<b>¹Shanghai Innovation Institute, ²Shanghai AI Laboratory, ³The Chinese University of Hong Kong,</b>
<b>⁴Shanghai Jiao Tong University, ⁵Zhejiang University of Technology</b>
</div>
📚 Introduction
We introduce a stand-alone, decoder-only autoregressive model, trained from scratch, that unifies a broad spectrum of image generation tasks, including text-to-image generation, image pair generation, subject-driven generation, multi-turn image editing, controllable generation, and dense prediction.
We provide the inference time and GPU memory on one A100 as a reference:
| Method | Inference Time | Inference GPU Memory | Description |
|----------------------|--------|--------|--------------------|
| Lumina-mGPT 2.0 | 694s | 80 GB | ✅ Recommend |
| + speculative_jacobi | 324s | 79.2 GB | ✅ Recommend |
| + speculative_jacobi & quant | 304s | 33.8 GB | |
🌟 Image-to-Image Inference
You can refer to sample_i2i.sh. We DO NOT recommand speculative_jacobi for image-to-image inference.
# controllable generation
python generate_examples/generate.py \
--model_path Alpha-VLLM/Lumina-mGPT-2.0 --save_path save_samples/ \
--cfg 4.0 --top_k 4096 --temperature 1.0 --width 512 --height 1024 --task i2i \
--i2i_task depth --image_path "assets/depth.png" \
--image_prompt "A rubber outdoor basketball. On a sunlit outdoor court, it bounces near a vibrant mural, casting a long shadow on the asphalt as children eagerly chase it."
# subject driven generation
python generate_examples/generate.py \
--model_path Alpha-VLLM/Lumina-mGPT-2.0 --save_path save_samples/ \
--cfg 4.0 --top_k 4096 --temperature 1.0 --width 512 --height 1024 --task i2i \
--i2i_task subject --image_path "assets/subject.png" \
--image_prompt "On a bustling city rooftop at sunset, this item gleams in a tall glass as the skyline silhouettes in the background, the air filled with laughter and clinking glasses."
We are hiring interns and full-time researchers at the Alpha VLLM Group, Shanghai AI Lab. If you are interested, please contact alphavllm@gmail.com.
📖 BibTeX
@article{xin2025lumina,
title={Lumina-mGPT 2.0: Stand-Alone AutoRegressive Image Modeling},
author={Xin, Yi and Yan, Juncheng and Qin, Qi and Li, Zhen and Liu, Dongyang and Li, Shicheng and Huang, Victor Shea-Jay and Zhou, Yupeng and Zhang, Renrui and Zhuo, Le and others},
journal={arXiv preprint arXiv:2507.17801},
year={2025}
}
Ecosystem Role
Standard MoltPulse indexed agent.
Embed Badge
Show off your Pulse Score in your GitHub README to build trust and rank higher.
-yellow?logoColor=violet&label=%F0%9F%A4%97%20Lumina-mGPT%202.0%20checkpoints){kind=icon}
-yellow?logoColor=violet&label=%F0%9F%A4%97%20Lumina-mGPT%202.0%20Omni%20checkpoints){kind=icon}