Skip to main content

UI review · screenshot bug hunting

Hand UI screenshots to a visual understanding model to pick issues across four dimensions and score them; page errors and images that fail to load can also be located with a screenshot plus code.

Disclaimer: The content shown on this page is generated by an AI model and is for reference only. It does not constitute any guarantee of results. Actual output and Credits consumption are subject to "Usage Details" in the console.
After staring at your own pages long enough, you go numb — problems are right there and you don't see them. Hand a screenshot to the model and have it pick issues across color / font / hierarchy / background and give a score — it's more sensitive to "the overall feel is off" than the human eye; page errors and images that fail to load can also be located by giving it a screenshot along with the code. This walkthrough runs a review on the first version and the revision of a "Parents' Medication Manager" page. Call chain: qwen3.7-plus (visual understanding) → feed the UI screenshot + a four-dimension review checklist (color / font / hierarchy / background) → per-item judgment + priority improvements + score.

Actual output

Review run on the first version and the revision of the same "Parents' Medication Manager" page. Submitted materials:

First version (dot-grid background + dark green title bar + multi-color time-slot bar)

UI first version

Revision (single accent color + solid background)

UI revision

Steps: Four-dimension UI review

Send the UI screenshot along with the instruction below to qwen3.7-plus:
You are a UI/UX designer. This is a screenshot of a "Parents' Medication Manager" web app
I built. Please give me a review report from the perspective of aesthetics and usability.

Please check item by item along the four dimensions below. For each, give a judgment of
[OK] / [Needs improvement] / [Has a problem] and point out the specific location:
1. Color: How many colors are used overall? Is the palette restrained and unified? Is the
   color semantics correct (e.g., is a warning color used where a warning is needed)?
2. Font and layout: Is the font consistent? Are there misalignments, text being squeezed,
   or ugly line wraps?
3. Visual hierarchy: What does the eye land on first when entering the page? Is the
   primary vs. secondary clear? Are multiple elements competing for attention?
4. Background and decoration: Is the background clean? Are there redundant textures,
   decorations, or gradients that interfere with reading?

Finally, provide:
- An overall score (1-10)
- The top 3 priority improvements sorted by "impact × fix cost"
- A one-sentence summary of the impression this UI gives

Be specific and direct, no pleasantries.
Overall score: 6.5 / 10

Priority improvements (sorted by impact × fix cost), in the model's own words:

1. Remove the background grid (fix cost very low, impact very high)
   "The page uses a background with a prominent dot grid. This kind of background is
     usually for design mockup or note-taking apps, and it looks very out of place in a
     'medication management' tool for middle-aged and elderly users. It not only makes
     the interface look unprofessional and un-serious, but also adds visual noise that
     interferes with the core information."

2. The yellow safety alert box at the top is too eye-catching
   "It is even more prominent than the '0/4 taken today' progress" — stealing the show.

3. The left time-slot bar (orange/purple/blue)
   "Slightly oversaturated, a bit harsh," competing with the core information for
   attention.

It also pointed out: the privacy text "font size too small and color too light," "contrast
too low, easy to be completely ignored."

One-sentence impression: like a "design mockup" rather than a "serious tool for middle-aged
and elderly users."

Notes

  • Consumption: This playbook involves visual understanding (qwen3.7-plus) single-image review. The screenshot itself is converted into image_tokens and counted toward input, taking up the vast majority of the input side; the text prompt is only a small fraction. The larger and clearer the screenshot, the higher the input consumption — a normal webpage screenshot size is enough for UI review; no need to use oversized images.
  • How to use its output: It's great as a bug-hunting lead — an out-of-place background, a scene-stealing alert box, or harsh colors are exactly the kinds of problems you go numb to after staring too long, and it picks them out at a glance.
  • But don't treat it as an acceptance report: In testing, it can misread details that require precise counting (e.g., calling seven or eight colors "within 5," or missing a serif font), and it may misread the meaning of data on the page. Still confirm each suggestion against the actual UI.
  • A cheaper approach: Write the color, font, and hierarchy requirements into the prompt up front (see the "Parents' Medication Manager" card) to save a lot of rework.
Model Playground
Fine-tuning
Model Compression
Statistics and Monitoring
Support