HiDream

HiDream.ai's open-source 17B image foundation model with a sparse Mixture-of-Experts DiT, widely adopted across ComfyUI and Civitai.

Image Generation Open Source Has API Open Source
Researched · Published · Reviewed
RECATOOLS Score
7 / 10
Founded
2023
HQ
Beijing, China
Users
280K+ model downloads (first month)
Launched
May 2025

Overview

HiDream (HiDream-I1) is an open-source 17B-parameter image generation foundation model from Beijing-based HiDream.ai, released in May 2025. It uses a sparse Diffusion Transformer with a dynamic Mixture-of-Experts architecture for high quality at low cost, and ships in Full, Dev, and Fast variants. An editing model (HiDream-E1.1) followed in July 2025. The weights ranked near the top of Hugging Face trending within a week and passed 280K downloads in a month, and are widely used in ComfyUI, Civitai, and tools like Recraft. HiDream.ai also runs the Vivago creative platform.

Advertisement

Pricing

Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.

Free
Free
Open weights (HiDream-I1 Full/Dev/Fast, E1.1) free to download and self-host from Hugging Face.

What you can produce with HiDream

  • Open weights under permissive licensing, with native ComfyUI support from close to release
  • State-of-the-art open-model scores on GenEval, DPG and HPSv2.1 at launch
  • Full, Dev and Fast variants, plus FP8 and community repacks widely available on Civitai
  • ⚠️ The Dev variant, not the Full one, is what practitioners actually recommend — several reviewers report fp8 or Q4 GGUF at around 28 steps giving better real-world results than the heavy Full model for the compute it costs
  • ⚠️ VRAM-hungry at roughly 17B with a large text encoder, and it never displaced Flux as the community default; prompt adherence and aesthetics drew praise from some and shrugs from others
  • No longer the flagship either: HiDream.ai has since shipped the E1.1 editor and a distilled 8B HiDream-O1 that ranks near the top of open-weight arena boards, making I1 the older tier of a moving lineup
Advertisement
RECATOOLS Verdict

What this is for: Open-source, cost-efficient high-quality text-to-image generation and editing that runs well in community pipelines.

Who this is for: ComfyUI and Civitai users, researchers, and developers wanting a capable open image model to self-host and fine-tune.

Availability: Free open weights on Hugging Face (HiDream-ai org); also usable via the Vivago platform. Self-hostable.

Independent AI-assisted assessment by RECATOOLS.

What people say

HiDream-I1 earned real credibility in the open-source image community on release: the 17B sparse-MoE DiT posted state-of-the-art open-model scores on GenEval, DPG, and HPSv2.1, got native ComfyUI support almost immediately, and spread fast across Civitai (including FP8 and community repacks). ComfyUI Wiki and outlets like Helpful Tiger framed it as a serious free alternative worth trying, and its permissive licensing plus Full/Dev/Fast variants made it easy to slot into existing pipelines.

Hands-on reception was more mixed than the benchmarks suggest. A recurring practitioner complaint is that the Full model is heavy and often not worth the compute—several reviewers report the Dev variant (fp8 or Q4 GGUF, ~28 steps) actually gives better real-world results—and its ~17B size with a large text encoder makes it VRAM-hungry compared to lighter options. Prompt adherence and aesthetics drew praise from some and shrugs from others, and it never displaced Flux as the community default.

HiDream.ai has since shipped newer models (the HiDream-E1.1 editor and a distilled 8B HiDream-O1 that ranks near the top of open-weight arena boards), making I1 the older tier of a moving lineup rather than the company's current flagship.

Summary of public user & expert reviews, compiled by RECATOOLS.

Frequently asked questions

It ships in Full, Dev and Fast, trading quality against cost, and a separate editing model (HiDream-E1.1) followed in July 2025.
It is built for community pipelines — ComfyUI, Civitai and tools like Recraft — and passed 280,000 downloads within a month of release.
State-of-the-art open-model results on GenEval, DPG and HPSv2.1 at release, from a 17B sparse Diffusion Transformer with a dynamic mixture-of-experts design.

About this listing

Researched on
Published on
Last reviewed

This entry was compiled from publicly available data including HiDream's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with HiDream unless explicitly stated.

Data accuracy

Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.

For the latest details, please refer to HiDream directly →

Spotted something out of date? Suggest an update →

Advertisement