MantouAI Rebuilds the E-Commerce Visual Pipeline: One Agent That Handles Main Images, Detail Pages, Virtual Models, and Video Storyboards
The platform demonstrates a workflow-level approach to AI media generation, where the Agent manages proposals, batch generation, and iterative editing instead of just rendering images on a prompt. Its visual baseline card and floor-level editing are concrete patterns for keeping AI output consistent across a multi-image asset set, which is the real blocker for turning single-image generators into production tools.
MantouAI is an AI production platform for Chinese e-commerce merchants that treats the whole visual asset pipeline as one connected workflow instead of a single prompt-to-image tool. New tasks start with the Agent asking questions and outputting three creative proposals with a free visual comparison card; only after the user confirms does it charge credits. A "visual baseline card" locks color, lighting, and style across the set, and every generated page is split into editable sections that can be regenerated, locally edited, reordered, inserted, or deleted without re-running the whole batch.
Product information is reused through a "product archive" that can be auto-populated from Taobao/Tmall pages via a userscript. For video, the pipeline goes from purpose and platform choice to a storyboard script with timeline, shot size, camera movement, and voiceover text, then renders the final clip with support for multiple languages and aspect ratios. Pricing is credit-based: a 1K main image costs about ¥4, tasks default to 4 free images, failures don't deduct credits, and downloads unlock permanently for 10 credits per image or per set.
The platform is currently invite-only and gives 1000 free credits on registration, enough for 20 main-image tasks. It targets store owners, design studios, agencies, brands, and cross-border sellers, while acknowledging that pixel-perfect or complex composite work still needs manual Photoshop finishing.
The "visual baseline card" is a notable pattern: it fixes a global style reference before generating individual images, which directly addresses the storefront consistency problem of mismatched main images and detail pages.
The pricing model changes the incentive structure for AI media tools: proposals are free, per-section edits cost a flat 10 credits, and failures don't deduct, so users can iterate cheaply instead of paying per pixel.
Invitation-only access plus credit bundles starting at ¥9,688 signals a B2B commercial focus, not a consumer toy; the cost math treats a ¥4 main image against an ¥800 outsourcing quote, not against a hobbyist's allowance.
The Agent's multi-turn questioning before generation is a workflow choice that acknowledges garbage-in-garbage-out: it forces the user to clarify selling points and materials instead of letting a vague prompt produce a pretty but unusable image.
The lack of API access, stated openly in the FAQ, means the platform is intentionally a closed product for now, which limits its usefulness for developers who want to integrate it into their own pipelines.