AnalysisVisual AINovember 12, 2026

Fizgig v9.7.0 adds H3-Image LoRA training with top likeness scores

Read original source →github.com

After 12 hours of side-by-side MiniMax H3 Image trainings, the best previews scored closer to the subject than all but two of their own 45 photos on the hardest dataset (few headshots, mostly medium and wider shots).

1 source

More stories today

Open the live feed