AI Video Restyler and Style Transfer

The Problem It Solves

In creative and content production, there is a frequent need to quickly transform existing video footage (e.g., live-action) into a specific artistic style (e.g., anime, ink wash, cyberpunk) while strictly maintaining the original video's character identities, core actions, camera movements, composition, and overall rhythm. Traditional keyframe redrawing or fully automated video stylization often struggles to balance complete style transformation with high content fidelity, risking character distortion, loss of motion, or deviation in cinematic intent. This skill specifically addresses this tension: achieving a precise balance between style transfer and content preservation in the context of generative video restyling.

How the Skill Works

The core capability of this skill is executing a complete video style restyling pipeline. Its operation is based on deep analysis and structured preservation of the source video's content, rather than applying a simple global filter. The key steps are as follows:

1. Source Video Analysis: The process begins by inspecting and logging the source video's physical properties: MIME type, resolution, aspect ratio, duration, and byte size. Key elements such as characters, silhouettes, key actions, camera motion (e.g., pan, tilt, zoom), composition, scene, primary color palette, and the source audio intent must be identified.
2. Constructing a Preservation Brief & Edit Instruction: Based on the analysis, a clear preservation brief is generated, listing all core visual elements that must remain unchanged in the restyled output. A concise positive edit instruction is crafted, focusing solely on the one primary visual style change to be delivered for this run, to avoid instruction pollution.
3. Model Selection & Validation: Call the beatra.models.list endpoint with capability: "video_edit" to verify if a current model card accepts the source video's actual attributes (type, dimensions, duration, etc.). Unless a specific available model is user-selected, the model is kept as "auto" for optimal matching.
4. Submission & Status Polling: After presenting the confirmed restyling route, source video, references (if used), edit instruction, model behavior, and billing boundaries, a request is frozen with an opaque and stable billing.net_charged_credits client_request_id. The beatra.videos.edit tool is then called exactly once to submit the task. Subsequently, task status is polled using beatra.tasks.get until it reaches a terminal state.
5. Result Review & Delivery: Upon task completion, the returned video assets or links are delivered. The review focuses on: style match fidelity, character and object structural integrity, shot continuity, action and temporal consistency, source audio results, aspect ratio, and whether the actual output duration matches the source video.

All these operations interact with the Beatra API via the bundled scripts/mcp_client.py script, ensuring a consistent execution environment.

Scope and Important Notes

This skill has clear usage boundaries:
- It is focused on restyling an entire existing video. If the request is to remove AI-generated artifacts and make the footage more photorealistic, the request should be routed to the video photorealization workflow. If the request involves adding footage before or after a segment, the video continuation workflow should be used.
- Any paid modification to a submitted task (e.g., changing the source video, modifying the edit instruction, adjusting the model or aspect ratio) constitutes a new logical billing job, requiring a new client_request_id and fresh confirmation.
- The processing of source audio (e.g., background noise, dialogue) is best-effort; results may vary by model and consistency with the original audio is not guaranteed.
- The style match and structure preservation of the output depend on model capability and instruction quality; these are review checkpoints, not promises of exact preservation.