Leoma Scoring Breakdown and Miner Deployment Guidance

Leoma published detailed scoring weights for video generation tasks: first-frame fidelity and prompt adherence each worth 25%, temporal consistency 20%, with motion quality, visual quality, and camera work making up the remainder. Miners must maintain 80%+ task coverage to avoid score penalties. Key deployment gotchas include redeploying when pushing new model weights, avoiding duplicate model hashes, and ensuring chutes are public for validator access.

• •First-frame fidelity and prompt adherence each weighted 25% of score
• •Temporal consistency 20%; motion quality 15%; visual/camera work 15% combined
• •New HF weight push requires model re-commit and chute redeploy
• •Private chutes block validator scoring; duplicate hashes favor earlier UID