AI design feedback that measures before it judges
No reviewer free tonight, and chatbots call everything clean. Upload one screen: code measures its contrast exactly, then an AI names up to three fixes, each tied to a principle.
- Free, no signup
- Seconds, not days
- Never stored
PNG, JPG or WebP at its real size, 1x or 2x. Or paste with Ctrl+V. A mockup of several screens shrunk into device frames gets worse feedback.
Your screen goes to an AI model to write the points. ux360.design does not keep it, and only hosts that do not train on it can run the model. Please do not upload screens that show real people's data.
Your critique
Measured by code
Exact, from the pixels, not the AI. Letters match the coral highlights on your screen. For every line's ratio, open the contrast checker.
Worth a look
Written by an AI and graded by a designer before launch: about 6 in 10 of its points held up, so read them as a second opinion. Numbers match the yellow highlights.
Working well
What is an AI design critique?
An AI design critique reviews a screenshot of your interface and tells you what to fix. This one does it in two passes. First, code in your browser measures what can be measured: the contrast of every line of text, whether a heading is outranked by text below it, line spacing and line length. Then an AI model, GLM 5.3 Flash, is given those results as settled facts and asked only for what code cannot judge.
Why not just paste your screen into a chatbot?
| What you need | A general chatbot | This critique |
|---|---|---|
| Contrast and text size | Estimated from pixels, and often wrong on a small or shrunk screen | Measured line by line, with the ratio and the nearest passing color |
| A short list to act on | Ten polite points, all weighted the same | Three at most, in order: the positions that graded most reliable |
| Where the problem is | Described in words | Marked on your screen |
| Why it matters | Rarely named | Each point tied to the principle behind it, with a fix |
| Cost | An account, often a subscription | Free, no signup |
How it works
- Upload one screen. Drop, paste or choose a PNG, JPG or WebP.
- Code measures it. OCR finds every line of text on your device and measures its contrast, size and spacing. Anything below the WCAG minimum is marked on your screen with a darker or lighter color that would pass.
- The AI reviews what code cannot. Whether the main action gets the most weight, whether things look tappable, how items are grouped and lined up, leftover placeholder copy, meaning carried by color alone, and too many typefaces.
- You get three points at most. Each is placed on your screen, says why it matters to the person using the design, and links to the principle behind it: emphasis, affordance, proximity and more.
How far should you trust it?
Before this page went up, the AI was tested on real student screens and on designs with planted flaws, and a designer graded every point it made. On single screens about 6 in 10 of its points were right, and the first two it listed were right 7 times in 10, which is why it shows three at most. It is weakest on spacing and grouping, and now and then calls a deliberate crop a flaw. The measured checks do not guess: they caught every contrast failure on the same screens.
Use it as a second opinion that arrives in seconds, before you ask a person. A critique from someone who knows the brief and the people using the design will always see more.
Questions
Is the AI design critique free?
Yes. There is no account, no signup and no limit per person. It runs on a small daily budget, so on a very busy day the AI points can pause until midnight UTC. The measured checks never pause, because they run in your browser.
What should I upload?
One screen at its real size: a PNG, JPG or WebP export of a single phone, tablet or desktop screen. A mockup that shows several screens shrunk inside device frames gets worse feedback, because the text is too small to measure and the AI starts judging the frame instead of the design.
How is this different from asking ChatGPT or Claude for design feedback?
A chatbot estimates contrast and text size from pixels, and in our tests models called text faint when a screen was only shrunk. Here, code measures those exactly first and the AI is told the results are settled. It is then limited to what needs judgement, returns three points at most, and ties each one to the design principle behind it, with a fix.
Does it check accessibility?
Partly. It measures the contrast of every line of text against WCAG 2.2 success criterion 1.4.3 and suggests a passing color for each failure, and the AI looks for meaning carried by color alone. It does not test keyboard use, screen readers or focus order, so it is not an accessibility audit.
Does it store my design?
No. The text reading and the contrast measurement happen on your device. For the AI points the screen is sent, shrunk to 1568px on its long side, through OpenRouter to Parasail, which runs the model, under settings that allow only hosts that do not train on what they receive. ux360.design keeps nothing.
Which design principles does it check?
Emphasis and hierarchy (does the main action get the most weight), affordance (do tappable things look tappable), proximity and alignment (are related things grouped and lined up), repetition (do repeated parts match), color (is meaning carried by color alone), unity (too many typefaces), plus leftover placeholder copy.
Read more on visual hierarchy and designing for accessibility, or see how every principle fits together in the fundamentals, in order. The privacy policy covers the tools in full.