Wan 2.2 consistency experiment chains three 14s clips into one action scene
Tokyo_Jab · reddit · 2026-09-25
Creator TokyoJab shares H3 Consistency Experiment #2 ("Long Walk Home"), targeting character consistency and fast action speed. Method: generate a 14-second clip, then use its last 3 seconds (at half resolution) plus a character sheet and car reference as input for the next generation — chaining three 14-second clips in one generation is enough to keep continuity without losing consistency. Workflow to be posted soon.
More from Multimodal
- ElevenLabs MCP x Claude: Voiceovers to Branded Ads in a Single Chat — iamfakhrealam · 2026-09-25
- Grok 4.7 pairs with Blender to build surprisingly detailed Raptor engine 3D models — CurieuxExplorer · 2026-09-25
- Google launches Gemini 3.8 Flash TTS and Flash-Lite TTS, its most expressive audio models yet — CurieuxExplorer · 2026-09-25
- MiniMax H3 video generation shows persistent redness and excess contrast users can't fix — kyahinaamrakhe-1 · 2026-09-25
- Indie creator shares a storyboard workflow for AI-generated action sequences — Educational_Wash_448 · 2026-09-25
- WanPE: 397B Prompt Enhancer Boosts 30s Video Preference by 50.86 Points — Yubo Zhu · 2026-09-25