Migrate text2video models to use unified runtime API

[gtong-nv]

CLI / Web UI
↓
DemoSpec + InferenceConfig
↓
T2VDemoAdapter
↓
T2VRuntime (one selected heavyweight pipeline)
↓
T2VSession (one finite prompt rollout)
↓
StepResult video chunks
├─ replay: shared MP4 output path
└─ WebRTC: shared session manager, encoder, transport, UI

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[greptile-apps]

Greptile Summary

The PR migrates the text-to-video demos onto the unified runtime and shared WebRTC serving lifecycle.

• Adds an app-owned T2V adapter, runtime, session, presets, runner configuration, and replay/WebRTC launch paths.
• Adds reusable browser controls for finite prompt-driven generations, playback, recording, and downloads.
• Extends the WebRTC manager so multiple generations can run over one persistent peer connection.
• Registers the T2V demo as a workspace package and runner plugin.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains in the eligible follow-up review scope.

No blocking failure remains.

Important Files Changed

Filename	Overview
apps/t2v_demo/app.py	Adds typed replay/WebRTC launch orchestration, prompt updates, and generated-artifact HTTP endpoints.
apps/t2v_demo/runtime.py	Implements the unified model adapter, heavyweight reusable runtime, finite inference session, and MP4 artifact collection.
apps/t2v_demo/presets.py	Defines app-owned Causal-Forcing, Self-Forcing, and Cosmos pipeline presets.
flashdreams/flashdreams/serving/webrtc/manager.py	Adds opt-in persistent peer connections that wait for and execute successive generations.
flashdreams/flashdreams/serving/webrtc/web/request_session.js	Adds prompt-generation controls, repeated-generation activation, recording, playback, and download behavior.
apps/t2v_demo/pyproject.toml	Packages the T2V application and registers its runner configuration entry point.

Sequence Diagram

sequenceDiagram
  participant User
  participant UI as CLI / Web UI
  participant Adapter as T2VDemoAdapter
  participant Runtime as T2VRuntime
  participant Session as T2VSession
  participant Output as MP4 / WebRTC

  User->>UI: Select backend and submit prompt
  UI->>Adapter: DemoSpec + InferenceConfig
  Adapter->>Runtime: Create selected pipeline
  Runtime->>Session: Start finite rollout
  loop Autoregressive blocks
    Session->>Session: Generate and finalize block
    Session-->>Output: StepResult video chunk
  end
  Session->>Output: Finalize recorded artifact
  Output-->>User: Replay, stream, or download

Loading

_{Reviews (4): Last reviewed commit: "fix precommit" | Re-trigger Greptile}