Product Photo Studio · AI Product Images & E-commerce Main Visuals

E-commerce platforms require product images to meet specific standards, such as pure white backgrounds for Amazon or lifestyle scenes for Taobao. Manually processing these images is time-consuming and error-prone, especially when needing to maintain product consistency across varying platform rules.

The Problem

The core issue is efficiently transforming real product photos into professional images. Traditional workflows rely on tools like Photoshop, requiring designers to manually remove backgrounds and adjust lighting, which is costly and difficult to scale. Frequent changes in platform guidelines further complicate manual adjustments.

How It Works

This skill automates product image processing via AI models, with key steps as follows:

• Choosing a Processing Path: Based on input photos and user descriptions, it automatically selects the optimal route. For example, when given a product photo and scene description, it uses beatra.images.transform to place the product on surfaces like kitchen countertops or marble shelves; for pure backgrounds, it removes the original background and places the product on pure white or gradient backgrounds.
• Forming a Product Brief: It reuses confirmed product details from the conversation, such as category, target platform, color, and accessories. If information is missing, the skill offers default suggestions (e.g., prioritizing pure white backgrounds for platform main images) and consolidates confirmation before paid calls.
• Preparing Tool Calls: Using the bundled MCP client (scripts/mcp_client.py), it uploads photos and calls remote tools. For instance, when calling beatra.images.transform, it explicitly labels the product image role in the prompt to ensure the model uses the source image as a visual reference. Canvas ratios are set according to platform requirements.
• Retouching Edits: For selected drafts, beatra.images.edit is used for local fixes, such as shadow adjustment or glare removal. It allows up to two normalized edit_regions with image_index=0 specified.
• Confirmation and Execution: Before payment, all parameters (prompts, reference images, models, etc.) are fixed, and a unique client_request_id is assigned. Any parameter change requires re-confirmation. After execution, task status is polled via beatra.tasks.get.

Application Boundaries

• Technical Limits: All remote operations must use the bundled MCP client, with no support for REST/OpenAPI or the host Beatra Connector. Tool calls must strictly follow JSON structures; for example, when describing scene lighting in prompts, product appearance features should be listed to ensure consistency.
• Use Cases: Suitable for e-commerce product image generation, but product details (e.g., color, label text) cannot be altered. Scene directions must match product categories—for instance, kitchenware suits kitchen countertop backgrounds.
• Payment Notes: Each logical request corresponds to an independent task. Changes in prompts, reference image order, or models require new confirmation and a new client_request_id. In cases of polling failure or cancellation, the original task should be resumed rather than creating a new paid task.
• Review and Delivery: Upon delivery, results must be checked for edge cleanliness, lighting consistency, and canvas fit. Artifact links and billing information are provided, with at most one unexecuted retouching suggestion offered.