| | --- |
| | license: apache-2.0 |
| | language: |
| | - en |
| | base_model: |
| | - Qwen/Qwen-Image-Edit-2511 |
| | - Qwen/Qwen-Image-Edit-2509 |
| | library_name: diffusers |
| | pipeline_tag: image-to-image |
| | tags: |
| | - qwen-image-edit |
| | - fp8 |
| | - vllm |
| | - compressed-tensors |
| | - art |
| | --- |
| | |
| | # **Qwen-Image-Edit-AIO-FP8** |
| |
|
| | > Qwen-Image-Edit-AIO-FP8 is an FP8-compressed edition of the Qwen-Image-Edit series developed by the Qwen team at Alibaba, designed to deliver high-precision multimodal diffusion-based image editing with significantly reduced memory consumption and accelerated inference performance. Built upon the advanced capabilities introduced in Qwen-Image-Edit-2509, including native multi-image conditioning that enables seamless fusion of multiple references such as identity, product, and environment within a single coherent generation pipeline, and further enhanced by the large-scale MMDiT-based architecture of Qwen-Image-Edit-2511 with its 20B-parameter backbone for superior identity consistency and minimal drift during iterative refinement, this FP8 release preserves structural control, text-in-image editing accuracy, and ControlNet compatibility while dramatically improving deployment efficiency on modern hardware. Optimized for industrial design workflows, high-fidelity multi-person composition, material replacement, geometric reasoning, and annotation-aware generation, Qwen-Image-Edit-AIO-FP8 maintains professional-grade lighting and viewpoint stability, supports integrated community LoRA adaptations, and enables scalable production-ready image editing pipelines with lower VRAM requirements and minimal quality degradation compared to full-precision checkpoints. |
| |
|
| | --- |
| |
|
| | ## Download FP8 |
| |
|
| | - `Qwen-Image-Edit-2509 @ FP8` - [Download](https://huggingface.co/prithivMLmods/Qwen-Image-Edit-AIO-FP8/resolve/main/Qwen-Image-Edit-2509-FP8-E5M2.safetensors?download=true) |
| | - `Qwen-Image-Edit-2511 @ FP8` - [Download](https://huggingface.co/prithivMLmods/Qwen-Image-Edit-AIO-FP8/resolve/main/Qwen-Image-Edit-2511-FP8-E4M3FN.safetensors?download=true) |
| |
|
| | ## Quick Start with Diffusers 🧨 |
| |
|
| | ### Install the required packages |
| |
|
| | ```py |
| | transformers # - transformers@v4.57.6 |
| | torch # - torch@v2.9.1+cu128 |
| | diffusers # - diffusers@v0.37.0.dev0 |
| | accelerate # - accelerate@v1.12.0 |
| | ``` |
| |
|
| | ## Qwen-Image-Edit-AIO-FP8 [Demo] |
| |
|
| | ```py |
| | import torch |
| | from diffusers import QwenImageEditPipeline # or your compatible pipeline |
| | |
| | model_path = "./Qwen-Image-Edit-AIO-FP8" |
| | |
| | pipe = QwenImageEditPipeline.from_pretrained( |
| | model_path, |
| | torch_dtype=torch.bfloat16 |
| | ) |
| | |
| | pipe.to("cuda") |
| | |
| | # Example usage |
| | # outputs = pipe(image=input_image, prompt="Your edit prompt") |
| | ``` |
| |
|
| | > [!IMPORTANT] |
| | This repository follows the same release notes, terms and conditions, and license as the original model page, [Qwen-Image-Edit](https://huggingface.co/Qwen/Qwen-Image-Edit-2509). |
| |
|
| | > [!IMPORTANT] |
| | This repository also contains experimental LLM compressed (compressed-tensors) editions of the Qwen image edit models. Feel free to ignore those editions if you are struggling with them or unable to mount them. |
