LTX 2.5 Fast runs stuck at dispatch with queue_position 0

APIs
1

On October 6, 2026, we observed inconsistent start delays for six-second videos using lightricks/ltx-2.5-fast. All times below are UTC.

1. Run never started for approximately 20 minutes

Run: run_01M48XSZ9B8DYZ8SM3VSR8WTQV
Format: 6 seconds, 720p, 16:9
Created: 15:37:36.378513Z
We cancelled it at 15:57:43.480021Z.

Repeated status polling showed these selected fields:

{

"status": "queued",

"progress": 0,

"queue_position": 0,

"error": null,

"capacity_wait": {

"status": "delayed",

"message": "Your request is taking longer than usual to start. We're still trying."

}

}

Its events reached run.dispatching at 15:37:44.054997Z, but there was no run.started event. Cancellation confirmed started: false and credits_charged: 0.

During the stall:

  • The complete account run list showed this was the only active run.
  • The account had 942 provider credits.
  • The public status reported API, scheduler and worker pool operational.

2. Another 720p run waited almost six minutes

Run: run_01M48Z2AETRV7VKEYP4KZ6TBT3

Provider timestamps show 5 minutes 53 seconds waiting, followed by only 16.8 seconds rendering.

3. A later 480p run completed much faster

Run: run_01M48ZFS96WRAS8BZKVMAEC6AS

It waited approximately 6 seconds and rendered in 9.2 seconds. This comparison does not establish that resolution caused the delays.

Please trace these runs and clarify:

  1. What blocked dispatch on the first two?
  2. What exactly does queue_position: 0 mean?
  3. Were model capacity, cold starts, account limits, or stalled worker assignments involved?
  4. Does cancelling an uncharged queued run and submitting one fresh attempt help, or simply return it to the same capacity pool?
  5. What queue-start latency should we expect for customer-facing API use?
2

Hi, latency varies on our platform due to a variety of factors, including whether you are an enterprise user, current demand to our platform, and downstream GPU provider capacity. Cancelling and re-submitting will not help, and non-enterprise users should expect latency anywhere between instant and ~10 minutes for video generation during peak demand times.

3

Okay understood. How can we get an enterprise plan? What is the minimum cost? Does this reduce/eliminate latency?

Also - do you have a rough idea on general peak demand times?

4

Hi, for enterprise info, please fill out the form at Enterprise AI Video & Realtime Avatar API | Protoface and a team member will contact you within a business day. Enterprise users receive some prioritization, but latency is never eliminated — like the vast majority of platforms, we operate with probabilistic latency targets for enterprise customers (ex. p9X < XX minutes). I don’t have any information on peak demand times — the exact time that p95 demand occurs varies every day.

5

Understood thank you for the info