# Luma — Pushing the Limit of Efficient Inference-Time Scaling with Terminal Velocity Matching

- Company: Luma (lumalabs.ai)
- Announced: 2025-11-26
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lumalabs.ai/news/tvm
- Record: https://forck.live/items/7976-pushing-the-limit-of-efficient-inference-time-scaling-with-terminal-velocity
- Subject: Ray / Dream Machine

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.

## Evidence

Verbatim from https://lumalabs.ai/news/tvm:

> Pushing the Limit of Efficient Inference-Time Scaling with Terminal Velocity Matching

---

Record: https://forck.live/items/7976-pushing-the-limit-of-efficient-inference-time-scaling-with-terminal-velocity
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md