Descriptions:
HiDream-O1, the MIT-licensed open-source image generation model, is now natively supported in ComfyUI, and the Nerdy Rodent channel provides a hands-on walkthrough of building a complete workflow from scratch. The tutorial uses the base model (not the distilled dev variant, which lacked ComfyUI support at time of recording) and introduces the channel’s signature “rodent method” — organizing nodes into color-coded modular groups to keep workflows readable and maintainable.
The video covers the full range of HiDream-O1’s capabilities: basic text-to-image using the dedicated HiDream latent image node (required due to specific supported resolutions), multi-reference image input via the HiDream-O1 reference images node, image-to-image with the RES multistep sampler at 0.6 denoise, inpainting with manually crafted masks, and upscaling to 2560×2560 via both image and latent upscale methods. Practical settings throughout include CFG 3 and 28 steps, with notes that usable compositions emerge in as few as 8 steps on the base model.
HiDream-O1 is benchmarked informally against Ernie Image and Z Image on identical prompts. HiDream leads on prompt adherence and in-image text generation, while Ernie Image produces better fur and fine detail at lower resolutions (1440×1440), and Z Image — though weaker on prompt following — is highlighted as a useful post-processing partner for toning down HiDream’s tendency toward oversaturated, plasticky outputs.
📺 Source: Nerdy Rodent · Published May 15, 2026
🏷️ Format: Tutorial Demo
