
model_path = os.path.join(prefix, int4_model)
break
if model_path is None:
model_path = int4_model
transformer = NunchakuT5EncoderModel.from_pretrained(model_path)
transformer.forward = types.MethodType(svdquant_t5_forward, transformer)
transformer.shared = WrappedEmbedding(transformer.shared)
clip.cond_stage_model.t5xxl.transformer = (
transformer.to(device=device, dtype=dtype) if device.type == "cuda" else transformer
)
return (clip,)
comfyui/nodes/preprocessors/__init__.py deleted 100644 → 0
View file @ 804a6d30
from .depth import FluxDepthPreprocessor
comfyui/nodes/preprocessors/depth.py deleted 100644 → 0
View file @ 804a6d30
import os
import folder_paths
import numpy as np
import torch
from image_gen_aux import DepthPreprocessor
class FluxDepthPreprocessor:
@classmethod
def INPUT_TYPES(s):
model_paths = ["LiheYoung/depth-anything-large-hf"]
prefix = os.path.join(folder_paths.models_dir, "checkpoints")
local_folders = os.listdir(prefix)
local_folders = sorted(
[
folder
for folder in local_folders
if not folder.startswith(".") and os.path.isdir(os.path.join(prefix, folder))
]
)
model_paths = local_folders + model_paths
return {
"required": {
"image": ("IMAGE", {}),
"model_path": (
model_paths,
{"tooltip": "Name of the depth preprocessor model."},
),
}
}
RETURN_TYPES = ("IMAGE",)
FUNCTION = "depth_preprocess"
CATEGORY = "SVDQuant"
TITLE = "FLUX.1 Depth Preprocessor"
def depth_preprocess(self, image, model_path):
prefix = os.path.join(folder_paths.models_dir, "checkpoints")
if os.path.exists(os.path.join(prefix, model_path)):
model_path = os.path.join(prefix, model_path)
processor = DepthPreprocessor.from_pretrained(model_path)
np_image = np.asarray(image)
np_result = np.array(processor(np_image)[0].convert("RGB"))
out_tensor = torch.from_numpy(np_result.astype(np.float32) / 255.0).unsqueeze(0)
return (out_tensor,)
comfyui/pyproject.toml deleted 100644 → 0
View file @ 804a6d30
[project]
name = "svdquant"
description = "SVDQuant ComfyUI Node. SVDQuant is a new post-training training quantization paradigm for diffusion models, which quantize both the weights and activations of FLUX.1 to 4 bits, achieving 3.5× memory and 8.7× latency reduction on a 16GB laptop 4090 GPU. GitHub: https://github.com/mit-han-lab/nunchaku"
version = "0.1.5"
license = { file = "LICENSE.txt" }
dependencies = []
requires-python = ">=3.10, <3.13"
#[project.urls]
#Repository = "https://github.com/mit-han-lab/nunchaku"
# Used by Comfy Registry https://comfyregistry.org
[tool.comfy]
PublisherId = "lmxyy1999"
DisplayName = "svdquant"
Icon = "https://raw.githubusercontent.com/mit-han-lab/nunchaku/bfd9aa3ae84f51a26414f1600d725a0098472820/assets/logo.svg"
comfyui/requirements.txt deleted 100644 → 0
View file @ 804a6d30
GPUtil
diffusers>=0.32.2
accelerate
sentencepiece
protobuf
huggingface_hub
comfyui/workflows/svdq-flux.1-canny.json deleted 100644 → 0
View file @ 804a6d30
{
"last_node_id": 38,
"last_link_id": 76,
"nodes": [
{
"id": 3,
"type": "KSampler",
"pos": [
1290,
40
],
"size": [
315,
262
],
"flags": {},
"order": 11,
"mode": 0,
"inputs": [
{
"name": "model",
"localized_name": "model",
"label": "model",
"type": "MODEL",
"link": 71
},
{
"name": "positive",
"localized_name": "positive",
"label": "positive",
"type": "CONDITIONING",
"link": 64
},
{
"name": "negative",
"localized_name": "negative",
"label": "negative",
"type": "CONDITIONING",
"link": 65
},
{
"name": "latent_image",
"localized_name": "latent_image",
"label": "latent_image",
"type": "LATENT",
"link": 66
}
],
"outputs": [
{
"name": "LATENT",
"localized_name": "LATENT",
"label": "LATENT",
"type": "LATENT",
"links": [
7
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "KSampler"
},
"widgets_values": [
875054580097021,
"randomize",
20,
1,
"euler",
"normal",
1
]
},
{
"id": 35,
"type": "InstructPixToPixConditioning",
"pos": [
1040,
50
],
"size": [
235.1999969482422,
86
],
"flags": {},
"order": 10,
"mode": 0,
"inputs": [
{
"name": "positive",
"localized_name": "positive",
"label": "positive",
"type": "CONDITIONING",
"link": 67
},
{
"name": "negative",
"localized_name": "negative",
"label": "negative",
"type": "CONDITIONING",
"link": 68
},
{
"name": "vae",
"localized_name": "vae",
"label": "vae",
"type": "VAE",
"link": 69
},
{
"name": "pixels",
"localized_name": "pixels",
"label": "pixels",
"type": "IMAGE",
"link": 70
}
],
"outputs": [
{
"name": "positive",
"localized_name": "positive",
"label": "positive",
"type": "CONDITIONING",
"links": [
64
],
"slot_index": 0
},
{
"name": "negative",
"localized_name": "negative",
"label": "negative",
"type": "CONDITIONING",
"links": [
65
],
"slot_index": 1
},
{
"name": "latent",
"localized_name": "latent",
"label": "latent",
"type": "LATENT",
"links": [
66
],
"slot_index": 2
}
],
"properties": {
"Node name for S&R": "InstructPixToPixConditioning"
},
"widgets_values": []
},
{
"id": 8,
"type": "VAEDecode",
"pos": [
1620,
40
],
"size": [
210,
46
],
"flags": {},
"order": 12,
"mode": 0,
"inputs": [
{
"name": "samples",
"localized_name": "samples",
"label": "samples",
"type": "LATENT",
"link": 7
},
{
"name": "vae",
"localized_name": "vae",
"label": "vae",
"type": "VAE",
"link": 60
}
],
"outputs": [
{
"name": "IMAGE",
"localized_name": "IMAGE",
"label": "IMAGE",
"type": "IMAGE",
"links": [
9
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "VAEDecode"
},
"widgets_values": []
},
{
"id": 9,
"type": "SaveImage",
"pos": [
1850,
40
],
"size": [
828.9535522460938,
893.8475341796875
],
"flags": {},
"order": 13,
"mode": 0,
"inputs": [
{
"name": "images",
"localized_name": "images",
"label": "images",
"type": "IMAGE",
"link": 9
}
],
"outputs": [],
"properties": {},
"widgets_values": [
"ComfyUI"
]
},
{
"id": 32,
"type": "VAELoader",
"pos": [
1290,
350
],
"size": [
315,
58
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [],
"outputs": [
{
"name": "VAE",
"localized_name": "VAE",
"label": "VAE",
"type": "VAE",
"links": [
60,
69
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "VAELoader"
},
"widgets_values": [
"ae.safetensors"
]
},
{
"id": 26,
"type": "FluxGuidance",
"pos": [
700,
50
],
"size": [
317.4000244140625,
58
],
"flags": {},
"order": 7,
"mode": 0,
"inputs": [
{
"name": "conditioning",
"localized_name": "conditioning",
"label": "conditioning",
"type": "CONDITIONING",
"link": 41
}
],
"outputs": [
{
"name": "CONDITIONING",
"localized_name": "CONDITIONING",
"label": "CONDITIONING",
"type": "CONDITIONING",
"shape": 3,
"links": [
67
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "FluxGuidance"
},
"widgets_values": [
30
]
},
{
"id": 34,
"type": "DualCLIPLoader",
"pos": [
-80,
110
],
"size": [
315,
122
],
"flags": {},
"order": 1,
"mode": 0,
"inputs": [],
"outputs": [
{
"name": "CLIP",
"localized_name": "CLIP",
"label": "CLIP",
"type": "CLIP",
"links": [
62,
63
]
}
],
"properties": {
"Node name for S&R": "DualCLIPLoader"
},
"widgets_values": [
"clip_l.safetensors",
"t5xxl_fp16.safetensors",
"flux",
"default"
]
},
{
"id": 23,
"type": "CLIPTextEncode",
"pos": [
260,
50
],
"size": [
422.84503173828125,
164.31304931640625
],
"flags": {},
"order": 4,
"mode": 0,
"inputs": [
{
"name": "clip",
"localized_name": "clip",
"label": "clip",
"type": "CLIP",
"link": 62
}
],
"outputs": [
{
"name": "CONDITIONING",
"localized_name": "CONDITIONING",
"label": "CONDITIONING",
"type": "CONDITIONING",
"links": [
41
],
"slot_index": 0
}
],
"title": "CLIP Text Encode (Positive Prompt)",
"properties": {
"Node name for S&R": "CLIPTextEncode"
},
"widgets_values": [
"A robot made of exotic candies and chocolates of different kinds. The background is filled with confetti and celebratory gifts."
],
"color": "#232",
"bgcolor": "#353"
},
{
"id": 19,
"type": "PreviewImage",
"pos": [
1127.9403076171875,
554.3356323242188
],
"size": [
571.5869140625,
625.5296020507812
],
"flags": {},
"order": 9,
"mode": 0,
"inputs": [
{
"name": "images",
"localized_name": "images",
"label": "images",
"type": "IMAGE",
"link": 26
}
],
"outputs": [],
"properties": {
"Node name for S&R": "PreviewImage"
},
"widgets_values": []
},
{
"id": 18,
"type": "Canny",
"pos": [
744.2684936523438,
566.853515625
],
"size": [
315,
82
],
"flags": {},
"order": 8,
"mode": 0,
"inputs": [
{
"name": "image",
"localized_name": "image",
"label": "image",
"type": "IMAGE",
"link": 76
}
],
"outputs": [
{
"name": "IMAGE",
"localized_name": "IMAGE",
"label": "IMAGE",
"type": "IMAGE",
"shape": 3,
"links": [
26,
70
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "Canny"
},
"widgets_values": [
0.15,
0.3
]
},
{
"id": 38,
"type": "ImageScale",
"pos": [
379.69903564453125,
565.2651977539062
],
"size": [
315,
130
],
"flags": {},
"order": 6,
"mode": 0,
"inputs": [
{
"name": "image",
"localized_name": "image",
"type": "IMAGE",
"link": 75
}
],
"outputs": [
{
"name": "IMAGE",
"localized_name": "IMAGE",
"type": "IMAGE",
"links": [
76
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "ImageScale"
},
"widgets_values": [
"nearest-exact",
1024,
1024,
"center"
]
},
{
"id": 7,
"type": "CLIPTextEncode",
"pos": [
323.8695068359375,
387.9589538574219
],
"size": [
425.27801513671875,
180.6060791015625
],
"flags": {
"collapsed": true
},
"order": 5,
"mode": 0,
"inputs": [
{
"name": "clip",
"localized_name": "clip",
"label": "clip",
"type": "CLIP",
"link": 63
}
],
"outputs": [
{
"name": "CONDITIONING",
"localized_name": "CONDITIONING",
"label": "CONDITIONING",
"type": "CONDITIONING",
"links": [
68
],
"slot_index": 0
}
],
"title": "CLIP Text Encode (Negative Prompt)",
"properties": {
"Node name for S&R": "CLIPTextEncode"
},
"widgets_values": [
""
],
"color": "#322",
"bgcolor": "#533"
},
{
"id": 17,
"type": "LoadImage",
"pos": [
6.694743633270264,
562.3865966796875
],
"size": [
315,
314.0000305175781
],
"flags": {},
"order": 2,
"mode": 0,
"inputs": [],
"outputs": [
{
"name": "IMAGE",
"localized_name": "IMAGE",
"label": "IMAGE",
"type": "IMAGE",
"shape": 3,
"links": [
75
],
"slot_index": 0
},
{
"name": "MASK",
"localized_name": "MASK",
"label": "MASK",
"type": "MASK",
"shape": 3,
"links": null
}
],
"properties": {
"Node name for S&R": "LoadImage"
},
"widgets_values": [
"robot.png",
"image"
]
},
{
"id": 36,
"type": "SVDQuantFluxDiTLoader",
"pos": [
823.9686279296875,
-126.4416275024414
],
"size": [
395.6002197265625,
106
],
"flags": {},
"order": 3,
"mode": 0,
"inputs": [],
"outputs": [
{
"name": "MODEL",
"localized_name": "MODEL",
"type": "MODEL",
"links": [
71
],
"slot_index": 0
}
],
"properties": {
"Node name for S&R": "SVDQuantFluxDiTLoader"
},
"widgets_values": [
"mit-han-lab/svdq-int4-flux.1-canny-dev",
"disable",
0
]
}
],
"links": [
[
7,
3,
0,
8,
0,
"LATENT"
],
[
9,
8,
0,
9,
0,
"IMAGE"
],
[
26,
18,
0,
19,
0,
"IMAGE"
],
[
41,
23,
0,
26,
0,
"CONDITIONING"
],
[
60,
32,
0,
8,
1,
"VAE"
],
[
62,
34,
0,
23,
0,
"CLIP"
],
[
63,
34,
0,
7,
0,
"CLIP"
],
[
64,
35,
0,
3,
1,
"CONDITIONING"
],
[
65,
35,
1,
3,
2,
"CONDITIONING"
],
[
66,
35,
2,
3,
3,
"LATENT"
],
[
67,
26,
0,
35,
0,
"CONDITIONING"
],
[
68,
7,
0,
35,
1,
"CONDITIONING"
],
[
69,
32,
0,
35,
2,
"VAE"
],
[
70,
18,
0,
35,
3,
"IMAGE"
],
[
71,
36,
0,
3,
0,
"MODEL"
],
[
75,
17,
0,
38,
0,
"IMAGE"
],
[
76,
38,
0,
18,
0,
"IMAGE"
]
],
"groups": [],
"config": {},
"extra": {
"ds": {
"scale": 1.5863092971714992,
"offset": [
170.04223120944968,
209.5374167314878
]
},
"node_versions": {
"comfy-core": "0.3.24"
}
},
"version": 0.4
}
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment