About GenDesigns

What GenDesigns is

You describe a screen, or a whole app, the way you would describe it to a colleague. GenDesigns picks a visual theme, generates each screen as HTML and Tailwind CSS, and lays them out on a canvas you keep editing by talking to it. Ask for a darker theme, a different layout, one more screen, and it reworks what is already there.

The output is markup, not a flat picture. That matters when the design has to leave the tool: export a single screen as standalone HTML or the whole project as a ZIP, and a developer can open it and read it. The people who get the most out of it are usually the ones without a design tool already open — founders putting a deck together, product managers sketching a flow, engineers who need something to build against.

More detail on how the generator works, and what it costs on the pricing page.

Who builds GenDesigns

GenDesigns is built and maintained by the GenDesigns Team. The same people who ship the product write the benchmarks, the guides, and the comparison pages on this site — there is no separate content operation, and nothing here is outsourced to writers who have never used the tool.

That cuts both ways, and it is the reason the rest of this page exists. Research written by the people selling the product is only worth reading if the method is visible and the incentives are stated out loud. So here they are.

How we test things

Most "we tested the AI models" posts are one person generating one output per tool and writing up whichever one impressed them. That is a screenshot collection, not a test. LLM output varies enough between runs that a single generation tells you almost nothing. Our method is built around that fact:

• Identical prompts, identical instructions. Every model or tool in a comparison gets the same prompt and the same system instructions. No per-model prompt tuning, because tuning a prompt to a favorite is how you get the result you wanted before you started.
• Multiple runs, then average. Each prompt runs several times per model and the scores are averaged. In our 2026 LLM comparison that meant 10 prompts across 5 models, 3 runs each — 150 generations — which is also what let us measure run-to-run consistency as a criterion in its own right.
• Fixed viewing conditions. Every output is rendered in a browser at the same width, 390px for mobile work, so nothing wins by being judged on a roomier screen than its competitors.
• Criteria written down before the results come in. Visual quality, layout logic, code quality, component accuracy, consistency — each weighted, each with a stated definition, all fixed before scoring starts.
• Independent scoring by more than one person. Outputs are scored separately by a designer, a developer, and a product manager, and the scores are averaged. Three people disagreeing is signal; one person deciding is taste.
• The bad results get published too. Every model in the benchmark has its worst output named and described alongside its best, including the models we like.

The full method, the ten prompts verbatim, and the per-criterion scores are all in the LLM UI generation benchmark. If you want to check our work, you have everything you need to run it yourself.

How we write about our own product

We sell an AI design tool and we publish pages comparing AI design tools. Those two facts sit uncomfortably together, so rather than claim a neutrality nobody would believe, here is what we hold ourselves to:

• We say when GenDesigns is in the list. If our product appears in a comparison, it is labeled as ours. You should read our entry with exactly the skepticism that deserves.
• Competitors are described by what they are good at. Including the cases where the honest answer is that another tool is the better fit. A comparison where we win every row is a comparison nobody should trust.
• We link out to the tools we compare against. Go and look at them.
• Prices and features go stale. Anything we quote was accurate when the page was written or last updated. Vendors change plans without telling us; check theirs before you decide.
• Corrections get made, not argued about. If we have described your product wrongly, email us and we will fix the page.

Contact us

Questions about the method, corrections, press, partnerships, or product support all go to the same inbox — a real person reads it.

Email: hello@gendesigns.ai
On X: @gen_designs_ai