Recreating a viral fight video with MiniMax H3: full reference-generation workflow
MixZealousideal9359 · reddit · 2026-09-04
Reddit user MixZealousideal9359 recreated a viral X video using MiniMax-H3-Pruned-Ref-Delta-Fused-r1024-ComfyUI (0.4MP, 5 steps), producing a 15-second painterly gouache 2.5D fight sequence.
The post shares the full reference-generation setup: four subject definitions lock the character's identity and wardrobe, the abandoned concrete pool fight-pit environment and lighting, the opponent's physique, and the black ink-vine tattoo effect — with a retention analysis showing all references fully preserved across 8 shots.
A useful end-to-end example of structured subject definitions for character consistency in long sequences.
More from Multimodal
- Synthesia Assistant rolls out to all plans: full brand videos from a prompt — Scobleizer · 2026-09-04
- Could Astra generate full 3D scenes and spark a new non-imagegen AI art genre? — flowersslop · 2026-09-04
- WebMCP demo: an AI agent DJs in the browser with hundreds of direct tool calls — No_Guide_8697 · 2026-09-04
- Saudi HUMAIN launches free Voice model with Saudi dialect STT/TTS and developer SDK — aziz4ai · 2026-09-04
- DLSS 5 ships: one-step pixel diffusion generates photorealistic 4K frames in real time — ctnzr · 2026-09-04
- User edits cinematic Tesla Cybercab video entirely with Grok Build — elonmusk · 2026-09-04