MiniMax-H3 LoRA Test: T2VA Works Best, FF2VA Requires Merging Character and Background Sheets
Lanky_Conclusion_749 · reddit · 2026-08-16
Reddit user LankyConclusion749 rented GPU time on RunPod to test fal/MiniMax-H3-Realism-People-LoRA and shared detailed notes. Findings: frequent artifacts, some weight files completely failed, Turbo LoRAs had poor quality, and both REF2VA and FF2VA methods failed. T2VA worked well, with LoRA weight sweet spot between 0.3-0.8. Key finding: for FF2VA, merging character sheet and background sheet into the first frame is the only reliable way to eliminate warping and achieve clean outputs.
More from Multimodal
- ImageNet should retire: DiT method rankings negatively correlate between ImageNet and text-to-image — YouJiacheng · 2026-10-16
- Open-source anatomy-reveal skill turns Claude into an exploded-view motion design studio — msharmas · 2026-10-16
- Flux 3 Klein vs Flux 2 Klein hands-on: the new base model 'barely usable' for T2I — Both-Rub5248 · 2026-10-16
- USTC paper uses speculative sampling to cut latency in interactive video generation — ustc · 2026-10-16
- 800,000 Gaussian splats run in WebGPU; Casado sees near-infinite scaling — generativist · 2026-10-16
- Nano Banana 2.1 image generation now live in Perplexity Computer for all users — ChrisUniverse · 2026-10-16