Metraipolis › archive › dc_25033f296e7d
LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes — arXiv:2609.03796
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.03796
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T01:12:06.110Z
- SHA-256 of the archived text
edf4674b0c012f3b9f57abe14f77c343f7a85bc3eca600090bdeba6d253c5521- Stable URL of this page
https://metraipolis.com/documento/dc_25033f296e7d
How to cite
LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes — arXiv:2609.03796 — arXiv (en), https://arxiv.org/abs/2609.03796 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_25033f296e7d (Metraipolis archive dc_25033f296e7d, sha256 edf4674b0c012f3b9f57abe14f77c343f7a85bc3eca600090bdeba6d253c5521)
BibTeX
@misc{dc25033f296e7d,
title = {LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes — arXiv:2609.03796},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.03796},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_25033f296e7d; sha256 edf4674b0c012f3b9f57abe14f77c343f7a85bc3eca600090bdeba6d253c5521},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.03796. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We introduce LLaDA-Image, a unified framework that pairs a 6B Diffusion Transformer (DiT) trained from scratch with a frozen vision-language understanding module built on the LLaDA2.0-Mini diffusion language model backbone. Instead of relying heavily on paired image-text data from the beginning, we first build a strong visual generative prior through image-only pre-training and mid-training. The generation pipeline comprises 220M samples, 98 of which are real images. For efficient and scalable optimization, we use parameter-free RMSNorm throughout the DiT together with the Muon optimizer. The resulting unified model produces highly photorealistic images while accurately following fine-grained editing instructions. We further distill LLaDA-Image into LLaDA-Image-Turbo, enabling fast inference in 2-4 sampling steps. On Qwen-Image-Bench, LLaDA-Image achieves overall scores of 53.53 and 53.38 on the English and Chinese tracks, respectively, setting a new state-of-the-art among open-source models on both tracks. To support further research on capable and efficient generative models, we release our model weights, training code, and detailed recipes. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-04 in cs.AI). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.03796 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-17T00:28:03.002Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-04?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).