HKU-MMLab

HKU-MMLab / Macro

Public

The official repo of "MACRO: Advancing Multi-Reference Image Generation with Structured Long-Context Data"

macro400k.github.io ai benchmark cv data
19
0
100% credibility
Found Mar 27, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

MACRO provides a dataset, benchmarks, and fine-tuned models for generating images from multiple reference pictures across tasks like customization, illustration, spatial, and temporal.

How It Works

1
🌐 Discover MACRO

You stumble upon this exciting tool for blending multiple reference pictures into stunning new images.

2
📥 Grab ready examples

Download sample reference images and pre-trained AI models to get started right away.

3
✨ Test your first blend

Use the built-in sample to generate an image from several references and see the magic happen.

4
🖼️ Add your own pictures

Upload your favorite reference images and type a simple description of what you want to create.

5
🎨 Generate your image

Hit generate and watch as the AI crafts a beautiful new picture inspired by all your references.

🏆 Perfect multi-ref art

Enjoy your custom image that perfectly combines multiple inspirations, ready to share or use!

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Macro?

Macro delivers a Python-based dataset and benchmark for multi-reference image generation, tackling the challenge of creating visuals from 1-10+ reference images plus text prompts across customization, illustration, spatial, and temporal tasks. Download the structured data from Hugging Face, grab fine-tuned checkpoints for models like Bagel, OmniGen2, and Qwen-Image-Edit, then run quick inference scripts on samples or batch-eval your outputs with LLM scoring via OpenAI or Gemini APIs. It's built for advancing long-context image synthesis, complete with training pipelines that mix in text-to-image data.

Why is it gaining traction?

Unlike scattered multi-image datasets, Macro organizes long-context refs into clear brackets (1-3, 4-5, 6-7, 8+ images) with dynamic resolution handling, making it dead simple to benchmark models fairly. Devs dig the drop-in scripts for single-image tests or full benchmark runs, plus auto-generated training configs—no more wrestling raw data. Forget gaming hacks like github macro fortnite or macro recorder tools; this powers real structured gen.

Who should use this?

Multimodal AI researchers benchmarking reference-conditioned models, image editing devs handling style transfer or scene composition from photo collections, or teams fine-tuning open diffusion models on custom multi-ref setups. Ideal for academics citing the arXiv paper or prototyping apps beyond single-prompt T2I.

Verdict

Grab it for research if you're into multi-ref gen—docs are solid, HF integration seamless, but with 19 stars and 1.0% credibility score, it's raw academia, not battle-tested prod. Pair with official github actions for CI to iterate fast.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.