MiniMax H3 AI Video Creation

Specific Problem

Video creation often starts with a rough brief, where creators may have only a scene description, an image, or vague ideas, making it hard to generate high-quality videos tailored to publishing scenarios. Manually refining prompts is time-consuming and error-prone, while parameters like resolution, duration, aspect ratio, and cost lack transparent control during generation. Common AI video tools often suffer from unstable output quality, limited format support, or inability to precisely match brand visual requirements.

Core Capabilities and Workflow

This skill leverages the MiniMax H3 model through a structured workflow to transform creative inputs into publishable videos. Core capabilities include:

• Intelligent Brief Enhancement: Running beatra.videos.enhance_prompt on rough descriptions to generate production-ready prompts, automatically filling in details like camera movement, lighting, rhythm, and sound.
• Multi-Format Generation Routing: Selecting tools based on input type:
• beatra.videos.generate (text-to-video)
• beatra.videos.animate (image-to-video)
• beatra.videos.interpolate (start/end frame control)
• beatra.videos.generate_from_references (reference-based generation)
• Specification Control and Cost Transparency: Enforcing model="minimax-h3" with support for 720p/2K resolution, 4-15 second duration, and aspect ratios like 16:9. Before each generation, beatra.models.list retrieves real-time pricing, with costs calculated per output second (e.g., text-to-video) or separately for input/output seconds (e.g., reference-based generation), and an estimate is generated.

The key steps follow a technical workflow:
1. Route Selection: Matching tools to the creative starting point, avoiding routing to unrelated functions like video editing.
2. Brief Convergence: Reusing existing information (subject, camera movement, brand details) to form a focused prompt; for reference materials, marking each reference's purpose in array order, e.g., “Image 1 is the featured product”.
3. Cost Confirmation: Presenting an entry card with real-time estimates, resolution, duration, etc., requiring user confirmation of balance (entry package ¥29/11,000 points) before creating a stable client_request_id.
4. Execution and Review: Calling tools via scripts/mcp_client.py, polling tasks with beatra.tasks.get; delivering artifact links, dimensions, task IDs, and billing.net_charged_credits, with modification directions based on subject consistency, audio-visual alignment, etc.

Applicable Boundaries and Notes

• Platform Dependency: All operations must be performed through the bundled scripts/mcp_client.py, with no support for REST/OpenAPI fallback or host Beatra Connector calls.
• Function Limitations: Only suitable for MiniMax H3 model capabilities; does not support video editing, extension, or advanced parameters like negative prompts and random seeds.
• Cost Control: The registration bonus of 600 points is usually insufficient for a single video; pre-top-up is required. Billing is based on real-time model card prices, and user confirmation must be obtained before generation.
• Material Requirements: Reference-based generation accepts up to five images, three videos, and three audio clips, with total duration limits of 15 seconds; local files must be uploaded via beatra.assets.upload.
• Error Handling: Errors like insufficient_balance are relayed with the charge URL, but the skill will not automatically retry generation.