Vigglorious Studio: chunked reference frames and guide keyframes fix character drift in long AI videos
Tablaski · reddit · 2026-10-09
Redditor Tablaski open-sourced Vigglorious Studio, a node suite tackling character drift in long Viggle-style AI videos.
Key findings and approach:
- Chunking workflow fix: instead of reusing one reference image across all chunks (>15s videos), picking a well-swapped reference frame per chunk greatly improves stability.
- Guide keyframes fight drift: H3-style guide keyframes reinforce character accuracy—one face-swapped guide supplied facial identity even when the main reference had no face.
Suite features:
- Video manager: load/trim source, adjust resolution and FPS, pick per-chunk reference frames, add guide keyframes
- Character manager: head and full-body references across angles
- Automatic head/body swaps using Qwen Image 2.1 and BFS LoRAs (replaceable)
- Optional SAM3 masks to confine guide influence to head/hair/person
- Per-sampling-step guide attention strengths and a visual sigma-curve editor
Author frames it as a starting point—optimal sampler/sigma/guide-strength combos remain for the community to explore.
More from Multimodal
- GPT-6 Luna Shows Surprising Skill at Spotting AI-Generated Images — Angaisb_ · 2026-10-09
- KAIST's LongTake: long-horizon teacher forcing keeps 30-60s video generation dynamic — kaist-ai · 2026-10-09
- AI image of the day: GPT-6 Image 2.5 recreates 1940s Chinese village newsreel — DeryaTR_ · 2026-10-09
- Star Wars AI songs strike again: Qui-Gon gets his own 'Gin and Juice' anthem — aronchick · 2026-10-09
- MiniMax H3 zombie short film goes viral on Reddit — gabxav · 2026-10-09
- Seedance 2.5 AI video praised as a "masterpiece" in viral share — BLUECOW009 · 2026-10-09