MiniMaxAI open-sourced MiniMax-H3-Image-PE, a prompt expansion model for MiniMax-H3-Image
MiniMaxAI released MiniMax-H3-Image-PE on Hugging Face, a prompt expansion model that turns short requests into detailed image descriptions and aspect-ratio recommendations for MiniMax-H3-Image. The model is fine-tuned from Qwen3.8-27B in FP4 and released under the MiniMax-H3-Image Community License Agreement.
China context
- Original name
- 稀宇科技
- Outside China
- Open weights · huggingface.co
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can access the MiniMax-H3-Image-PE weights on Hugging Face and integrate prompt expansion into their image generation pipelines.
- For investors
- The release of prompt expansion weights under a community license may indicate MiniMax's strategy to encourage adoption of its image generation models among developers.
MiniMaxAI released MiniMax-H3-Image-PE on Hugging Face. The model provides prompt expansion for MiniMax-H3-Image, converting short text instructions into detailed image descriptions and aspect-ratio recommendations. It supports text-to-image and image editing workflows. The text-to-image prompt expansion weights are released, fine-tuned from Qwen3.8-27B in FP4, along with task prompts, system prompts, and a client example. Image editing prompt expansion is still under development and only the prompt is provided. The model is part of MiniMax-H3-Image and released under the MiniMax-H3-Image Community License Agreement.
MiniMax-H3-Image-PE is fine-tuned from Qwen3.8-27B in FP4 precision. The repository recommends serving with SGLang version 0.5.23 or higher and xgrammar 0.2.7 or higher. XGrammar constrains output to a JSON schema, and a reasoning parser separates reasoning from the final answer. Default request parameters include temperature 1.0, top_p 0.95, top_k 20, max_tokens 4096, and reasoning effort xhigh. The model outputs an expanded caption and aspect ratio in JSON format.
Developers outside China can now use MiniMax-H3-Image-PE to improve image generation prompts for MiniMax-H3-Image, potentially reducing the need for manual prompt engineering. The release of prompt expansion weights under a community license may lower the barrier to integrating MiniMax's image generation capabilities into third-party applications.
MiniMax-H3-Image-PE can help developers generate more detailed and effective prompts for MiniMax-H3-Image, potentially improving image quality and reducing iteration time. The open release of weights under a community license may encourage experimentation and integration.
The image editing prompt expansion is still under development, with only the prompt released. A specific signal to watch is whether MiniMaxAI releases the image editing PE weights and whether the model gains adoption in image generation workflows outside China.