ONE-Lab/

MultiRef-benchmark

🏠 Project Page | 💻 GitHub | 📖 Paper | 🤗 MultiRef-Dataset MultiRef-Benchmark is a rigorous evaluation framework for controllable image generation models that use multiple visual references.

MultiRef-Benchmark is designed to test a model's ability to integrate diverse visual elements from multiple sources, a common task for digital artists but a significant challenge for current generative models. The benchmark comprises 1,990… See the full description on the dataset page: https://huggingface.co/datasets/ONE-Lab/MultiRef-benchmark.