# River-Zhang/ICEdit

[NeurIPS 2025] Image editing is worth a single LoRA! 0.1% training data for fantastic image editing! Surpasses GPT-4o in ID persistence~ MoE ckpt released! Only 4GB VRAM is enough to run!

Repository: https://github.com/River-Zhang/ICEdit
Canonical: https://ross.abutalabs.com/products/icedit
Homepage: https://river-zhang.github.io/ICEdit-gh-pages/
Language: Python
License: NOASSERTION
License Family: other
Topics: diffusion, diffusion-models, diffusion-transformer, editing-image, image-editing, dit, in-context, gpt4o, gpt4oimage
Last push: 2025-12-19T19:08:02+00:00

## Health v2 (maintenance only)
Score: 45/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 58, release rhythm 35, longevity 35
- inputs: {"age_days": 491, "days_push": 257, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2102, forks 111 (observed 2026-08-28T04:06:13.711019+00:00)

## What it is
ICEdit (In-Context Edit) is a research framework for instruction-based image editing built on large-scale Diffusion Transformers, using a LoRA-MoE hybrid tuning strategy with only ~1% trainable parameters and 0.1% of the training data of prior methods. It includes inference and training code, ComfyUI nodes, and a Gradio demo, and can run with as little as 4GB VRAM.

## Use cases
- edit images with natural language instructions
- change objects or attributes in a photo via text prompts
- run instruction-based image editing on a low-VRAM GPU
- train a custom image editing LoRA
- multi-turn precise photo editing
- open-source alternative to GPT-4o image editing

## When to choose
- you need instruction-following image editing with strong identity persistence
- you have limited GPU memory (as low as 4GB with quantization)
- you want to train your own editing LoRA cheaply
- you want a ComfyUI-integrated editing workflow

## When to avoid
- you need text-to-image generation from scratch rather than editing existing images
- you need a fully permissive license (license is custom/unspecified)
- you need commercial-grade support or guaranteed stability
- you work outside diffusion/DiT-based pipelines

## Facets
- artifact type: library
- maturity: active
- function: image-processing, machine-learning, deep-learning, llm-inference
- domain: image-processing, artificial-intelligence, deep-learning
- platform: python
- tags: diffusion-transformer, lora, instruction-based-image-editing, comfyui, moe, gradio-demo, research-code, linux, gpu, docker

## Member repositories
- River-Zhang/ICEdit (main) score 45

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:13.711019+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:54:15.821709+00:00, confidence not recorded.
  - readme: https://github.com/River-Zhang/ICEdit (fetched 2026-08-28T04:06:13.711019+00:00, sha 15d133ab3470)
  - homepage: https://river-zhang.github.io/ICEdit-gh-pages/ (fetched 2026-08-29T10:34:30.110207+00:00, sha cca8d14a5764)
- Data as of 2026-08-30T08:39:29.467469+00:00.