Nexisgen Clarifies Miner Scoring on Data Quality

Nexisgen explained its core incentive mechanism: miners curate and host datasets (400 clips with metadata), validators train LoRA adapters on each dataset and score results via VBench video evaluation. Miner rewards depend solely on how well their dataset improves model performance, not compute power. Data quality and diversity are the competitive edge; top-5 miners by score receive emission.

• •Miner role: curate 400-clip datasets, host on R2. No training required.
• •Validator role: pull datasets, train LoRA adapters, generate and score eval videos.
• •Scoring metric: miner reward tied to downstream model performance on VBench.
• •Competitive advantage is data curation and taste, not GPU resources.