From the source
Luma presented Terminal Velocity Matching (TVM), a method for efficient inference-time scaling, and challenged readers to distinguish between images generated by 4-step TVM and 50x2-step diffusion.
Pushing the Limit of Efficient Inference-Time Scaling with Terminal Velocity Matching
From the source
Pushing the Limit of Efficient Inference-Time Scaling with Terminal Velocity Matching
lumalabs.ai
- Announced
- 26 Nov 2025
- First seen by forck
- 03 Sep 2026 14:37 UTC (≥280d after)
