A single tool gives you one kind of image at a time, but what you usually need is a whole set. This article is about chaining them together on the canvas. To get there: Creative Studio → 🎨 Canvas.
Canvas plus the Design Agent
It's an infinite surface — scroll to zoom, drag empty space to pan. You can drop in uploads, word art, layout templates and past assets, and output from other tools can be pushed straight onto it.
Describe what you want to the 🤖 Design Agent on the right and it returns an execution plan first; nothing runs and no credits are spent until you press ▶. Typing @ lets you reference images from Assets, products from your product library, or existing skills, and you can attach your shop's brand style. Models are routed automatically in the background. Select any image and in-place actions appear — repaint, erase, outpaint, remove background, edit, annotated edit, convert to video — while 🧩 Split layers decomposes a finished image into background, subject and editable text layers. Only use it on images you own or are licensed to use.
36 official skill cards
The Skills tab holds 36 official cards across four categories — e-commerce, social, brand and video: compliant white-background hero images, feature detail images, the Amazon five-image set, Xiaohongshu seeding covers and more. Tap a card and it builds a plan from that recipe; cards that need a product photo will ask you to upload one first. A plan that worked well can be saved with ⭐ Save as skill and replayed as-is, or recommended to your team.
Workflows: one sentence, the whole chain
The Workflows tab holds official multi-step pipelines — nine of them today, each card labeled with its step count and estimated credits:
- Full listing set from one photo (25 steps): compile the product facts → retouch a white-background hero as the anchor for the whole chain → 2 more main images → generate all 8 slots of the detail page one by one.
- Product film from one photo (31 steps): a 15-second, 9-shot breakdown → a nine-panel storyboard → a still per shot → a clip per shot → stitched in shot order (stitching is free).
- Plus chains for apparel launches, Amazon A+ modules, beauty and personal care detail sets, poster text swaps and more.
Use this to recreate produces a plan directly; Open in node editor loads it as a node graph you can adjust step by step. Steps that carry your original photo get the fidelity constraint injected; in rare cases the model still drifts, so just re-run, and failed steps release their credits. See product fidelity.
Node editor: wire up your own chain
Enter it from Workflow Mode in the canvas top bar. There are three source nodes (upload image, pull from canvas, product info) and nine step types: copy generation, AI image, poster image, model try-on, video generation, clip stitching, background swap, image edit and cutout. Drag from an output dot to the next node's input to connect them; cycles and type mismatches are rejected. Downstream prompts can reference upstream output with [image N]. Select a node, press ▶, and a confirmation card lists the charge for each step and the total — nothing runs until you confirm. Copy generation and clip stitching are free, and a failed step only re-runs that step.
Exporting
You can export PNG (video nodes are excluded), a layered PSD (text stays editable, so you can keep working in Photoshop), or the artboard cropped to exact dimensions. One known rough edge: some older cross-origin images may not embed into the exported file. When you see the skip warning, re-add those few images with upload image and export again. Canvases can be saved to the cloud and picked back up under My canvas.