Beat AI Video Studio

The Problem It Solves

When creating short videos, moving from a vague idea or disparate assets to a finished piece involves numerous decision points: How do you choose the correct generation path based on your current assets (text, images, video clips)? How do you ensure the output strictly adheres to your intended shot design, subject action, and rhythm? How do you precisely control the input and output of each API call without confusing billing stages? This skill systematizes these complex decisions into a clear, controllable workflow.

Core Capabilities & Workflow

This skill is built around seven mutually exclusive video generation routes, each tailored to a specific asset combination and target operation:
- Text-Dominant, No Still Frame: Prioritizes prompt enhancement via beatra.videos.enhance_prompt or generating a keyframe with beatra.images.generate. These are two independent paid starting points.
- Single Image as Opening Frame: Uses beatra.videos.animate for image-to-video generation.
- Single Image as End Frame (Optional First Frame): Uses beatra.videos.interpolate for frame interpolation.
- Multiple Reference Assets: Uses beatra.videos.generate_from_references to guide new video creation.
- Modifying Existing Video: Uses beatra.videos.edit.
- Extending Existing Video: Uses beatra.videos.extend, noting that the duration parameter specifies the final total length.

Key workflow steps include:
1. Asset & Intent Review: Inspect all accessible assets for true attributes (type, size, duration, etc.), and clarify the publishing scenario, action, camera movement, and must-preserve elements.
2. Route Selection & Model Inquiry: Select the single appropriate route based on assets, then call beatra.models.list to obtain the real-time model capability card (e.g., text_to_video, image_to_video) to determine constraints like duration and resolution.
3. Phased Payment & Confirmation: Before any generation operation, the corresponding video admission card or bonus confirmation card must be presented, showing a real-time cost estimate and top-up link based on the current model card. A client_request_id is generated and the request is submitted only after user confirmation.
4. Execution & Delivery: Once the payload is frozen, submit it once using the bundled MCP tool. Poll the task to completion via beatra.tasks.get, deliver the artifact, and recommend a next step (e.g., edit, extend, or regenerate) based on inspection results.

Applicable Boundaries & Caveats

• No Assembly: Multi-shot sequences should be planned as separate videos; this skill does not provide timeline assembly.
• Strict Route Adherence: Do not automatically enter video generation after prompt enhancement or keyframe creation; they are distinct paid steps.
• Prioritize Asset Inspection: Prefer user-provided assets. Upload refers only to transmission; local paths cannot be passed directly to remote tools.
• Payment Isolation: Every paid call (whether for text enhancement, image generation, or video generation) requires its own confirmation process. A single confirmation for an entire plan does not authorize any subsequent video generation calls.