Open-source ComfyUI workflow: 4K 600+ frame background replacement with MiniMax H3 tiled diffusion
Emotional_Example_12 · reddit · 2026-10-22
Open-source Bruxosdovfx-SyncTile-MiniMax-H3 v5.4 splits large/long videos into overlapping regions (spatial tiles × temporal windows); every model call predicts all regions and blends them into one canvas with a single denoising trajectory — no tile seams, no chained clips, and VRAM stays at one region's footprint, enabling 4K 600+ frame background replacement. Ships with a one-click Windows installer (auto-installs ComfyUI custom nodes, models, LoRAs with resume) plus a manual path requiring ffmpeg. Includes a Background Replace v3 LoRA — a rank-concatenation merge of five community LoRAs (1.91GB BF16, strength 1.0), structure-validated but not yet generation-tested. Technique draws on MultiDiffusion (ICML 2023). Full workflow JSON included.
More from Multimodal
- Image-to-video in ComfyUI without a local GPU: Forma custom nodes, pay-as-you-go from $5 — no3us · 2026-10-22
- Midjourney style codes --sref 1028313209 4064731875: chunky flat illustration demo — sergeantsref · 2026-10-22
- 'The Red Envelope' — a short film created entirely locally with CastWorkshop — CastWorkshop · 2026-10-22
- Claude skill rebuilds AI-generated video as a playable game in 45 minutes, then recreates it in Unreal — MattVidPro · 2026-10-22
- Surreal 'ruins of tomorrow' AI art series wows with megastructure aesthetics — piotrbinkowski · 2026-10-22
- Open-source huashu-motion turns Claude Code into a launch-film studio, 1.3k stars — AlchainHust · 2026-10-22