MiniMax H3's high prompt adherence creates stiff, frozen videos lacking subtle motion
DifficultAd5938 · reddit · 2026-08-23
A user observed that MiniMax H3's high prompt adherence results in stiff, wooden videos because the model omits natural, subtle movements unless explicitly prompted. Testing a 'slop sway' LoRA added life and realism by introducing random character adjustments. The author compares this to models like LTX and Wan, which trade prompt adherence for more generative details, and suggests a generation setting to tune this tradeoff between creativity and strictness.
More from Models
- Grok Voice 3.0 Tops Design Arena Speech Leaderboard, Up 7.4 Points — XFreeze · 2026-10-16
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16
- Meta open-sources Muse Spark: top US open model, scores just below DeepSeek — bindureddy · 2026-10-16
- TRL adds day-0 support for fine-tuning Muse Spark Open's 816B MoE with LoRA on 64 H100s — QGallouedec · 2026-10-16
- Meta's Muse Spark Open ships detailed eval docs; researcher pushes machine-readable EEE standard — evijit · 2026-10-16
- Baseten serves Meta's Muse Spark Open on day 0 at $1.25/M input tokens — baseten · 2026-10-16