Training Character LoRAs for Minimax using Ostris AI Toolkit
No_Statement_7481 · reddit · 2026-08-27
The author shares practical experience training character LoRAs (voice and likeness) for Minimax using the Ostris AI Toolkit.
Key Config & Data:
- Hardware: RTX 5090 requires offloading; RTX 6000 preferred. Dual-dataset strategy (images + video) for voice.
- Params: 2000-2400 steps (80-90 mins); Learning rate 0.0002; Differential Guidance on (Level 3); LoRA weight 0.65-0.85.
- Data: 15-60 images (1024x1024, 20+ recommended) for likeness; 6-12 videos (512x512) for voice.
- Automation: Used Qwen3 VL8b for auto-captioning.
The demo successfully recreated the character Enid from Wednesday, bypassing the model's bias towards Jenna Ortega.
More from Multimodal
- Open-source LoRA turns MiniMax H3 into seamless 360-degree orbit videos from a single photo — camenduru · 2026-10-15
- Neon Noir Elegance prompt template goes viral with striking pink-blue two-tone outputs — LudovicCreator · 2026-10-15
- Krea 2 Turbo + px16 LoRA generates game-ready pixel art in 5s, where GPT Image 2.5 fails — Angaisb_ · 2026-10-15
- How Goldfish Made Its Kung Fu Scenes: Real Nanquan Choreography + Seedance 2.5 — JunieLauX · 2026-10-15
- Custom px16 LoRA makes Krea 2 Turbo churn out pixel icons at 5s each — Angaisb_ · 2026-10-15
- Prompt share: 'Crimson Footsteps' makes one red element pop in monochrome AI photos — azed_ai · 2026-10-15