MiniMax H3 REF + Fizgig Nodes: As SOTA (Ref2Img \ Img2Img) True Image Editing Model, Native 4096x1536 in 1 Minute, WF + Prompt for Every Example.
retained text
WF + Prompts - [https://civitai.red/posts/31291351](https://civitai.red/posts/31291351) Nodes - [https://github.com/shootthesound/ComfyUI-Fizgig-H3-Still/tree/main](https://github.com/shootthesound/ComfyUI-Fizgig-H3-Still/tree/main) \- 4096x1536, 18 Steps in 1Min (5060 Ti 16gb). \- Stellar Prompt \\ Spatial Understanding. \- Don't need LLM processed prompts for every image. \- Can imitate vast amount of styles based on REF. \- Great world knowledge, smart with concepts. \- Sharp looking images, low amount of artifacts. \- Way faster than Qwen 2.1 at 2+ IMG References.
document fetched 2026-09-29 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wsb4h3/minimax_h3_ref_fizgig_nodes_as_sota_ref2img