Skip to content

feat: support ordered multi-image references for video generation - #3

Draft
giaan1702 wants to merge 2 commits into
suncodevn:mainfrom
giaan1702:feat/multi-reference
Draft

giaan1702 wants to merge 2 commits into
suncodevn:mainfrom
giaan1702:feat/multi-reference

Conversation

@giaan1702

Copy link
Copy Markdown

Related to #1.

Video generation currently accepts one image reference. This adds an optional ordered reference_images list with per-image label/role guidance, preserving attachment order and repeated image positions. Existing single-reference requests retain their behavior; conflicting reference modes are rejected.

Includes validation, owned-file resolution, combined decoded byte enforcement, job persistence, dispatcher/client integration, OpenAPI definitions, usage documentation, and six regression tests covering ordering, repeated attachments, ownership, metadata, byte limits, and idempotency.

Validation: 30/30 architecture + Studio tests pass; syntax check passes for 63 files; git diff whitespace check passes. Tests use offline fixtures/local HTTP servers. No new paid generation was submitted, and distributed load was not validated.

Draft for API review: the 100-item list cap is a local resource bound, not a verified upstream maximum. Combined images remain subject to MAX_IMAGE_MB (default 20 MiB). Labels/roles are prompt guidance, not first/last-frame controls. See docs/video-image-references.md.

This PR is independent of persistent generation sessions and does not include its API field, ComfyUI connector, or machine-specific configuration.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant