MantouAI Rebuilds the E-Commerce Visual Pipeline: One Agent That Handles Main Images, Detail Pages, Virtual Models, and Video Storyboards
E-commerce visuals are being rebuilt by Agents: one platform handles main images, detail pages, real shots, replicas, and videos
It's not "type a prompt and wait for an image" — it's breaking the real workflow of an e-commerce designer into steps an AI can take.
People working in e-commerce know this cycle:
- Outsourcing a set of main images: quoted from ¥800, wait 3 days, go through 5 rounds of revisions;
- A detail page needs 12 sections, and the designer's schedule is booked until next month;
- Want to replicate a competitor's viral visual style, but worried about being called out for copying;
- Want real-life scene shots, which require a studio, models, location, and post-production;
- Short videos are worse: script, storyboard, voiceover, editing — one pipeline needing four different suppliers.
The bigger headache is style inconsistency: the main image uses one color scheme, the detail page another, the model's face changes, and the store looks like three different brands.
Recently I did a deep trial of a platform called MantouAI, an e-commerce visual production platform. What it does is not "yet another AI drawing tool" — it rebuilds the entire pipeline mentioned above. This article explains the product's capabilities, the actual workflow, output quality, and cost all at once.
1. What it actually is
MantouAI is an AI visual production platform aimed at e-commerce merchants, covering:
| Task type | What it can do | Status |
|---|---|---|
| Main image | Single main image / image set | Live |
| Detail page | Long-form detail page, supports 1~16 sections | Live |
| Replica | Reference a competitor image / template, replace with your own product | Live |
| Real shot | Realistic scene images without an actual shoot | Live |
| Video | Idea → storyboard script → final video, supports voiceover | Live |
| SKU | SKU image production | Under construction |
The biggest difference from "one-prompt-to-image" is that it is a complete production pipeline:
Product archive
→ AI understands product
→ Outputs 3 creative proposals
→ User selects a proposal
→ Generates a visual baseline card
→ Layer-by-layer batch generation
→ Section-level secondary editing
→ Long-image stitching preview
→ Unlock download / share
The workspace has four parts: left is the task and product list, middle is the parameter form and Agent chat area, right is the generation results area, top is the credits top-up entry. The overall interaction is closer to "chatting with a designer" than "tuning parameters to pull a gacha."
2. Five capabilities that actually save time
1. See 3 proposals first, then decide whether to spend money
When you create a task, the Agent doesn't silently generate 8 images. Instead, it talks to you over several rounds, outputs 3 creative proposals, and generates a visual comparison card for free.
Each proposal explains the style direction, color scheme, and key highlights, and the recommended one carries a badge. You compare them directly, then click 【Confirm Proposal, Start Generating】.
This design solves the most painful e-commerce issue: if the direction is wrong, everything after it is wasted. Spending 20~40 seconds to look at proposals beats spending 8 minutes generating a pile of useless images.
After you confirm a proposal, the system first generates a visual baseline card to lock the color tone, lighting, and style for the entire set, then generates layer by layer. This is the key to keeping the whole visual set consistent.
2. Product archive reuse + one-click Taobao/Tmall import
Filling in product name, selling points, materials, and specs for every task is pure grunt work. MantouAI uses a product archive to consolidate this:
- Basic info: name, category, brand, price band, target audience, SKU code;
- Product details: core materials, usage scenarios, spec sheet, SKU table, product highlights;
- Product assets: 4 groups of images — main image, detail, specs, certifications.
The more complete the archive, the higher the system's completeness score, and the more stable the generated output. When you later create main image, detail page, or video tasks, you select the product archive at the top and it backfills automatically.
Even simpler: product link import. After installing a userscript in your personal center, open a Taobao/Tmall product page and click once — the product info syncs into the archive automatically.
3. Don't redo the whole set for dissatisfaction — section-level editing
This is the part closest to a real workflow, in my opinion. Generated results are organized by "sections" (floors), and each can be operated on individually:
- Preview / History: switch between V1, V2... and use whichever version you want;
- Regenerate: change resolution or model, only redo this section (10 credits per operation, no deduction on failure);
- Edit image: say "change the background to light wood, make the text one size bigger" — partial edit (10 credits per operation);
- Insert section: have the AI generate a new section, or upload a local image and insert it directly;
- Drag to reorder: hold a section card and drag; the long image and full set downloads stitch in the new order automatically;
- Delete section: delete a section you don't want.
This means you don't need to re-run the entire 12-section detail page just to fix a corner badge.
4. Virtual models keep the person's look consistent across the whole store
When a person needs to appear, you can upload a dedicated model image in "My Models":
- Create up to 20 models;
- Each model can have up to 12 reference images;
- In main image, video, and other tasks, select a model and generation maintains identity consistency.
For categories that strongly depend on a person appearing — apparel, home goods, beauty — this is far more practical than "randomly generating a nice face each time."
5. Video is not just a single clip
MantouAI's video pipeline is more complete than image generation:
- Choose video purpose (main image video / product promotion / grass-roots planting / plot-driven promotion);
- Choose target platform (Douyin, Xiaohongshu, Video Account, Kuaishou, Taobao, Tmall, JD, Pinduoduo, TikTok, etc.);
- Set voiceover language (Mandarin, Cantonese, English, Russian, Japanese, Korean, Vietnamese, Thai), on-screen person, duration, aspect ratio, resolution;
- AI recommends 3 video proposals;
- Generate storyboard script: timeline, visual content, shot size, camera movement, person actions, voiceover copy broken down shot by shot;
- Script can be manually edited, or have AI redo it;
- After confirming the script, generate the final video; continue chatting to modify if unsatisfied.
Video is billed per second, and the estimated cost is shown before confirming the script — you won't find out the price after it's done.
3. Real workflow test: main image
Running through a main image task step by step:
Step 1: New task, fill parameters
Upload at least 3 product images (max 10, each ≤5MB), fill in product name, selling points, specs, materials, choose the number of main images (1~8), aspect ratio, resolution (1K/2K/4K), model, language, platform type.
There are two execution modes:
- Proposal mode: the Agent gives proposals first, you confirm, then generate;
- Quick mode: skip proposal confirmation and generate directly.
Step 2: Chat with the Agent
After submitting, the Agent thinks for 20~40 seconds and may ask for more info. If there's nothing to add, reply "continue."
Step 3: Choose a proposal
3 proposals plus a free visual comparison card. Choose one and confirm. A popup clearly states how many credits this deduction will cost; confirmation charges.
Step 4: Wait for generation
A main image task defaults to 4 free generated images. The full set takes about 8 minutes. If generation fails, paid images are automatically refunded, and free images automatically retry for free.
Step 5: Edit and download
Section-level preview, regenerate, partial edit, insert, and drag-to-reorder are all supported. In the long-image preview area you can uncheck sections you don't want, then stitch with one click.
Downloads support PNG / JPG / WebP / PSD formats. PSD over 30000px automatically converts to PSB (large document); WebP is limited when the long image exceeds 16383px. These details are handled.
Unlocking a single image or the full set is required to download (10 credits per image, 10 credits per set). After unlocking, it's permanent, unlimited downloads, no watermark. If you've already unlocked individual sections, the full-set unlock won't charge twice.
There is also a very useful feature: login-free sharing. Click 【Share】 on the task detail page to generate a read-only link. Colleagues or clients can view the preview without logging in (unlocked sections show a watermark). The link works for 7 days and can be revoked anytime.
4. Price: get the math straight first
🎁 New users get 1000 free credits on registration, limited-time trial. At 50 credits per main image 1K generation, that's enough to run 20 main image tasks free — you can walk through the whole process before deciding whether to pay.
Credit packages:
| Package | Price | Credits | Bonus credits |
|---|---|---|---|
| Classic | ¥9688.00 | 120,000 | — |
| Premium | ¥18888.00 | 240,000 | — |
| Professional | ¥46666.00 | 600,000 | — |
Enterprise bulk purchases and long-term cooperation can be negotiated separately; the package is matched to actual usage.
For a 1K main image, 50 credits is about ¥4 per generation (varies slightly by package), and it includes 4 free generated images by default. Add partial edits and download unlocks, and the cost per main image set is still far below outsourcing. More importantly, the time cost: from submission to output is about 8 minutes instead of 3 days.
Note: prices and rules above are as of the time of writing; refer to the platform page for actual display.
5. Who should use it
- Taobao / Tmall / Pinduoduo store owners: no dedicated designer, but need to launch new products weekly;
- Design companies / studios: need to deliver main images, detail pages, and videos in bulk, using AI to compress the production cycle;
- Agency / e-commerce service providers: run multiple stores in parallel and use production capacity to take more orders;
- Brand owners: need unified visual style while controlling outsourcing costs;
- Cross-border sellers: need multi-language voiceovers and multi-platform aspect ratios, covering TikTok, Amazon, and similar scenarios.
It's also fair to state what it's not for: projects with pixel-level precision or complex compositing that have extremely high requirements will still need AI plus final manual adjustments in Photoshop. The platform's FAQ explicitly recommends combining with manual retouching for complex products.
6. Common questions
Q: What if the generated product details don't match the original image?
Upload clearer product images from multiple angles; emphasize "keep product details unchanged" in the design requirements; for complex products, fine-tune with PS at the end.
Q: How long does one set take?
Image tasks take about 8~10 minutes, replica tasks 5~8 minutes, videos depend on duration.
Q: Can multiple tasks run at the same time?
Parallel tasks are supported; the concurrency limit depends on the package.
Q: Do previews have watermarks?
Unlocked images carry a preview watermark; after unlocking individual images or the full set, downloaded files have no watermark.
Q: Is there API access?
Current focus is on product feature iteration; for custom needs, talk to the account manager.
7. How to try it
MantouAI currently uses invitation-based activation: contact the account manager to get an account and initial password. Registration gives 1000 free credits for a limited-time trial — you can run the main image, detail page, replica, and real-shot workflows fully without charging first.
If you're struggling with any of these:
- Every product launch means repeatedly hunting for a designer and waiting for scheduling;
- Main images and detail pages don't match stylistically;
- Want to replicate competitor visuals without starting from scratch;
- Short video outsourcing costs too much;
I suggest running a real product through a main image task. The 3 proposals are free to view, and if you don't confirm, you don't pay — that's the lowest-cost way to verify.
If this article was helpful, feel free to like and bookmark. Comment below about your biggest pain point in e-commerce visuals, and I'll add some tested parameter settings and pitfalls from real usage.