Runware/hidream-i1-dev
073
1---2license: mit3tags:4- image-generation5- HiDream.ai6language:7- en8pipeline_tag: text-to-image9library_name: diffusers10---11 1213 14`HiDream-I1` is a new open-source image generative foundation model with 17B parameters that achieves state-of-the-art image generation quality within seconds.15 16<span style="color: #FF5733; font-weight: bold">For more features and to experience the full capabilities of our product, please visit [https://vivago.ai/](https://vivago.ai/).</span>17 18## Key Features19- ✨ **Superior Image Quality** - Produces exceptional results across multiple styles including photorealistic, cartoon, artistic, and more. Achieves state-of-the-art HPS v2.1 score, which aligns with human preferences.20- 🎯 **Best-in-Class Prompt Following** - Achieves industry-leading scores on GenEval and DPG benchmarks, outperforming all other open-source models.21- 🔓 **Open Source** - Released under the MIT license to foster scientific advancement and enable creative innovation.22- 💼 **Commercial-Friendly** - Generated images can be freely used for personal projects, scientific research, and commercial applications.23 24## Quick Start25Please make sure you have installed [Flash Attention](https://github.com/Dao-AILab/flash-attention). We recommend CUDA version 12.4 for the manual installation.26```27pip install -r requirements.txt28```29Clone the GitHub repo:30```31git clone https://github.com/HiDream-ai/HiDream-I132```33 34Then you can run the inference scripts to generate images:35 36```python37# For full model inference38python ./inference.py --model_type full39 40# For distilled dev model inference41python ./inference.py --model_type dev42 43# For distilled fast model inference44python ./inference.py --model_type fast45```46> **Note:** The inference script will automatically download `meta-llama/Meta-Llama-3.1-8B-Instruct` model files. If you encounter network issues, you can download these files ahead of time and place them in the appropriate cache directory to avoid download failures during inference.47 48## Gradio Demo49 50We also provide a Gradio demo for interactive image generation. You can run the demo with:51 52```python53python gradio_demo.py 54```55 56## Evaluation Metrics57 58### DPG-Bench59| Model | Overall | Global | Entity | Attribute | Relation | Other |60|-----------------|-----------|-----------|-----------|-----------|-----------|-----------|61| PixArt-alpha | 71.11 | 74.97 | 79.32 | 78.60 | 82.57 | 76.96 |62| SDXL | 74.65 | 83.27 | 82.43 | 80.91 | 86.76 | 80.41 |63| DALL-E 3 | 83.50 | 90.97 | 89.61 | 88.39 | 90.58 | 89.83 |64| Flux.1-dev | 83.79 | 85.80 | 86.79 | 89.98 | 90.04 | 89.90 |65| SD3-Medium | 84.08 | 87.90 | 91.01 | 88.83 | 80.70 | 88.68 |66| Janus-Pro-7B | 84.19 | 86.90 | 88.90 | 89.40 | 89.32 | 89.48 |67| CogView4-6B | 85.13 | 83.85 | 90.35 | 91.17 | 91.14 | 87.29 |68| **HiDream-I1** | **85.89**| 76.44 | 90.22 | 89.48 | 93.74 | 91.83 | 69 70### GenEval71 72| Model | Overall | Single Obj. | Two Obj. | Counting | Colors | Position | Color attribution |73|-----------------|----------|-------------|----------|----------|----------|----------|-------------------|74| SDXL | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | 0.23 |75| PixArt-alpha | 0.48 | 0.98 | 0.50 | 0.44 | 0.80 | 0.08 | 0.07 |76| Flux.1-dev | 0.66 | 0.98 | 0.79 | 0.73 | 0.77 | 0.22 | 0.45 |77| DALL-E 3 | 0.67 | 0.96 | 0.87 | 0.47 | 0.83 | 0.43 | 0.45 |78| CogView4-6B | 0.73 | 0.99 | 0.86 | 0.66 | 0.79 | 0.48 | 0.58 |79| SD3-Medium | 0.74 | 0.99 | 0.94 | 0.72 | 0.89 | 0.33 | 0.60 |80| Janus-Pro-7B | 0.80 | 0.99 | 0.89 | 0.59 | 0.90 | 0.79 | 0.66 |81| **HiDream-I1** | **0.83**| 1.00 | 0.98 | 0.79 | 0.91 | 0.60 | 0.72 |82 83### HPSv2.1 benchmark84 85| Model | Averaged | Animation | Concept-art | Painting | Photo |86|-------------------------|----------------|------------|---------------|--------------|------------|87| Stable Diffusion v2.0 | 26.38 | 27.09 | 26.02 | 25.68 | 26.73 |88| Midjourney V6 | 30.29 | 32.02 | 30.29 | 29.74 | 29.10 |89| SDXL | 30.64 | 32.84 | 31.36 | 30.86 | 27.48 |90| Dall-E3 | 31.44 | 32.39 | 31.09 | 31.18 | 31.09 |91| SD3 | 31.53 | 32.60 | 31.82 | 32.06 | 29.62 |92| Midjourney V5 | 32.33 | 34.05 | 32.47 | 32.24 | 30.56 |93| CogView4-6B | 32.31 | 33.23 | 32.60 | 32.89 | 30.52 |94| Flux.1-dev | 32.47 | 33.87 | 32.27 | 32.62 | 31.11 |95| stable cascade | 32.95 | 34.58 | 33.13 | 33.29 | 30.78 |96| **HiDream-I1** | **33.82** | 35.05 | 33.74 | 33.88 | 32.61 |97 98 99## License Agreement100The Transformer models in this repository are licensed under the MIT License. The VAE is from `FLUX.1 [schnell]`, and the text encoders from `google/t5-v1_1-xxl` and `meta-llama/Meta-Llama-3.1-8B-Instruct`. Please follow the license terms specified for these components. You own all content you create with this model. You can use your generated content freely, but you must comply with this license agreement. You are responsible for how you use the models. Do not create illegal content, harmful material, personal information that could harm others, false information, or content targeting vulnerable groups.101 102 103## Acknowledgements104- The VAE component is from `FLUX.1 [schnell]`, licensed under Apache 2.0. 105- The text encoders are from `google/t5-v1_1-xxl` (licensed under Apache 2.0) and `meta-llama/Meta-Llama-3.1-8B-Instruct` (licensed under the Llama 3.1 Community License Agreement).