Music video VFX ask: swapping cars for elephants while keeping lip sync intact
gizy843 · reddit · 2026-11-01
- A Russian hip-hop music video creator asks for workflow advice: replace a background car with another car, then a pack of Dalmatians, clear the scene, and have an elephant walk in — all while keeping the original body motion and lip sync intact without the face turning to mush.
- Tested so far: Minimax h3 (a mess), LTX 2.5 ripple (okay from afar, no close-up detail), Wan 2.2 Animate (better motion, weak lip sync and prompt adherence). He also saw someone rotoscope the face for protection in Wan 2.2 and asks for the best current ComfyUI setup.
More from Multimodal
- Beginner tutorial: MiniMax H3 + Pika Flex LoRA generates stereo-audio video in 60s — No-Sleep-4069 · 2026-11-01
- Pika launches H3 Flex: faster-than-real-time video model, up to 80% cheaper than H3Max — techNmak · 2026-11-01
- DiffusionART-Flash open-sourced: RL for diffusion models with 25x fewer training timesteps — YangYou1991 · 2026-11-01
- Redditor shares full storyboard prompt for a 3-minute robot short film made with Sonata 5.5 — Ford-X · 2026-11-01
- Reddit's viral AI short 'Gnomes' Balloon' charms with a gnome and glowing jack-o'-lantern — love1008 · 2026-11-01
- Shizuku AI unveils in-house singing lip sync tech, beating uLipSync in quality — bdsqlsz · 2026-11-01