Nightshift: open-source ComfyUI video studio drafts by day, finishes the same take overnight locally
Huge_Relation2996 · reddit · 2026-10-30
Built for Comfy's Open Source Higgsfield challenge, Nightshift is an Apache-2.0 AI video studio running fully local in ComfyUI — images, video, voice and songs all generated with open weights via the Comfy SDK.
Highlights:
- Hardware-aware model recipes: reads your rig (DGX Spark gets LTX 2.5, Krea 2, FLUX.2 Klein, MiniMax H3; 8GB cards get lighter models) and reuses models you already have
- Fast drafts by day, the exact same take finished overnight: it saves the draft's joint audio+video latent and continues from it at 2x resolution — same motion and sound, not just the same seed
- Night finals queued grouped by model so each loads once
- What-if films, effects (ignite, freeze, gold, shatter), photo-to-camera-moves, product-ad and music-video presets, plus an editor with voice, music and captions
More from Multimodal
- Open-source image-to-3dlab fixes garbled text in image-to-3D, runs fully locally — Scobleizer · 2026-10-31
- AI movies weekly: voice actors ratify AI protections with 94.95% yes vote, cold box office — lmoroney · 2026-10-31
- Chained sref codes generate a cohesive 'smoking cat' crayon illustration series — sergeantsref · 2026-10-31
- Local Edit Model Battle: Qwen Image 2.1 vs FLUX 3 Klein vs FIBO 2 in ComfyUI — AuzBoss · 2026-10-31
- Getting Garments Right in MiniMax H3 ref2va Workflows Is Harder Than It Looks — __MichaelBluth__ · 2026-10-31
- Seedance demo wows Reddit with cinematic golden-hour BMX clip — BubblyPea8444 · 2026-10-31