HiDream-O1-Image

8B image-generation model by HiDream-ai, pairing a Reasoning-Driven Prompt Agent with a Qwen3-VL text component. FP8 builds from community redistributors land near 10 GB, BF16/FP16 near 18–20 GB. Nineteen cards carry a recipe here, from a 12 GB floor up to the 5090, both Radeon boards included. It is also the only image model in this catalogue released under a plain MIT licence.