Templates for CI image, video and 3D

[vecnode]

Templates:

• text-to-image
• text-to-video 2D
• text-to-video 3D
• video-to-video

(define sub-modules too)

[vecnode]

Quick status + next steps for these CI templates:

What is in place already

• There is a clear list of target templates (text-to-image, text-to-video 2D/3D, video-to-video), and the core run logic is moving toward a single orchestrated pipeline with centralized configuration.
• Recent issues are pushing toward reusable config objects and a modular pipeline, which is a good foundation for declarative CI templates.

What remains to fully resolve this issue

• For each template type (text-to-image, text-to-video 2D, text-to-video 3D, video-to-video):
  • Define a minimal config file (or Python settings object) that fully describes that template (prompt, model, sampler, IO paths, etc.).
  • Implement a small wrapper or entrypoint that takes the template config and invokes the shared pipeline/orchestrator.
• Add CI jobs that:
  • Run each template with small, fast settings suitable for CI (reduced resolution/steps).
  • Assert basic success (no crash, outputs written to expected location).
• Keep CI/template logic separate from diffusion math and device management so that templates remain stable even as internals evolve.

This will turn the bullet list in this issue into concrete, versioned templates that exercise the main use cases in a repeatable way, both locally and in CI.