Seamless action scenes in ComfyUI: FL2V + Minimax H3 guide workflow on a 3090
VisionWithin · reddit · 2026-08-27
The author experiments with ComfyUI's FL2V node plus "Add Guide for Minimax H3" for seamless action scenes: a one-second guide video is enough for a visually seamless result, though audio must be produced in post, guided only by the video frames and prompting.
Everything was made with 16:9 480p videos and an 8-step turbo LoRA; one generation takes 2.5 minutes on an RTX 3090, leaving plenty of time to redo shots and refine prompts. Main limitation: no subject consistency except by chance—more serious work needs the Ref2V node for guided generation. Verdict: a lot of fun.
More from Multimodal
- Open-source LoRA turns MiniMax H3 into seamless 360-degree orbit videos from a single photo — camenduru · 2026-10-15
- Neon Noir Elegance prompt template goes viral with striking pink-blue two-tone outputs — LudovicCreator · 2026-10-15
- Krea 2 Turbo + px16 LoRA generates game-ready pixel art in 5s, where GPT Image 2.5 fails — Angaisb_ · 2026-10-15
- How Goldfish Made Its Kung Fu Scenes: Real Nanquan Choreography + Seedance 2.5 — JunieLauX · 2026-10-15
- Custom px16 LoRA makes Krea 2 Turbo churn out pixel icons at 5s each — Angaisb_ · 2026-10-15
- Prompt share: 'Crimson Footsteps' makes one red element pop in monochrome AI photos — azed_ai · 2026-10-15