HiDream-I1 Open Source Release - Next Generation Image Generation Model

news

HiDream.ai releases HiDream-I1, a new open-source text-to-image model with 17B parameters that outperforms existing open-source models in multiple benchmarks, supporting high-quality image generation in various styles

HiDream-I1 Demo

HiDream.ai officially open-sourced their latest text-to-image model HiDream-I1 on April 7, 2025. With 17B parameters, this model can generate high-quality images within seconds and has achieved leading scores in multiple benchmark tests.

Model Features

  • ✨ Superior Image Quality - Supports realistic, cartoon, artistic, and various other image styles, achieving SOTA results in HPSv2.1 evaluation, highly aligned with human aesthetic preferences
  • 🎯 Excellent Prompt Following - Performs exceptionally well in GenEval and DPG benchmarks, surpassing all other open-source models
  • 🔓 Fully Open Source - Released under MIT license, supporting scientific research and innovative applications
  • 💼 Business Friendly - Generated images can be freely used for personal projects, scientific research, and commercial applications

Available Model Versions

HiDream.ai offers three versions of the HiDream-I1 model to meet different needs:

| Model Name | Inference Steps | Features | Repository Link | |

  • |

|

-- |

| | HiDream-I1-Full | 50 | Complete version, best quality | 🤗 HiDream-I1-Full | | HiDream-I1-Dev | 28 | Distilled version, balance of efficiency and effectiveness | 🤗 HiDream-I1-Dev | | HiDream-I1-Fast | 16 | Speed version, suitable for real-time generation needs | 🤗 HiDream-I1-Fast |

Technical Highlights

HiDream-I1 employs a Mixture of Experts (MoE) architecture DiT model, combining dual-flow MMDiT blocks with single-flow DiT blocks, efficiently allocating computing resources through dynamic routing mechanisms. The model integrates multiple text encoders, including OpenCLIP ViT-bigG, OpenAI CLIP ViT-L, T5-XXL, and Llama-3.1-8B-Instruct, significantly enhancing semantic understanding capabilities.

Benchmark Results

DPG-Bench

| Model | Overall | Global | Entity | Attribute | Relation | Other | | -- |

  • |

|

|

|

--
PixArt-alpha
SDXL
DALL-E 3
Flux.1-dev
SD3-Medium
Janus-Pro-7B
CogView4-6B
HiDream-I1

GenEval

| Model | Overall | Single Obj. | Two Obj. | Counting | Colors | Position | Color attribution | | -- |

  • |

-- |

-- |

-- |

|

-- |

-- | | SDXL | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | 0.23 | | PixArt-alpha | 0.48 | 0.98 | 0.50 | 0.44 | 0.80 | 0.08 | 0.07 | | Flux.1-dev | 0.66 | 0.98 | 0.79 | 0.73 | 0.77 | 0.22 | 0.45 | | DALL-E 3 | 0.67 | 0.96 | 0.87 | 0.47 | 0.83 | 0.43 | 0.45 | | CogView4-6B | 0.73 | 0.99 | 0.86 | 0.66 | 0.79 | 0.48 | 0.58 | | SD3-Medium | 0.74 | 0.99 | 0.94 | 0.72 | 0.89 | 0.33 | 0.60 | | Janus-Pro-7B | 0.80 | 0.99 | 0.89 | 0.59 | 0.90 | 0.79 | 0.66 | | HiDream-I1 | 0.83 | 1.00 | 0.98 | 0.79 | 0.91 | 0.60 | 0.72 |

HPSv2.1 benchmark

| Model | Average | Animation | Concept-art | Painting | Photo | | -- |

  • |

|

-- |

--
Stable Diffusion v2.0
Midjourney V6
SDXL
Dall-E3
SD3
Midjourney V5
CogView4-6B
Flux.1-dev
stable cascade
HiDream-I1

Getting Started

To try using the HiDream-I1 model, please visit the project's GitHub repository for detailed installation and usage guides: HiDream-I1 GitHub Repository

You can also download model weights directly from Hugging Face, choosing the version that suits your needs.

Tip: Using this model requires installing Flash Attention, with CUDA 12.4 version recommended. The model also uses Llama-3.1-8B-Instruct, so please ensure you have adequate network connection and disk space.

HiDream-I1 Open Source Release - Next Generation Image Generation Model | ComfyUI Wiki