HiDream-O1-Image combines generation, editing and multi-reference personalization, with skeleton and layout guidance for directing character scenes. It is a promising local option for character art, though consistency still needs practical testing against a human-approved character sheet and new poses. Full and Dev use different inference settings; the authors recommend Full for editing. Save the checkpoint and settings with approved assets so later revisions use the same setup. The optional prompt agent adds a separate local model or an external API, with its own resource needs and terms.
Commercial use & license
The official card lists MIT for HiDream-O1-Image code and models. Check optional backend-model terms and rights to all reference assets separately.
Adult content rules
The card does not establish whether explicit sexual content is permitted. Local execution still requires checking asset rights and any optional backend terms.
What you need
CUDA-capable GPU, the complete selected checkpoint and the official Python dependencies. Install flash-attn or disable its use as the setup guide requires; measure memory for your chosen workflow. The optional local prompt agent also needs separate Gemma weights.
Ranks are recommendations for the task described, based on reviewed documentation. Model ranks cover non-explicit development work. We have not run practical benchmarks. Check content policies and commercial terms separately. How the ranks work →
Sources & further reading +
Last checked 5 October 2026.