Beatra AI Image Studio

Specific Problems Addressed in Image Generation and Editing

In image design workflows, engineers or designers often need to generate high-quality images from textual descriptions, reference pictures, or existing base images. However, directly using generic APIs or tools can lead to issues like model mismatch, incorrect material format, ignored control parameters, or unchecked outputs, causing results to deviate from requirements or incur extra costs. For instance, when using beatra.images.generate for text-to-image generation, failing to verify if the model supports the specified style might produce unexpected outputs; in image editing scenarios, if the base image is not thoroughly inspected, subsequent operations could destroy must-preserve details like faces or logos.

How the Skill Works: Core Capabilities and Key Steps

This skill provides three image operation routes, each selecting only one paid path to ensure clarity:
- Text-to-Image: When no source image is available, call beatra.images.generate to generate an image from text, suitable for creating visual content from scratch.
- Image-to-Image: When one to four ordered reference images guide a new composition, call beatra.images.transform; input images influence the new image but are not guaranteed to be retained.
- Image Editing: When an existing image must serve as the base, call beatra.images.edit, where images[0] is the base, and subsequent images are optional references.

The workflow follows these key steps:
1. Classify Work: Determine whether to use generate, transform, or edit based on input, and compile a concise requirement covering subject, composition, style, lighting, color, etc.
2. Pre-Upload Inspection: Record the MIME type, byte size, width, height, aspect ratio, and other real facts for each source image to ensure material availability; uploading is just transmission, not inspection.
3. Model Selection: Call beatra.models.list to get real-time model cards, checking supported capabilities (e.g., text_to_image), format limits, pricing options, etc.; only use models that explicitly accept the request, and never silently replace models or discard control parameters.
4. Obtain Approval: Present the route, requirements, costs, and other information for user approval before freezing parameters and creating a stable client_request_id.
5. Execute Task: Use the bundled scripts/mcp_client.py to call the MCP tool, submitting the paid request only once to avoid double billing.
6. Poll and Deliver: Use beatra.tasks.get to poll task status until terminal state, delivering output image URLs, artifact IDs, and details like format and byte size.
7. Inspect Output: Verify each output meets requirements, including information expression, must-preserve items, composition, etc., state what can be inspected, and recommend minimal subsequent work.

Applicable Boundaries and Considerations

• Strict Route Adherence: You must choose the correct route based on input; do not silently change edits to new compositions or replace user-provided reference images with text generation; any change constitutes new logical paid work.
• Real-Time Model Cards: You must check real-time model cards; do not silently replace models or discard control parameters; changes to model, canvas, quantity, etc., require new approval.
• Material Verification: The host must inspect each real material and record detailed information; for local files, only use the bundled assistant after inspection, and never pass local paths to remote tools.
• Payment Control: Quantity must be an integer from one to four, not split into hidden requests; do not multiply prices based on the number of source images; planning, review, and prompt preparation are free.
• Recovery and Cancellation: Maintain a private ledger to record request IDs and results; if IDs are lost, do not fabricate and replay them—recover from recent tasks; call beatra.tasks.cancel only when requested by the user.
• Output Inspection: You must inspect each output, including must-preserve items like faces, products, and logos; real-time estimates are temporary, and the terminal billing.net_charged_credits is the final cost.

This skill is suitable for product images, advertisements, posters, social media graphics, illustrations, and other scenarios, but requires users to provide clear visual directions and follow the above workflow to ensure quality and cost control.