# HiDream.ai

> HiDream.ai is an AI research lab building open-source image generation foundation models. There are three models — HiDream-O1-Image (the flagship: a natively unified, pixel-level Unified Transformer for text-to-image generation, instruction editing, and subject-driven personalization, up to 2048×2048), HiDream-E1.1 (instruction editor), and HiDream-I1 (sparse DiT-MoE foundation). All are released under the MIT License. Each model ships in different versions as checkpoints.

## Models

### HiDream-O1-Image — flagship, unified (MIT)
Pixel-level Unified Transformer: raw pixels, text, and task conditions in one shared token space; no VAE, no separate text encoder.
- **1.5** — latest flagship release.
- **1.0** (open source, from the technical report):
  - [8B checkpoint](https://huggingface.co/HiDream-ai/HiDream-O1-Image): efficient; reaches parity with models several times its size.
  - [Dev checkpoint](https://huggingface.co/HiDream-ai/HiDream-O1-Image-Dev): distilled, 28 steps at guidance 1.0; debuted #8 (highest open-weights) on the Artificial Analysis Text-to-Image Arena.
  - Pro (200B+): scaled-up, state-of-the-art on the report benchmarks.

### HiDream-E1.1 — instruction editor (17B, MIT)
Identity-preserving, multi-step, up to ~1 megapixel; built on HiDream-I1.
- [E1.1 (current)](https://huggingface.co/HiDream-ai/HiDream-E1-1)
- [E1 (original)](https://huggingface.co/HiDream-ai/HiDream-E1-Full)

### HiDream-I1 — foundation, sparse DiT-MoE (17B, MIT)
- [Full](https://huggingface.co/HiDream-ai/HiDream-I1-Full): highest quality.
- [Dev](https://huggingface.co/HiDream-ai/HiDream-I1-Dev): distilled, faster iteration.
- [Fast](https://huggingface.co/HiDream-ai/HiDream-I1-Fast): few-step, high throughput.

## Code & weights

- [GitHub organisation](https://github.com/HiDream-ai)
- [HiDream-O1-Image repo](https://github.com/HiDream-ai/HiDream-O1-Image)
- [Hugging Face organisation](https://huggingface.co/HiDream-ai)

## Research

- HiDream-O1-Image technical report (arXiv:2605.11061): https://arxiv.org/abs/2605.11061
- HiDream-I1 paper (ACM MM 2025, arXiv:2505.22705): https://arxiv.org/abs/2505.22705

## Run it

- diffusers: `from diffusers import HiDreamO1ImagePipeline`
- ComfyUI: native nodes (Comfy-Org/ComfyUI)
- Hosted: Hugging Face Spaces, WaveSpeed, fal.ai