Step1X Edit

Step1X Edit is an open-source image editing model for developers and researchers who want to edit existing pictures with written instructions on their own GPU hardware. It handles changes to individual objects as well as the overall appearance of an image. The code uses the Apache 2.0 license.

Its editing range includes adding or removing subjects, replacing objects, changing backgrounds and adjusting colors or materials. It can also alter facial expressions and poses, retouch portraits, and change or remove text within an image. Style edits include pixel art, clay-like surfaces and oil painting effects, so the same model covers both targeted corrections and broader visual changes.

The model combines a multimodal language model, which processes the reference image and editing request, with a diffusion decoder that produces the edited picture. It works with Hugging Face Diffusers. For people comparing image editing models, the accompanying GEdit-Bench evaluates tasks drawn from real user requests and includes English and Chinese evaluations.

Local inference uses a CUDA-based GPU setup, and the published performance tests use NVIDIA H800 GPUs. The project provides ways to reduce GPU memory use, distribute inference across multiple GPUs with xDiT, and accelerate it with TeaCache. Developers can fine-tune the model with LoRA and use those weights for specialized edits; an included LoRA targets damaged hands in anime images.