AI Video Continuation and Extension

The Problem It Solves

In video production, creators often face the challenge of naturally extending existing short videos. For instance, when a short video requires continuation of an action, extension of a moment, or a smoother transition at the end, traditional editing or concatenation methods might disrupt visual coherence. Specific problems include adding a prelude at the beginning, creating a natural conclusion at the end, extending a key action shot, or completing a reveal process without introducing abrupt jumps or mismatched elements. This often involves timeline expansion, but simple copy-paste can lead to rhythm breaks or content inconsistencies.

How the Skill Works

This skill addresses the above issues through core capabilities and key steps:

• Core Capability: Focused on generating transitional footage at the beginning or end of a source video to achieve natural continuation. It applies to continuing the next action, extending a moment, extending establishing shots, adding a prelude, completing a reveal, or creating a natural ending, with the core being context-aware generation of coherent new content based on the source video.

• Key Steps:
  1. Review Source Video: Record the actual MIME type, width, height, aspect ratio, duration, and byte size of the video. Identify characters, posture and action direction, camera movement, composition, shot feel, lighting, color, environment, and source audio intent. For example, if the video is MP4 format with a 10-second duration, confirm its aspect ratio as 1920x1080.
  2. Build Continuity State: Based on the continuation direction (forward or backward), extract key states from the end or beginning of the source video, including characters, posture, action direction, camera movement, composition, shot feel, lighting, color, environment, and source audio intent. Then specify a continuation beat—the unique next event that the new footage should achieve, such as "character turns and starts walking."
  3. Call Model List: Use beatra.models.list with {"capability":"video_extend"} to confirm that a model card accepts the source clip's actual MIME type, dimensions, duration, and byte size. Also check real-time upper limits for final duration, direction control, optional reference material limits, and billing basis.
  4. Choose Extension Duration: The requested billing.net_charged_creditsduration is the duration of the final returned video, not the seconds to add. It must be an integer exceeding the credible source duration and within the real-time upper limit. If the source video has reached or exceeded the limit, this skill cannot proceed and must inform the user truthfully.
  5. Execute API Call: Freeze all parameters with a stable client_request_id, then call beatra.videos.extend exactly once. Afterward, poll the same task with beatra.tasks.get until termination, delivering the returned video assets or links. When reviewing results, check seamlessness, motion direction, character consistency, camera and lighting coherence, sound behavior, final duration, and whether the new event naturally progresses.

Applicable Boundaries and Notes

• Inapplicable Scenarios: This skill is dedicated to continuation, not other operations. If the request involves redrawing the video into a new visual style, route to the video redraw workflow; modifying existing frames requires the video editing workflow; using two fixed images as start and end requires the first-last frame workflow. Additionally, it does not apply to concatenating multiple clips but rather adding transitional footage to a single source video.
• Input Requirements: Must provide an accessible source video, continuation direction (beginning or end), and the next visual beat. Reuse known scene, atmosphere, character, camera, and source audio preferences. The credible source duration is itself a hard planning input, as the extension duration is based on it.
• Billing and Changes: Extension is a paid stage. Changing the source video, continuation direction, next beat, final duration, model, or control constitutes new logical paid work, requiring a new ID and new confirmation. Each duration is the final returned video duration; it never indicates seconds to add. If the creation response is lost, replay the byte-identical frozen payload with the same stage ID.
• Model and Limits: The model defaults to auto, but real-time upper limits must be checked. If the source video duration already exceeds the limit, this skill cannot be used. When calling beatra.videos.extend, parameters must be correct to avoid duplicate submissions.
• Security and Updates: The skill uses the bundled scripts/mcp_client.py for automatic updates, but update failures do not affect existing functionality. Configuring or invoking the host Beatra Connector is prohibited; only the bundled client is used.