Realtime Videos · Models
mock-stream
A CPU test model with the same command contract as the real models. Build and test your client against it first.
Integration testing
Overview
mock-stream does no machine learning. It waits a configurable time per step and renders a gradient whose colour is derived from the prompt and seed, with the step number and prompt text drawn on it. A prompt change is therefore visible in the pixels, which makes it useful for checking that your commands arrive.
| Property | Value |
|---|---|
| Model name | mock-stream |
| Output | 320×180, 16 fps, one video track (main_video), no audio |
| Step | 4 frames per step after about 60 ms of simulated work (configurable per deployment) |
| Commands | set_prompt, set_paused, set_seed |
| Concurrency | One session per pod |
What to use it for
- Wiring up session creation, polling, and cleanup.
- Negotiating WebRTC and receiving frames.
- Verifying that a
set_promptsent on the data channel changes the stream.
Its timing says nothing about real models. Do not use it to estimate latency, capacity, or visual quality.
Tracks
| Name | Kind | Direction |
|---|---|---|
main_video | video | out (pod to client) |
Commands
| Command | Payload | Default | Notes |
|---|---|---|---|
set_prompt | prompt: string, at most 512 chars | a calm lake at sunrise | Takes effect at the next step. Marked for moderation. |
set_paused | paused: boolean | false | Holds generation on the last frame; the session keeps its slot. |
set_seed | seed: integer, 0 to 2147483647 | 0 | Changes the colour of the rendered gradient. |
Each command is an /events/set_<field> path in the OpenAPI document returned by GET /v1/realtime/sessions/{id}/schema, and a successful command is acknowledged as accepted.