HiDream-O1-Image.
HiDream-O1-Image is an open-source image model from HiDream.ai. Windpaint doesn't serve it yet, so there's no model id or price. Request early access to start with the live models.
- Status
- Coming soon
- Task
- Text to image, Image editing
- Input
- Prompt, optional reference images
- Output
- Images up to 2048 × 2048
- Developer
- HiDream.ai
- Parameters
- 8B
- Licence
- MIT
- Released
- May 2026
What the model does.
Paraphrased from the official model card.
HiDream-O1-Image is a unified image model from HiDream.ai. It works directly on pixels, with no separate VAE or text encoder, and covers text-to-image, instruction editing and subject-driven personalization in one 8B model.
It generates natively at up to 2048 × 2048 and its card highlights long, multi-region text rendering. It's published in an undistilled version and a distilled Dev version, and HiDream recommends the full model for editing.
Coming soon to Windpaint.
HiDream-O1-Image isn't served, so it has no model id, no price and no request example here. When it goes live, this page gets its tiers, prices and request fields like the live models have.
Want this model? Request early access. We're adding models as we add capacity.
Live for images today: Z-Image Turbo (text to image) and FLUX.2 [klein] 4B (text to image). See Image generation for what the API does now.
Is HiDream-O1-Image licensed for commercial use?
Licence: MIT.
Yes. MIT allows commercial use, including running the model behind a paid API, with no revenue or user caps.
- MIT covers both the code and the HiDream-O1-Image models (card License section and repo LICENSE, Copyright (c) 2026 HiDream.ai). The Dev-2604 card says the same for Dev-2604. MIT requires keeping the copyright and permission notice in copies. HiDream-O1-Image repo, LICENSE
- The optional local prompt agent uses Google's Gemma-4-31B-it, which requires accepting the Gemma licence on Hugging Face. HiDream-O1-Image model card
This is our reading of the licence, not legal advice. Read the licence text before you rely on it.
HiDream-O1-Image hardware requirements.
Published by HiDream.ai for running the weights yourself. These are the developer's figures, not Windpaint's.
- Published VRAM
- Not stated
- GPUs named
- A CUDA-capable GPU
- Applies to
- No VRAM figure on the card. flash-attn recommended; avoid PyTorch 2.9.x.
- The card says a CUDA-capable GPU is required for inference. It gives no VRAM figure. HiDream-O1-Image model card
- flash-attn is highly recommended. Without it, you have to edit
models/pipeline.pyto setuse_flash_attnto False, or inference fails to import the kernel. PyTorch 2.9.x is not recommended because of a linked upstream issue. HiDream-O1-Image model card - The optional local prompt-agent backend runs Gemma-4-31B-it, a separate 31B model with its own licence. The agent can instead call any OpenAI-compatible endpoint. HiDream-O1-Image model card
HiDream-O1-Image native resolution.
What the developer's own code and examples use.
- Generates directly at up to 2048 × 2048. The CLI defaults to 2048 × 2048, and other sizes snap to valid resolutions internally. HiDream-O1-Image model card
- For single-image edits,
--keep_original_aspectresizes the reference to a max size of 2048 and uses its dimensions for the output. HiDream-O1-Image model card
Recommended settings for HiDream-O1-Image.
Steps, guidance and defaults from the developer's card and code, for running the weights yourself.
- Full model: 50 steps, guidance scale 5.0, shift 3.0, default scheduler. HiDream-O1-Image model card
- Dev (distilled): 28 steps, guidance 0.0, shift 1.0, flash scheduler. For single-image edits, Dev defaults to the flow_match scheduler, which the card recommends for editing. HiDream-O1-Image model card
- The multi-reference personalization examples use
--shift 1. The default seed is 32, andnoise_scale_startandnoise_scale_endboth default to 7.5. HiDream-O1-Image model card - HiDream recommends the full model, not Dev, for editing tasks. HiDream-O1-Image model card
HiDream-O1-Image prompt guide.
From HiDream.ai's own prompt guide, prompt tooling and examples.
- For intricate, reasoning-heavy requests, run the official Reasoning-Driven Prompt Agent first. It reasons through layout, subject attributes, physical logic and text rendering, then rewrites the request as a self-contained English prompt. HiDream-O1-Image model card
- Dev-2604 (open-sourced May 14, 2026) is tailored to text-to-image. Its prompt agent talks to
HiDream-ai/Prompt-Refineserved through an OpenAI-compatible endpoint (vLLM). HiDream-O1-Image-Dev-2604 model card - Personalization accepts layout bounding boxes (
--layout_bboxes) and skeleton/openpose references alongside the subject images. HiDream-O1-Image model card
HiDream-O1-Image limitations.
As the developer states them.
- Needs a CUDA GPU. Without flash-attn, inference fails unless you edit the pipeline file. PyTorch 2.9.x is not recommended. HiDream-O1-Image model card
Questions.
Can I call HiDream-O1-Image on Windpaint today?
No. HiDream-O1-Image isn't served yet, so it has no model id, no price and no entry in the catalog endpoint. For images today, z-image-turbo and flux-2-klein-4b are live. There's no date for HiDream-O1-Image.
What's the difference between HiDream-O1-Image and HiDream-O1-Image-Dev?
Dev is the distilled version: 28 steps with guidance 0. The full model runs 50 steps at guidance 5.0. HiDream recommends the full model for editing.
Can HiDream-O1-Image edit images?
Yes. Give it one reference image and an instruction. With two or more references, it does subject-driven personalization.
What are the HiDream-O1-Image requirements?
A CUDA-capable GPU. flash-attn is recommended, and PyTorch 2.9.x should be avoided. The card doesn't give a VRAM figure.
How should I prompt HiDream-O1-Image?
For complex requests, the developers recommend running their prompt agent first. It rewrites your request into a self-contained English prompt that works out layout and any text to render.
How many parameters does HiDream-O1-Image have?
8B.
Is HiDream-O1-Image open source?
Yes. Its weights are licensed MIT, an OSI-approved open-source licence.
Can HiDream-O1-Image be used commercially?
Yes. MIT allows commercial use, including running the model behind a paid API, with no revenue or user caps. Its licence is MIT. That's our reading of the licence, not legal advice; check the licence text before you rely on it.
How do I ask for HiDream-O1-Image?
Request early access at app.windpaint.ai/register. We're letting people in gradually while we add models and capacity.