AI First-Last Frame Video Generation

Problem Scenario

When you have two definitive endpoint images (a first frame and a last frame) and need to generate a short video transition between them—such as for a transformation, day-night switch, product reveal, or scene shift—traditional video editing tools may not efficiently handle this directional transition. This skill specifically addresses the problem of creating short clips with directional motion paths from two static images, eliminating the manual effort of frame-by-frame editing.

Core Capability and Workflow

The core capability is transition generation based on first and last frames. It analyzes the features of both images, constructs a transition mapping, and invokes a video model to output a coherent motion clip. Key steps include:
- Image Analysis and Mapping Construction: Examine the MIME types, dimensions, character features, background differences, etc., of the first and last frames. Identify fixed elements (e.g., background structure) and changing elements (e.g., clothing or lighting), and define the motion direction (e.g., camera push-in) and landing时刻 on the last frame.
- Tool Invocation and Task Submission: Use beatra.models.list to confirm the model supports the attributes of both images, then submit the transition instruction via beatra.videos.interpolate. The payload includes first_frame and last_frame artifact IDs, a transition prompt (e.g., "Transform the subject from daytime to evening outfit with smooth motion"), duration (duration), and a client request ID (client_request_id).
- Result Processing and Polling: After task submission, poll using beatra.tasks.get until a terminal state is reached, retrieve the returned video assets or links, and verify that it starts from the first frame, ends at the last frame, and the intermediate motion meets expectations.

Applicable Boundaries and Notes

This skill is focused on transitions between two strict image endpoints and does not apply to other workflows: if you need to animate a single image, route to an image-to-video workflow; if you need to append frames before or after a video, route to a video continuation workflow. Note the following:
- Pricing Rules: Each transition is an independent paid stage; changing the first frame, last frame, transition instruction, or model triggers a new paid work with a new request ID.
- Endpoint Requirements: The first and last frames must be confirmed images; if an endpoint does not exist, it must be created and verified in a separate image generation stage first.
- Models and Tools: Defaults to model: "auto", but use beatra.models.list to confirm model support for the current image MIME types and dimensions. Only use the bundled MCP tools (e.g., beatra.videos.interpolate), avoiding configuration of host Connectors.
- Error Handling: If a task ID is lost, look up candidate tasks via client_request_id; cancel operations should only be called via beatra.tasks.cancel when requested by the user, and the terminal state confirmed with beatra.tasks.get.