Step1X Edit
An open-source AI image editing model runs on local CUDA GPUs, follows written editing requests and supports Diffusers and LoRA fine-tuning.

Step1X Edit edits existing images from written requests. It's for developers and researchers who want to run image editing on their own hardware, with control over the model and specialized training.
- Object and scene edits: Add, remove or replace subjects, swap backgrounds, and change colors or materials within an image.
- Portrait changes: Adjust expressions and gestures, retouch faces, or make targeted changes such as opening a subject's eyes.
- Text and style editing: Replace or remove words in pictures, or apply styles such as pixel art, clay and oil painting.
- Specialized fine-tuning: Train and use LoRA weights for particular editing tasks, with supplied weights for repairing malformed hands in anime images.
The model works with Hugging Face Diffusers. Local inference uses CUDA GPUs; published performance measurements use NVIDIA H800 hardware. The project includes reduced-memory inference, optional TeaCache acceleration and multi-GPU execution through xDiT. These paths still require GPU hardware.
Step1X Edit is open source under Apache 2.0. Its accompanying GEdit-Bench covers practical image editing requests, with English and Chinese evaluations for researchers comparing model results.