Problem
Model: fal-ai/ai-avatar and fal-ai/flashtalk
Date: 2026-05-15 ~07:00-09:00 UTC
Status: Multiple requests submitted, all stuck at IN_QUEUE with queue_position: 0 for 15+ minutes each. Never progressed to IN_PROGRESS or COMPLETED.
Steps to reproduce
- Submit a request to
fal-ai/ai-avatar with valid image_url and audio_url
- Poll status_url — returns
IN_QUEUE indefinitely
- Queue position stays at 0 permanently
- Request never completes, never fails — just hangs
Impact
- 4 separate requests submitted (3× ai-avatar, 1× flashtalk), all stuck
- Credits charged? Unknown — usage API returns 403
- No error message, no timeout, no failure notification
- No way to distinguish 'model is loading' from 'model is broken'
Expected behavior
- Request should enter
IN_PROGRESS within 60 seconds if model is available
- If model unavailable, should return an error rather than hanging indefinitely
- Status should reflect actual state (not
IN_QUEUE forever)
Suggestion
- Add model health endpoint:
GET /models/{model_id}/health → returns if model is currently serving
- Add maximum queue wait time with graceful failure message
- Expose model availability status in dashboard
Problem
Model:
fal-ai/ai-avatarandfal-ai/flashtalkDate: 2026-05-15 ~07:00-09:00 UTC
Status: Multiple requests submitted, all stuck at
IN_QUEUEwithqueue_position: 0for 15+ minutes each. Never progressed toIN_PROGRESSorCOMPLETED.Steps to reproduce
fal-ai/ai-avatarwith validimage_urlandaudio_urlIN_QUEUEindefinitelyImpact
Expected behavior
IN_PROGRESSwithin 60 seconds if model is availableIN_QUEUEforever)Suggestion
GET /models/{model_id}/health→ returns if model is currently serving