Research map / Vision-language and multimodal / Multimodal LLMs
Lumina-mGPT: Flexible Photorealistic Autoregressive Text-to-Image Generation
Dongyang Liu, Yi Xin, Shitian Zhao, Le Zhuo, Weifeng Lin, Xinyue Li, Qi Qin, Guangtao Zhai, Xiaohong Liu, Hongsheng Li, Yu Qiao, Peng Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| apply FP-SFT and Omni-SFT finetuning on Chameleon (mGPT init) | Chameleon · GenEval | Overall: 0.39 -> 0.56 | Improved |
| finetune Chameleon into Lumina-mGPT | Chameleon · MMBench | score: 19.80 -> 32.20 | Improved |
| finetune Chameleon into Lumina-mGPT | Chameleon · MME-p | score: 153.10 -> 976.85 | Improved |
| finetune Chameleon into Lumina-mGPT | Chameleon · POPE | score: 19.40 -> 70.43 | Improved |
| finetune Chameleon into Lumina-mGPT | Chameleon · SEEDBench-I | score: 30.50 -> 50.93 | Improved |
| finetune Chameleon into Lumina-mGPT | Chameleon · MMMU-val | score: 22.40 -> 27.11 | Improved |
| use AR Lumina-mGPT instead of diffusion Lumina-Next on same data | Lumina-Next · GenEval | Overall: 0.46 -> 0.56 | Improved |
| add progressive resolution stages (512, 768, 1024) in FP-SFT (ablation) | Lumina-mGPT · Qualitative image quality | visual artifacts and detail | Improved |
| add z-loss with weight 1e-5 during SFT (ablation) | Lumina-mGPT · SFT training stability | loss divergence | Improved |
| raise top-k for image decoding to 2000 instead of 5 (ablation) | Lumina-mGPT · Image generation | visual quality | Improved |
| increase sampling temperature from 0.7 to 1.0 (ablation) | Lumina-mGPT · Image generation | detail vs artifacts | Mixed |
| increase CFG scale (ablation) | Lumina-mGPT · Image generation | quality and stability | Mixed |
Closest papers by meaning
- JetFormer: An autoregressive generative model of raw images and text (ICLR 2025 · similarity 0.88)
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.88)
- MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis (AAAI 2025 · similarity 0.87)
- OmniGen-AR: AutoRegressive Any-to-Image Generation (NeurIPS 2025 · similarity 0.87)
- STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation (NeurIPS 2026 · similarity 0.87)
- Harmonizing Visual Text Comprehension and Generation (NeurIPS 2024 · 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.
