Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[CVPR 2025] Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models
| Date | Stars |
|---|---|
| 2026-07-31 | 364 |
| 2026-08-02 | 365 |
| 2026-08-06 | 365 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR 2025) <div align="center"> <a href="https://arxiv.org/abs/2503.18352"><img src="https://img.shields.io/static/v1?label=Arxiv&message=Diffusion-4K&color=red&logo=arxiv"></a>   <a href="https://huggingface.co/datasets/zhang0jhon/Aesthetic-4K"><img src="https://img.shields.io/static/v1?label=Dataset&message=Aesthetic-4K&color=yellow"></a>   </div> <p align="center" border-raduis="10px"> <img src="asset/demo.jpg" width="100%" alt="teaser_page2"/> </p> ## Updates [2025/06/03] We release the updated checkpoint [flux_wavelet_v2_sc](https://huggingface.co/zhang0jhon/flux_wavelet_v2_sc) using SC-VAE and Aesthetic-Train-V2. Please refer to our new arxiv paper [Ultra-High-Resolution Image Synthesis: Data, Method and Evaluation](https://arxiv.org/abs/2506.01331) for more details. The training code of SC-VAE is released at [sc-vae](https://github.com/zhang0jhon/diffusion-4k/tree/main/sc-vae). The Aesthetic-Train-V2 training set is released at [huggingface/Aesthetic-Train-V2](https://huggingface.co/datasets/zhang0jhon/Aesthetic-Train-V2). ## Introduction We present Diffusion-4K, a novel framework for direct ultra-high-resolution image synthesis using text-to-image diffusion models. The core advancements include: (1) **Aesthetic-4K Benchmark**: addressing the absence of a publicly available 4K image synthesis dataset, we construct Aesthetic-4K, a comprehensive benchmark for ultra-high-resolution image generation. We curated a high-quality 4K dataset with carefully selected images and captions generated by GPT-4o. Additionally, we introduce GLCM Score and Compression Ratio metrics to evaluate fine details, combined with holistic measures such as FID, Aesthetics and CLIPScore for a comprehensive assessment of ultra-high-resolution images. (2) **Wavelet-based Fine-tuning**: we propose a wavelet-based fine-tuning approach for direct training with photorealistic 4K images, applicable to various latent diffusion models, demonstrating its effectiveness in synthesizing highly detailed 4K images. Consequently, Diffusion-4K achieves impressive performance in high-quality image synthesis and text prompt adherence, especially when powered by modern large-scale diffusion models (e.g., SD3-2B and Flux-12B). ## Aesthetic-4K Dataset * Aesthetic-4K dataset is released at [huggingface/Aesthetic-4K](https://huggingface.co/datasets/zhang0jhon/Aesthetic-4K). <p align="center" border-raduis="10px"> <img src="asset/dataset.jpg" width="100%" alt="teaser_page2"/> </p> ## Installation Install the required packages: ``` pip install -r requirements.txt ``` Download the [dataset](https://huggingface.co/datasets/zhang0jhon/Aesthetic-4K) in *Aesthetic-4K* folder, and the pretrained latent diffusion models([SD3-2B](https://huggingface.co/stabilityai/stable-diffusion-3-medium-diffusers) and [Flux-12B](https://huggingface.co/black-forest-labs/FLUX.1-dev)) in *pretrain* folder. ## Training Modify the configuration in *ds_config.yaml* accordingly (*e.g.* num_processes), and then run the script for training Flux-12B: ``` sh train_flux.sh ``` <!-- Run the script for training SD3-2B: ``` sh train_sd3.sh ``` --> ## Checkpoint Flux-12B-WLF checkpoint is released at [flux_wavelet](https://huggingface.co/zhang0jhon/flux_wavelet). ## Image Generation Generate images with Flux-12B: ``` CUDA_VISIBLE_DEVICES=0 python test_flux.py --guidance_scale 5.0 --height 3072 --width 4096 --seed 0 --num_inference_steps 50 --checkpoint_path $CHECKPOINT_PATH --prompt $PROMPT ``` <!-- Generate images with SD3-2B: ``` CUDA_VISIBLE_DEVICES=0 python test_sd3.py --guidance_scale 7.0 --height 3072 --width 4096 --seed 0 --num_inference_steps 28 --checkpoint_path $CHECKPOINT_PATH --prompt $PROMPT ``` --> <!-- You can modify the paramters of height, width, guiduance scale and random seed for various images generation. ``` for height, width in [(4096, 4096), (4096, 3072),
Excerpt of 5,857 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5d8e409d29d22b4e, desc:latent diffusion