A quick Minimax H3 news round-up - 20th September 2026
retained text
Another quick Minimax H3 news and goodies round-up, for those who may have missed some items. -> A new "Depth-based camera re-control for MiniMax H3 + Viggle Meridian inside ComfyUI". https://github.com/NyckM/3d-Camera-control-H3-Minimax/tree/main -> A ComfyUI fix for what are said to be tile seam artifacts, a fix which looks as though it's on the verge of merging. I can't see them in the demo video, but I trust they're there. https://github.com/Comfy-Org/ComfyUI/pull/16422 https://www.reddit.com/r/StableDiffusion/comments/1wlltxm/fix_for_the_minimax_h3_vae_grid_tileseam_artifact/ -> A new 1980s Horror Movies LoRA, which also appears to offer useful 'night in the graveyard' lighting. Suggested trigger is *80s_horror* If you need some emaciated ghouls for your graveyard, I see there's also a new Weight Slider LoRA. https://huggingface.co/neph1/1980s_horror_movies_minimax_h3 https://huggingface.co/Playtime-AI/MM-H3-Weight_Slider -> A 1950s Scifi Minimax H3 LoRA. The trigger is *50s_scifi* and if you want the Technicolor look in the samples you have to prompt for it. https://huggingface.co/neph1/1950sScifiMinimaxH3 -> A new 'Simply Advanced MiniMax H3 workflow', which the maker has made me aware of and asked me to feature. Toggles decide if each media input is received as a reference object, or just set as a Guide, or both. There's also an LLM dashboard as a Subgraph... "which includes a 4-stage Ref2VA pipeline which divides the workload into more concise tasks, leaving the 4th stage with the sole task of just assembling the previous results". You can add your own LLM. https://www.reddit.com/r/comfyui/comments/1wk9qmg/simply_advanced_minimax_h3_workflow/ -> And finally, Qwen Image 2.1, available now for the very latest ComfyUI (release-day support added). The model's ability to generate a native 2k .PNG of precise typography, with an alpha channel, suggests it might be used to quickly output a video title, credits, inter-titles or end credits for use as reference in H3 videos. One might even prompt H3 to fade in or animate the still typography. The model's smallest Comfy file-set is 15Gb in total. Note the current official ComfyUI T2I workflow suggests a hefty 25 steps for a 2k image, so you may want to wait for the Turbo speed-ups and then see how well the typography holds up under these. https://huggingface.co/Comfy-Org/Qwen-Image-2.1 https://github.com/QwenLM/Qwen-Image-2.1/blob/main/prompt_rewrite/prompts/system_prompt_t2i.txt (official LLM-friendly prompt guide, to feed to a LLM) ~ OLD POSTS ~ https://old.reddit.com/r/comfyui/comments/1wku9uj/a_quick_minimax_h3_news_roundup_19th_september/ https://old.reddit.com/r/comfyui/comments/1wjywf7/a_quick_minimax_h3_news_roundup_18th_september/ https://old.reddit.com/r/comfyui/comments/1wj08mp/a_quick_minimax_h3_news_roundup_17th_september/ https://old.reddit.com/r/comfyui/comments/1wi3exl/a_quick_minimax_h3_news_roundup_16th_september/ https://old.reddit.com/r/comfyui/comments/1wh4tu3/a_quick_minimax_h3_news_roundup_15th_september/ https://old.reddit.com/r/comfyui/comments/1wgc4lj/a_quick_minimax_h3_news_roundup_15th_september/ (See 15th September post, for links to older posts) https://old.reddit.com/r/comfyui/comments/1w5i9iq/a_quick_minimax_h3_news_roundup_2nd_september_2026/ (See 2nd September post, for links to even older posts)
document fetched 2026-09-21 00:36 · canonical https://reddit.com/r/comfyui/comments/1wlnqid/a_quick_minimax_h3_news_roundup_20th_september