[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72243":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":36,"readmeContent":37,"aiSummary":38,"trendingCount":16,"starSnapshotCount":16,"syncStatus":39,"lastSyncTime":40,"discoverSource":41},72243,"PromptEnhancer","Hunyuan-PromptEnhancer\u002FPromptEnhancer","Hunyuan-PromptEnhancer","[CVPR 2026] PromptEnhancer is a prompt-rewriting tool, refining prompts into clearer, structured versions for better image generation.","https:\u002F\u002Fhunyuan-promptenhancer.github.io\u002F",null,"Python",3700,322,217,14,0,3,5,24,9,70.43,"Other",false,"main",true,[27,28,29,30,31,32,33,34,35],"hunyuan","hunyuan-image","image-editing","image-to-image","prompt","prompt-engineering","prompt-enhancer","text-to-image","vlm","2026-06-12 04:01:04","  \u003Cdiv align=\"center\">\n\n# PromptEnhancer: A Simple Approach to Enhance Text-to-Image Models via Chain-of-Thought Prompt Rewriting\n\n[**Linqing Wang**](https:\u002F\u002Fscholar.google.com\u002Fcitations?hl=en&view_op=list_works&gmla=AH8HC4z9rmDHYjp5o28xKk8U4ddD_n7BuMnk8UZFP-jygFBtHUSz6pf-5FP32B_yKMpRU9VpDY3iT8eM0zORHA&user=Hy12lcEAAAAJ) · \n[**Ximing Xing**](https:\u002F\u002Fximinng.github.io\u002F) ·\nZhiyong Xu · \n[**Yiji Cheng**](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=Plo8ZSYAAAAJ&hl=en) · \nZhiyuan Zhao · \nDonghao Li ·\nTiankai Hang ·\nZhenxi Li ·\n[**Jiale Tao**](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=WF5DPWkAAAAJ&hl=en) · \n[**QiXun Wang**](https:\u002F\u002Fgithub.com\u002Fwangqixun) · \n[**Ruihuang Li**](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=8CfyOtQAAAAJ&hl=en) · \nComi Chen ·\nXin Li · \n[**Mingrui Wu**](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=sbCKwnYAAAAJ&hl=en) · \nXinchi Deng · \nShuyang Gu · \n[**Chunyu Wang**](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=VXQV5xwAAAAJ&hl=en)\u003Csup>†\u003C\u002Fsup> · \n[**Qinglin Lu**](https:\u002F\u002Fluqinglin.weebly.com\u002F)\u003Csup>*\u003C\u002Fsup>\n\nTencent Hunyuan\n\n\u003Csup>†\u003C\u002Fsup>Project Lead · \u003Csup>*\u003C\u002Fsup>Corresponding Author\n\n\u003C\u002Fdiv>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2509.04545\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-arXiv:2509.04545-red?logo=arxiv\" alt=\"arXiv\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1949013083109459515\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F知乎-技术解读-0084ff?logo=zhihu\" alt=\"Zhihu\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHunyuanImage-2.1\u002Ftree\u002Fmain\u002Freprompt\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-PromptEnhancer_7B-blue?logo=huggingface\" alt=\"HuggingFace Model\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FPromptEnhancer\u002FPromptEnhancer-Img2img-Edit\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-PromptEnhancer_Img2Img_Edit-blue?logo=huggingface\" alt=\"HuggingFace Model\">\u003C\u002Fa>\n  \u003C!-- \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FPromptEnhancer\u002FPromptEnhancer-32B\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-PromptEnhancer_32B-blue?logo=huggingface\" alt=\"HuggingFace Model\">\u003C\u002Fa> -->\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPromptEnhancer\u002FT2I-Keypoints-Eval\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBenchmark-T2I_Keypoints_Eval-blue?logo=huggingface\" alt=\"T2I-Keypoints-Eval Dataset\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhunyuan-promptenhancer.github.io\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHomepage-PromptEnhancer-1abc9c?logo=homeassistant&logoColor=white\" alt=\"Homepage\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanImage-2.1\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FCode-HunyuanImage2.1-2ecc71?logo=github\" alt=\"HunyuanImage2.1 Code\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHunyuanImage-2.1\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-HunyuanImage2.1-3498db?logo=huggingface\" alt=\"HunyuanImage2.1 Model\">\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002Fx.com\u002FTencentHunyuan target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHunyuan-black.svg?logo=x height=22px>\u003C\u002Fa>\n\u003C\u002Fp>\n\n---\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fteaser-1.png\" alt=\"PromptEnhancer Teaser\"\u002F>\n\u003C\u002Fp>\n\n## Overview\n\nHunyuan-PromptEnhancer is a prompt rewriting utility that **supports both Text-to-Image generation and Image-to-Image editing**. It restructures input prompts while preserving original intent, producing clearer, structured prompts for downstream image generation tasks.\n\n**Key Features:**\n- **Dual-mode support**: Text-to-Image prompt enhancement and Image-to-Image editing instruction refinement with visual context\n- **Intent preservation**: Maintains all key elements (subject, action, style, layout, attributes, etc.) across rewriting\n- **Robust parsing**: Multi-level fallback mechanism ensures reliable output\n- **Flexible deployment**: Supports full-precision (7B\u002F32B), quantized (GGUF), and vision-language models\n\n## 🔥🔥🔥Updates\n\n- [2025-10-11] ✨ Release [PromptEnhancer-32B gradio](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FPromptEnhancer\u002FPromptEnhancer_32B).\n- [2025-09-30] ✨ Release [PromptEnhancer-Img2Img Editing model](https:\u002F\u002Fhuggingface.co\u002FPromptEnhancer\u002FPromptEnhancer-Img2img-Edit).\n- [2025-09-22] 🚀 Thanks @mradermacher for adding **GGUF model support** for efficient inference with quantized models!\n- [2025-09-18] ✨ Try the [PromptEnhancer-32B](https:\u002F\u002Fhuggingface.co\u002FPromptEnhancer\u002FPromptEnhancer-32B) for higher-quality prompt enhancement!\n- [2025-09-16] ✨ Release [T2I-Keypoints-Eval dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPromptEnhancer\u002FT2I-Keypoints-Eval).\n- [2025-09-07] ✨ Release [PromptEnhancer-7B model](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHunyuanImage-2.1\u002Ftree\u002Fmain\u002Freprompt).\n- [2025-09-07] ✨ Release [technical report](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.04545).\n\n## Installation\n\n### Option 1: Standard Installation (Recommended)\n```bash\npip install -r requirements.txt\n```\n\n### Option 2: GGUF Installation (For quantized models with CUDA support)\n```bash\nchmod +x script\u002Finstall_gguf.sh && .\u002Fscript\u002Finstall_gguf.sh\n```\n\n> **💡 Tip**: Choose GGUF installation if you want faster inference with lower memory usage, especially for the 32B model.\n\n## Model Download\n\n### 🎯 Quick Start\n\nFor most users, we recommend starting with the **PromptEnhancer-7B** model:\n\n```bash\n# Download PromptEnhancer-7B (13GB) - Best balance of quality and efficiency\nhuggingface-cli download tencent\u002FHunyuanImage-2.1\u002Freprompt --local-dir .\u002Fmodels\u002Fpromptenhancer-7b\n```\n\n### 📊 Model Comparison & Selection Guide\n\n| Model | Size | Quality | Memory | Best For |\n|-------|------|---------|--------|----------|\n| **PromptEnhancer-7B** | 13GB | High | 8GB+ | Most users, balanced performance |\n| **PromptEnhancer-32B** | 64GB | Highest | 32GB+ | Research, highest quality needs |\n| **32B-Q8_0 (GGUF)** | 35GB | Highest | 35GB+ | High-end GPUs (H100, A100) |\n| **32B-Q6_K (GGUF)** | 27GB | Excellent | 27GB+ | RTX 4090, RTX 5090 |\n| **32B-Q4_K_M (GGUF)** | 20GB | Good | 20GB+ | RTX 3090, RTX 4080 |\n\n### Standard Models (Full Precision)\n```bash\n# PromptEnhancer-7B (recommended for most users)\nhuggingface-cli download tencent\u002FHunyuanImage-2.1\u002Freprompt --local-dir .\u002Fmodels\u002Fpromptenhancer-7b\n\n# PromptEnhancer-32B (for highest quality)\nhuggingface-cli download PromptEnhancer\u002FPromptEnhancer-32B --local-dir .\u002Fmodels\u002Fpromptenhancer-32b\n\n# PromptEnhancer-Img2Img-Edit (for image editing tasks)\nhuggingface-cli download PromptEnhancer\u002FPromptEnhancer-Img2img-Edit --local-dir .\u002Fmodels\u002Fpromptenhancer-img2img-edit\n```\n\n### GGUF Models (Quantized - Memory Efficient)\n\nChoose one based on your GPU memory:\n\n```bash\n# Q8_0: Highest quality (35GB)\nhuggingface-cli download mradermacher\u002FPromptEnhancer-32B-GGUF PromptEnhancer-32B.Q8_0.gguf --local-dir .\u002Fmodels\n\n# Q6_K: Excellent quality (27GB) - Recommended for RTX 4090\nhuggingface-cli download mradermacher\u002FPromptEnhancer-32B-GGUF PromptEnhancer-32B.Q6_K.gguf --local-dir .\u002Fmodels\n\n# Q4_K_M: Good quality (20GB) - Recommended for RTX 3090\u002F4080\nhuggingface-cli download mradermacher\u002FPromptEnhancer-32B-GGUF PromptEnhancer-32B.Q4_K_M.gguf --local-dir .\u002Fmodels\n```\n\n> **🚀 Performance Tip**: GGUF models offer 50-75% memory reduction with minimal quality loss. Use Q6_K for the best quality\u002Fmemory trade-off.\n\n## Quickstart\n\n### Using HunyuanPromptEnhancer (Text-to-Image)\n\n```python\nfrom inference.prompt_enhancer import HunyuanPromptEnhancer\n\nmodels_root_path = \".\u002Fmodels\u002Fpromptenhancer-7b\"\n\nenhancer = HunyuanPromptEnhancer(models_root_path=models_root_path, device_map=\"auto\")\n\n# Enhance a prompt (Chinese or English)\nuser_prompt = \"Third-person view, a race car speeding on a city track...\"\nnew_prompt = enhancer.predict(\n    prompt_cot=user_prompt,\n    # Default system prompt is tailored for image prompt rewriting; override if needed\n    temperature=0.7,   # >0 enables sampling; 0 uses deterministic generation\n    top_p=0.9,\n    max_new_tokens=256,\n)\n\nprint(\"Enhanced:\", new_prompt)\n```\n\n### Using PromptEnhancerImg2Img (Image Editing)\n\nFor image editing tasks where you want to enhance editing instructions based on input images:\n\n```python\nfrom inference.prompt_enhancer_img2img import PromptEnhancerImg2Img\n\n# Initialize the image-to-image prompt enhancer\nenhancer = PromptEnhancerImg2Img(\n    model_path=\".\u002Fmodels\u002Fyour-model\",\n    device_map=\"auto\"\n)\n\n# Enhance an editing instruction with image context\nedit_instruction = \"Remove the watermark from the bottom\"\nimage_path = \".\u002Fexamples\u002Fsample_image.png\"\n\nenhanced_prompt = enhancer.predict(\n    edit_instruction=edit_instruction,\n    image_path=image_path,\n    temperature=0.1,\n    top_p=0.9,\n    max_new_tokens=2048\n)\n\nprint(\"Enhanced editing prompt:\", enhanced_prompt)\n```\n\n### Using GGUF Models (Quantized, Faster)\n\n```python\nfrom inference.prompt_enhancer_gguf import PromptEnhancerGGUF\n\n# Auto-detects Q8_0 model in models\u002F folder\nenhancer = PromptEnhancerGGUF(\n    model_path=\".\u002Fmodels\u002FPromptEnhancer-32B.Q8_0.gguf\",  # Optional: auto-detected\n    n_ctx=1024,        # Context window size\n    n_gpu_layers=-1,   # Use all GPU layers\n)\n\n# Enhance a prompt\nuser_prompt = \"woman in jungle\"\nenhanced_prompt = enhancer.predict(\n    user_prompt,\n    temperature=0.3,\n    top_p=0.9,\n    max_new_tokens=512,\n)\n\nprint(\"Enhanced:\", enhanced_prompt)\n```\n\n### Command Line Usage (GGUF)\n\n```bash\n# Simple usage - auto-detects model in models\u002F folder\npython inference\u002Fprompt_enhancer_gguf.py\n\n# Or specify model path\nGGUF_MODEL_PATH=\".\u002Fmodels\u002FPromptEnhancer-32B.Q8_0.gguf\" python inference\u002Fprompt_enhancer_gguf.py\n```\n\n## GGUF Model Benefits\n\n🚀 **Why use GGUF models?**\n- **Memory Efficient**: 50-75% less VRAM usage compared to full precision models\n- **Faster Inference**: Optimized for CPU and GPU acceleration with llama.cpp\n- **Quality Preserved**: Q8_0 and Q6_K maintain excellent output quality\n- **Easy Deployment**: Single file format, no complex dependencies\n- **GPU Acceleration**: Full CUDA support for high-performance inference\n\n| Model | Size | Quality | VRAM Usage | Best For |\n|-------|------|---------|------------|----------|\n| Q8_0  | 35GB | Highest | ~35GB      | High-end GPUs (H100, A100) |\n| Q6_K  | 27GB | Excellent | ~27GB     | RTX 4090, RTX 5090 |\n| Q4_K_M| 20GB | Good    | ~20GB      | RTX 3090, RTX 4080 |\n\n## Usage Comparison\n\n| Model | Input Type | Use Case | Model Backend |\n|-------|------------|----------|---------------|\n| **HunyuanPromptEnhancer** | Text only | Text-to-Image generation | Transformers (7B\u002F32B) |\n| **PromptEnhancerImg2Img** | Text + Image | Image editing tasks | Transformers (32B) |\n| **PromptEnhancerGGUF** | Text only | Memory-efficient T2I | llama.cpp (quantized) |\n\n## Parameters\n\n### Standard Models (Transformers)\n- `models_root_path`: Local path or repo id; supports `trust_remote_code` models.\n- `device_map`: Device mapping (default `auto`).\n- `predict(...)`:\n  - `prompt_cot` (str): Input prompt to rewrite.\n  - `sys_prompt` (str): Optional system prompt; a default is provided for image prompt rewriting.\n  - `temperature` (float): `>0` enables sampling; `0` for deterministic generation.\n  - `top_p` (float): Nucleus sampling threshold (effective when sampling).\n  - `max_new_tokens` (int): Maximum number of new tokens to generate.\n\n### GGUF Models\n- `model_path` (str): Path to GGUF model file (auto-detected if in models\u002F folder).\n- `n_ctx` (int): Context window size (default: 8192, recommended: 1024 for short prompts).\n- `n_gpu_layers` (int): Number of layers to offload to GPU (-1 for all layers).\n- `verbose` (bool): Enable verbose logging from llama.cpp.\n\n### Image-to-Image Models (PromptEnhancerImg2Img)\n- `model_path` (str): Path to the pretrained Qwen2.5-VL model.\n- `device_map` (str): Device mapping for model loading (default: `auto`).\n- `predict(...)`:\n  - `edit_instruction` (str): Original editing instruction.\n  - `image_path` (str): Path to the input image file.\n  - `sys_prompt` (str): Optional system prompt (uses default if None).\n  - `temperature` (float): Sampling temperature (default: 0.1).\n  - `top_p` (float): Nucleus sampling threshold (default: 0.9).\n  - `max_new_tokens` (int): Maximum tokens to generate (default: 2048).\n\n## Citation\n\nIf you find this project useful, please consider citing:\n```bibtex\n@article{promptenhancer,\n  title={PromptEnhancer: A Simple Approach to Enhance Text-to-Image Models via Chain-of-Thought Prompt Rewriting},\n  author={Wang, Linqing and Xing, Ximing and Cheng, Yiji and Zhao, Zhiyuan and Donghao, Li and Tiankai, Hang and Zhenxi, Li and Tao, Jiale and Wang, QiXun and Li, Ruihuang and Chen, Comi and Li, Xin and Wu, Mingrui and Deng, Xinchi and Gu, Shuyang and Wang, Chunyu and Lu, Qinglin},\n  journal={arXiv preprint arXiv:2509.04545},\n  year={2025}\n}\n```\n\n## Acknowledgements\n\nWe would like to thank the following open-source projects and communities for their contributions to open research and exploration: [Transformers](https:\u002F\u002Fhuggingface.co\u002Ftransformers) and [HuggingFace](https:\u002F\u002Fhuggingface.co).\n\n## Contact\n\nIf you would like to leave a message for our R&D and product teams, Welcome to contact our open-source team. You can also contact us via email (hunyuan_opensource@tencent.com).\n\n## Github Star History\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=Hunyuan-PromptEnhancer\u002FPromptEnhancer&type=date&logscale&legend=top-left)](https:\u002F\u002Fwww.star-history.com\u002F#Hunyuan-PromptEnhancer\u002FPromptEnhancer&type=date&logscale&legend=top-left)\n","PromptEnhancer 是一个用于优化文本到图像生成的提示重写工具，通过改进提示语使其更清晰、结构化以提高图像生成质量。其核心技术特点在于采用链式思维重写机制，能够有效提升输入文本的表达力和准确性，进而增强基于文本的图像生成模型的表现。该项目适用于需要高质量图像输出的各种场景，如创意设计、数字艺术创作以及任何依赖于精确视觉内容生成的应用领域。开发语言为Python，并已在GitHub上获得广泛关注，是当前图像编辑与生成技术研究中的一个重要贡献。",2,"2026-06-11 03:40:59","high_star"]