Creator says Qwen 3.8 Flash Next writes better H3 video prompts than closed models
Turbulent_Blood5886 · reddit · 2026-10-11
A Reddit user who has spent two months on H3 video generation shares a workflow shift: closed models like Grok mostly transcribed his instructions and would omit emotion, expression, or action unless explicitly asked, leading to stiff 15-second clips and wasted generations.
After switching to Qwen 3.8 Flash Next to draft prompts, he lets the model fill in gaps and pick fitting emotions, says some clips came out better than he envisioned, and finds its suggestions for the next shot more creative than from supposedly smarter models. He ends by asking what others use to write H3 prompts.
More from Multimodal
- Suno 4.5 Song Meets Claude-Made Slopcore Video, Surprisingly Polished — technollama · 2026-10-11
- Dev asks who else is pushing Kling 4.0 Flash as hard as he is — techhalla · 2026-10-11
- Nano Banana 2.1's real win for ad makers is text rendering, not price — faraz0x · 2026-10-11
- Sonora Studio: free local web UI for ACE-Step 1.5 music generation on ComfyUI — baxxox · 2026-10-11
- WIRED tries Google Playground: three playable AI-generated games in one day — nordicinst · 2026-10-11
- Prompt share: minimalist black-ink character illustration template — azed_ai · 2026-10-11