Research map / Image generation / Editing and style transfer
MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · MagicBrush single-turn mask-free | CLIP-I: 0.8524 -> 0.9332 | Helped |
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · MagicBrush multi-turn mask-free | L1: 0.1584 -> 0.0964 | Helped |
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · MagicBrush single-turn mask-free | CLIP-T: 0.2764 -> 0.2781 | No clear effect |
| fine-tune HIVE on MagicBrush | HIVE · MagicBrush single-turn mask-free | L1: 0.1092 -> 0.0658 | Helped |
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · Human eval multi-choice consistency | times chosen best: 33 -> 51 | Helped |
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · Human eval individual image quality | Likert 1-5: 3.2 -> 3.6 | Helped |
| fine-tune InstructPix2Pix on MagicBrush | InstructPix2Pix · Human eval individual consistency | Likert 1-5: 3.0 -> 3.1 | No clear effect |
| multi-turn instead of single-turn editing | InstructPix2Pix w/ MagicBrush · MagicBrush mask-free | L1: 0.0625 -> 0.0964 | Hurt |
| more edit turns, fine-tuned model vs ground truth | Fine-tuned InstructPix2Pix · Human one-on-one image quality | preferred over GT (of 100): 19 -> 6 | Hurt |
Closest papers by meaning
- UltraEdit: Instruction-based Fine-Grained Image Editing at Scale (NeurIPS 2024 · similarity 0.90)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.89)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.89)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025 · similarity 0.88)
- Guiding Instruction-based Image Editing via Multimodal Large Language Models (ICLR 2024 · similarity 0.87)
- OmniEdit: Building Image Editing Generalist Models Through Specialist Supervision (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.