A store operator pastes a single flat product photo into a canvas, types "summer sale hero, soft daylight," and watches the background, copy, and model appear as separate, movable layers. No Photoshop, no designer on call. The tool is Gǎodìng (稿定设计), and the pitch is not "make a picture"—it is "finish the ad."
That reframing, from generation as novelty to generation as delivery, is where a lot of Chinese design tools quietly moved in 2025.
Key takeaways
- Gǎodìng (稿定设计) is a commercial design platform that relaunched as "Gǎodìng AI" on 19 September 2025 at Huawei Connect, built on Huawei Cloud's Token service.
- It reports more than 100 million individual users and 5 million enterprise accounts, mainly small and mid-sized businesses.
- Its core idea is layered generation: product, scene, and text are produced separately so each stays editable.
- A built-in AI design agent and an infinite canvas let users move from brief to finished asset in one workspace.
- Huawei Cloud optimization is reported to make text-to-image roughly 2x faster and text-to-video or image-to-video about 3.5x faster.
- It ranked second by web traffic among Chinese AI-creation tools in December 2025, behind only Dreamina.
From templates to agents
For years Gǎodìng (稿定设计) was known as a template library—thousands of layouts a non-designer could drop text into. The risk was obvious: templates scale poorly when every brand wants something distinct. The 2025 rebuild leans on generation instead of storage. According to China Daily's coverage of the Huawei Connect launch, the product covers inspiration gathering, creative stimulation, visual retouching, and design extension, aiming to take a brief from idea to deliverable.
The chief product officer, Mi Qipei, described the problem the team targeted: traditional design is slow, while naive generation is uncontrollable and needs repeated re-rolls plus manual fixes. Gǎodìng's answer was to grade content—product, scene image, marketing image—so each layer is generated on its own and stays adjustable.
How layered generation works
The workflow splits a marketing image into small pieces:
- Background — generated from the product alone, matching a chosen scene.
- Scene image — composed from background plus lighting and style cues.
- Marketing image — text and decoration added on top, still editable.
Because the layers are separate, a user can swap the headline without re-rendering the background, or change the season without reshooting the product. That is the practical difference between a one-shot image and a usable design file.
The agent and the canvas
Two features carry the experience. The first is an AI design agent (AI 设计 Agent) that, in the company's description, behaves like a colleague: it collects references, retrieves information, drafts the concept, and calls tools to produce the asset, all inside one conversation. The second is an infinite canvas where everything lives in one place, matching how creative people actually think.
Behind both sits Huawei Cloud's Token service, which supplies compute, model services, and development tooling. Gǎodìng (稿定设计) states the setup roughly doubled text-to-image speed and tripled multimodal video speed after optimization, though those figures are vendor-reported.
Scale and ranking
Gǎodìng's reach is its strongest claim. China Daily and the AICPB AI product list both cite more than 100 million individual users and 5 million enterprises served. In the December 2025 AI-creation web traffic report from QbitAI's index, Gǎodìng AI ranked second by monthly visits (above 3 million) among web-based creation products, behind Dreamina and ahead of LibLibAI and Canva's local edition.
That placement matters because it shows generation tools are being measured by usage, not press releases. A second-place finish in a contested category is a more honest signal than a funding headline.
Where it helps media teams
The clear fit is high-volume, low-differentiation output: e-commerce main images, detail pages, multi-language variants, and seasonal posters. For a cross-border seller running dozens of SKUs across markets, the value is throughput and consistency, not artistic surprise.
Why the workflow matters
The move from fixed templates to layered generation reflects a broader change in how Chinese small businesses approach design. Instead of hiring an agency or learning heavy desktop software, a store owner now expects to describe an outcome and receive an editable file within the same session. Layered output keeps the result a working asset rather than a one-off picture: when a promotion ends, the operator can keep the product layer and regenerate only the surrounding scene, which suits the fast, repetitive cadence of e-commerce better than one-shot generation.
Honest limitations
The layered approach is genuinely useful but bounded. Quality still depends on the underlying model's grasp of typography, brand rules, and local taste; outputs can look generic when prompts are thin. Speed claims are vendor-stated and tied to a specific cloud stack, so they may not hold on other infrastructure. Enterprise governance—asset rights, approval flows, and integration with existing design systems—is lighter than what dedicated creative suites offer. And like most generation tools, Gǎodìng (稿定设计) works best as a first draft engine, not a replacement for senior design review.
What readers can do now
- If you run e-commerce creative, test a single SKU through Gǎodìng's layered flow and compare the editable output against your current template process.
- Ask vendors for a reproducible speed benchmark on your own hardware before treating the 2x or 3.5x claims as fixed numbers.