MagicQuill
Interactive image editing system driven by brush strokes with AI intent prediction.
About
Brush strokes drive everything in MagicQuill: an add brush sketches new objects, a subtract brush removes them, and a color brush repaints regions, while a multimodal model based on LLaVA watches the strokes and predicts editing intent, proposing the text prompt automatically so editing feels like drawing rather than prompt engineering. The Ant Research system, accepted to CVPR 2025, executes edits through a BrushNet dual-branch inpainting pipeline on Stable Diffusion 1.5 checkpoints such as Realistic Vision and DreamShaper, orchestrated over ComfyUI internals. Running it locally means a conda, scripted, or Docker setup, roughly 25 GB of checkpoint downloads, and a GPU with at least 8 GB of VRAM, confirmed by the authors on an RTX 4070 laptop card; hosted demos on magicquill.art, Hugging Face Spaces, and ModelScope cover lighter hardware. An official ComfyUI node slots it into larger workflows, and MagicQuillV2 introduced layered visual cue editing in late 2025. The repository is licensed CC BY-NC 4.0, so commercial use requires separate permission from the authors.
Should you use MagicQuill?
Pick it when
You want to edit images by sketching: add, remove, or recolor elements with brushes while a built-in multimodal model guesses the prompt, for research, demos, or personal use on a GPU with 8GB or more.
Look elsewhere when
You need commercial use, which CC BY-NC 4.0 forbids without author permission, or lack 8GB VRAM and room for about 25GB of checkpoints. Its SD 1.5 checkpoints are an older base than the 12B and 20B instruction editors.
Alternatives to MagicQuill
- Krita AI Diffusion
GPL-3.0 and usable commercially, with brush and selection inpainting on a real canvas, but you write the prompts yourself.
- Qwen-Image-Edit
Apache-2.0 and a much larger 20B model for precise edits, but you describe changes in text instead of drawing them.
- IOPaint (Lama Cleaner)
Apache-2.0, runs on CPU, and handles quick removal with LaMa, but has no stroke-based intent prediction.
- Inpaint Anything
Click an object to remove or replace it under Apache-2.0, simpler than sketching but with less control over shape and color.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- AI Image/Video Editing
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Intermediate (3/5)
- License
- CC-BY-NC-4.0
- Minimum VRAM
- 8 GB
- Added
- Jul 29, 2026