CreaTrades.
From a product brief to a creative workflow.
Co-founder & backend lead / 2025 — ongoing
Smaller sellers need more than a generated image: they need a repeatable route from product references to usable assets.
Co-founder and backend lead. I connect task execution, model routing, saved workflows, assets and credit quotation.
120+ businesses served; 100,000+ generations. Company-reported, October 2026; not all businesses are paying customers.
Keep the lantern. Change the story.
A white-background reference tells us what the object is. An outdoor scene helps a seller communicate where it belongs. The work between them is about constraints as much as appearance.
Both images are from page 18 of the supplied CreaTrades brochure. This case explains the product workflow around that pair; it is not a record of one customer session. Team imagery; my contribution is product and backend engineering.
- 01
The brief
Use a recognisable product in a camping context. Separate the object’s fixed features from the scene choices.
- 02
The direction
Keep the reference available while exploring lighting, background and composition.
- 03
The review
Compare the handle, glass enclosure and silhouette against the input; review the scene separately.
- 04
The handoff
Save outputs and workflow context so the next revision can continue from an existing asset.
The work around the model.
- 01↓
Creative workspace
Brief, references, canvas & result review
- 02↓
Product & backend services
Task state · workflow persistence · routing · credit quotation
- 03↓
Queued generation
RabbitMQ → workers → external model providers
- 04
Results that can be reused
MySQL metadata · Redis state/cache · object storage
A generation request becomes a recoverable task.
I lead backend development and product direction. My responsibilities include task execution, provider routing, workflow persistence, asset/history services and credit quotation, with some frontend implementation.
Let the task outlive the request.
The architecture separates submission from queued generation. This adds task-state and result-storage work, but lets the product retain progress and outputs beyond one browser request. My backend work makes those states available to the workflow and asset interfaces.
The trade-off is more lifecycle coordination. A delayed response must remain distinguishable from a finished task; an output needs a stored reference before it can be reused.
Start with the seller’s work.
During business visits in 2025, I encountered product-image sets costing around RMB 1,000, lengthy production cycles and repeated revisions. That specific encounter gave us a problem to build around.
I expected image models to become more capable and affordable before video reached a similar price level. I chose to invest in the surrounding workflow and a product for sellers: reusable assets, team use and marketplace requirements.
CreaTrades has paying users and has been profitable after operating costs since August 2026. I lead its approved national entrepreneurship-practice project, with a CNY 100,000 allocation. Our team received Second Prize at GAISIC 2026.
Operating status and allocation are applicant-confirmed; the allocation is not a statement of cash received. Customer visits, training and follow-up are team activities.
Visit CreaTrades ↗From a brief to the next iteration.
Start with the product
Keep the product reference and the intended use in view.
An illustrated walkthrough using product screenshots and company-brochure examples. These images explain the workflow; they are not a single recorded generation session.
Watch the 60-second captioned film ▷
Download the film ↓Different products, different decisions.
Company/team examples from the supplied brochure. I develop the product and engineering that support this work.