Xiaohongshu Carousel Maker

The Specific Problem to Solve

When posting notes on Xiaohongshu (Little Red Book), carousel images are common for multi-image formats, but manual creation often leads to disjointed visuals—each image may vary in style or narrative, weakening the overall content expression. The core issue is: how to efficiently generate an ordered series of carousel images with visual consistency, ensuring the first image hooks clicks, subsequent images follow a logical flow, and the set shares the same subject, color scheme, and composition language, rather than producing isolated image variants.

How the Skill Works

This skill leverages the Beatra platform's MCP interface, offering three creation routes for different scenarios:

• Generate by Outline: When the note outline, page roles, and visual direction are clear, use beatra.images.generate to create the sequence in one go. For example, define roles like "hook cover" or "step explanation," then specify a unified canvas (default 2K 3:4) and safe area for titles.
• Synthesize from Photo Groups or References: Users can upload up to four reference images (in order), calling beatra.images.transform. Specify which images retain subjects and which only guide style, color, or composition— the system maintains series consistency.
• Refine a Selected Page: For a single generated page, use beatra.images.edit to modify up to two localized areas, ideal for tweaking titles or details.

The standard workflow emphasizes front-loaded planning: first freeze visual anchors (subject, color palette, lighting direction, composition language), then generate in the approved sequence. All remote operations must go through the bundled scripts/mcp_client.py client—directly calling the Beatra Connector or REST API is prohibited. Before generation, verify model capabilities (e.g., via beatra.models.list checking output_relationship: "sequence") and confirm payments in two steps: one for optional Xiaohongshu queries (60 credits each) and another for image generation. Tasks are polled with beatra.tasks.get, and delivery includes checks for page order, color unity, and title safe areas.

Applicable Boundaries and Notes

• Exclusively for Multi-Image Carousels: If only a single Xiaohongshu cover is needed, use zhongcao-cover-maker instead—do not mix workflows.
• Xiaohongshu Queries Are Optional and Paid: The skill supports querying public Xiaohongshu data (e.g., notes on the same topic, comments), but each query requires separate confirmation and payment (60 credits per query); the skill functions fully without them.
• Text Handling Requires Care: By default, space is reserved for post-generation text overlay, with no promise to generate Chinese titles. If text is requested, freeze the exact copy during confirmation and verify in the output.
• Security and Updates: The bundled client checks for updates silently (every 24 hours), installing from the official CDN and verifying files—failures do not affect the current task. Never fabricate data or pass off notes from other platforms as Xiaohongshu content.