Simple tricks for continuing MiniMax H3 videos: anchor on the last frame, crossfade the latents
martinerous · reddit · 2026-10-10
A developer shares lightweight tricks for continuing MiniMax H3 videos without heavyweight all-in-one workflows. Trick one: force the model to refresh scene objects from references (awkward, needs excuses to remove objects from view). Trick two: inject the last frame(s) into the reference model so the model anchors on it; minor end flicker usually vanishes after joining latents with a crossfade. Nodes and example workflows are open-sourced at ComfyUI-Martinodes, designed as drop-in additions to existing workflows.
More from Multimodal
- One prompt turned OpenAI's 722 math papers into a 92-second animated explainer via Opus — FinanceYF5 · 2026-10-10
- Dev recreates fictional ChatGPT UIs for Windows 98, XP, Vista and 7 — Midnight_Sun_BR · 2026-10-10
- MiniMax H3 removes people from videos inside ComfyUI — RobbaW · 2026-10-10
- BudgetPix: Google and UIUC's pixel diffusion model adapts compute per image, cutting tokens to 10% — CSProfKGD · 2026-10-10
- Free Female Face Prompt Builder adds natural-language-to-SD prompt conversion — dobelmont · 2026-10-10
- Model tuning could stop AI video from fighting 2D animation style, says Andrew Carr — andrew_n_carr · 2026-10-10