AI Image Generation Studio

The Problem Being Solved

Visual content creation often requires generating high-quality images from textual descriptions or existing visuals for use in product displays, advertising, poster design, or social media. Traditional methods may involve multiple tools, manual adjustments, or cumbersome processes, leading to inefficiency, inconsistent results, or difficulty meeting precise requirements. This skill aims to streamline text-to-image generation, image transformation, and image editing through a unified AI interface, ensuring outputs align with specific visual directions and publishing scenarios.

Core Capabilities and Workflow

The skill offers three core capabilities, automatically selecting the best route based on input type:

• Text-to-Image Generation (generate): When no source image is available, use beatra.images.generate to create images from pure text, ideal for conceptual illustrations or starting from scratch.
• Image Transformation (transform): With one to four ordered reference images guiding new composition, call beatra.images.transform. These inputs influence the new image's composition, style, or elements but do not guarantee any will serve as a base.
• Image Editing (edit): Using an existing image as the base for edits, invoke beatra.images.edit. images[0] is the base image, with subsequent images as ordered references for background modifications or detail adjustments.

The workflow follows a golden path:
1. Classify work as generate, transform, or edit, and compile requirements covering subject, composition, style, lighting, colors, publishing scenario, exclusions, and must-retain items.
2. Inspect each source image before upload for properties like MIME type, byte size, width, height, aspect ratio, alpha channel, and animation status to ensure data authenticity.
3. Call beatra.models.list to check real-time model cards, confirming supported capabilities (e.g., text_to_image, image_to_image, or image_edit), and verify input limits, control items, and pricing options.
4. Present the selected route, ordered sources and their roles, visual requirements, must-retain items, quantity, canvas, model behavior, and estimated costs, then freeze parameters after user approval.
5. Execute requests using the bundled script scripts/mcp_client.py, passing JSON parameters via standard input, avoiding host connectors or other fallback methods.
6. Poll task status, deliver each returned image's URL, artifact ID, and real dimensions, format, etc., reporting the final model and actual billed credits.
7. Check each output against requirements for information expression, subject retention, composition, style, etc., and recommend focused edits or regeneration, pending separate approval.

Applicable Boundaries and Notes

The skill is designed for tasks like product images, ads and brand visuals, posters, social media graphics, illustrations, conceptual art, photo variants, and background modifications, but has clear boundaries:

• Strict Route Selection: Each logical result selects only one paid image route, avoiding silent changes. Do not replace user-provided references with text generation or promise pixel-perfect retention.
• Hard Input Requirements: Non-empty visual direction is mandatory; transform requires one to four ordered reference images; edit requires a base image. Only ask for missing hard inputs or choices that substantially alter results.
• Model and Control Item Management: Checks based on real-time model cards are mandatory; no silent substitutions or discarded control items. Any control item changes (e.g., quantity, relationship, seed) require new approvals and constitute new paid work.
• Billing and Recovery Mechanisms: Billing is per successfully generated image; for partial successes in multi-output tasks, only successful images are charged. Recovery requires maintaining a ledger; do not fabricate request IDs or replay paid requests.
• Output Inspection Necessity: Check each output post-delivery to ensure compliance with must-retain items (e.g., faces, products, logos, text). Only recommend minimal subsequent work and await separate approval.