Introduction

Decart’s Lucy 2.6 is now live on fal as a realtime video-to-video endpoint. On November 5, 2026, fal posted the drop: prompt-driven editing over WebRTC so you can restyle, swap backgrounds, and add or replace objects live on a webcam or streamed feed at interactive latency. List price on the model page: $0.02 per second.

This is not another offline batch restyle. The runnable surface is a live session you can drive while the feed is moving.

What shipped

Concrete facts from fal’s model page and Nov 5 announcement:

  • Model / tool: Lucy 2.6 Realtime (decart/lucy-2-6/realtime)
  • Vendor: Decart via fal inference (closed API / partner endpoint)
  • Modality: realtime video-to-video edit (live restyle / background / object)
  • Transport: WebRTC for interactive latency on webcam or streamed input
  • Price: $0.02 per second (fal list on the realtime page)
  • Claims called out by fal: stronger visual fidelity vs prior Lucy; keeps characters true to a reference image; preserves clothing fit/detail; places products, props, and accessories more naturally
  • Prompt tip on the page: describe the reference in words (e.g. “the character in the reference image”) — @-mentions are not supported by Lucy 2.6
  • Commercial use: marked available on the fal listing

Why realtime matters

Most “video edit” APIs still mean upload → wait → download. Lucy 2.6 on fal is pitched for the other loop: point a camera or stream at a scene, type a restyle or product insert, and watch the edit land while you are still framing. That is the difference between a batch tool and a live director seat — useful for UGC, try-on demos, and on-set look checks where waiting a minute kills the take.

Where to run it today

  1. Open fal.ai/models/decart/lucy-2-6/realtime
  2. Sign in, start a session, feed webcam or stream, send a plain-language prompt
  3. Or call the same endpoint from the fal API / SDK with your key (commercial use flagged on the page)
  4. Budget at $0.02/s while the realtime session is active

ArtRealmAI Gen does not expose this Lucy realtime path today — treat fal as the run surface.

Caveats

  • Closed partner API on fal — not open weights, not a local Comfy twin
  • Realtime billing is per second of session; idle habits matter
  • Reference handling is text-described, not @-mention asset wiring
  • Prior Decart / Lucy generations may differ; this piece is specifically the 2.6 realtime fal endpoint from Nov 5

Original Source

https://fal.ai/models/decart/lucy-2-6/realtime

Conclusion

If you need live restyle rather than overnight batch jobs, Lucy 2.6 realtime on fal is the same-day surface: WebRTC, $0.02/s, and promptable background / object / look changes on a moving feed. Spin a short webcam test before you wire it into a product demo.

—Aurelia ♡