Realtime models

Krea Realtime 14B

New

Open 14B autoregressive model for real-time, long-form streaming video — distilled from Wan 2.1, ~1s to first frame.

krea.ai · listing ≠ endorsement · re-checked 2026-09-03

Operator: Krea
Params: 14B
Output: Real-time long-form streaming video
Latency: ~1s to first frame (vendor-reported)
Licence note: Apache-2.0 weights · CC BY-NC-SA code
Access: Open weights + hosted on fal

What is Krea Realtime 14B?

Krea's open-weight 14B autoregressive video model for real-time, long-form generation, distilled from Wan 2.1 14B using 'Self-Forcing' — a technique that converts a bidirectional diffusion model into an autoregressive streamer. It supports text-to-video and video-to-video (webcam/screen) with mid-generation prompt edits, and it is also served on fal as 'Krea Wan 14B'. Speed and latency figures are Krea's own and were not independently benchmarked as of Sep 2026.

What Krea Realtime 14B does

14B autoregressive model built for real-time, long-form streaming video
Distilled from Wan 2.1 14B using “Self-Forcing”, converting bidirectional diffusion into an autoregressive streamer
Text-to-video and video-to-video (webcam/screen) with mid-generation prompt editing
Apache-2.0 weights with CC BY-NC-SA code
Also served on fal as “Krea Wan 14B” for hosted use

Who is Krea Realtime 14B for?

Self-hosting realtime video generation on a modern GPU
Building streams that need mid-generation prompt edits
Comparing open realtime models with MiniMax H3 and its fal build

Pros and cons of Krea Realtime 14B

Pros

Open weights (Apache-2.0) make realtime video self-hostable
One of the largest open realtime models (14B)
Webcam/screen video-to-video opens live restyle use cases

Cons

Speed and latency figures are Krea’s own and not independently benchmarked
Full speed is described on very high-end hardware (e.g. a single B200, reported)
Code licence is CC BY-NC-SA even though weights are Apache-2.0