H3 Creative LoRAs Barely Work: Pruned Models Plus Turbo LoRAs Suspected
WaitAcademic1669 · reddit · 2026-11-16
A Reddit user reports that most creative LoRAs for H3 (likely a video/image generation model) have little to no effect, and motion-related LoRAs fail entirely.
The author mainly uses NSFW creative LoRAs and suspects the combination of pruned models with turbo LoRAs is the culprit, expressing nostalgia for WAN. The post is a troubleshooting request without a confirmed answer, but it flags a common pitfall: stacking acceleration LoRAs on pruned checkpoints can neutralize LoRA weights.
More from Multimodal
- Suno Studio helps Frost School of Music students turn drafts into studio tracks — suno · 2026-11-16
- Seedance 2.5 renders a library flying to the stars — prompt shared — LudovicCreator · 2026-11-16
- Midjourney SREF share: grainy B&W faces meet thick rainbow strokes (--sref 2991809003) — tisch_eins · 2026-11-16
- Futurama reimagined in 1971 style — an AI retro restyling demo — techhalla · 2026-11-16
- Dense Mexican-themed AI illustrations showcase current image model quality — egeberkina · 2026-11-16
- NovaGarb Plugin Gives NovaClaw Agents Voice and Media Generation — OkShock6767 · 2026-11-16