How to generate redesign candidates from current screens for UX researchers

Steps for UX researchers to turn current screens into redesign candidates, compare variations, and check outputs before sharing findings.

By 2Dex · 2026-10-18

2Dex is published by the makers of SageThumbs 2K, RēDesign and VectorMagik, which are compared here alongside other tools.

Start with current screens, define what each redesign should explore, generate several distinct candidates, then compare them against the original screen. RēDesign can send one UI screenshot to several AI models at the same time and show the returned HTML redesigns in a gallery next to the original.^[1]

Option	Candidate generation	Comparison and review	Cost
RēDesign	Sends one UI screenshot to several AI models at the same time and shows the returned HTML redesigns beside the original.^[1]	Shows two outputs with model names hidden and updates a model ranking from each pick.^[1]	RēDesign is free; the AI providers whose API keys are used are paid separately.^[1]
Other redesign tools	not published	not published	not published

What you need before you start

Generating redesign candidates from current screens takes a little preparation. Work through this list first.

The current screens. Gather the screenshots that represent the experience you want to explore. Keep the original versions available so every candidate can be compared with the starting point.
The research question. Write down what the redesign should help investigate, such as hierarchy, navigation, content emphasis, or visual direction.
The important constraints. Note requirements that a candidate should preserve, including key content, brand elements, interaction goals, or accessibility considerations.
A comparison method. Decide how candidates will be reviewed before generating them. A consistent set of questions makes differences easier to discuss.
A place to record findings. Keep candidate names, observations, and follow-up questions together so promising directions can be revisited.
A review step. Plan to inspect each output against the original screen rather than treating a generated candidate as a finished design.

Step by step

These steps work the same way whichever redesign tool you pick.

Choose the current screen. Start with a screen that has a clear research question behind it. Keep the original screenshot unchanged so it remains the reference point.
Describe what should be explored. Turn the research question into a short instruction. Say what should change and what should remain recognizable.
Ask for distinct directions. Create candidates that explore different answers to the same question. One direction might preserve the existing structure while another changes hierarchy or emphasis.
Keep the comparison fair. Use the same source screen and research goal when comparing candidates. Change one major design idea at a time when the research question calls for a focused comparison.
Generate the candidates. Let the chosen tool produce the variations. If it supports several models or multiple variants, decide how many directions are useful before starting the run.
Review each result beside the original. Look for changes in hierarchy, content emphasis, navigation, spacing, and other areas connected to the research question. Record observations rather than judging the candidate only by appearance.
Check accessibility concerns. Review text contrast and other requirements that matter to the study. A visually different candidate still needs to be considered against the constraints defined at the start.
Compare candidates without bias where possible. If the tool can hide model names during comparison, use that mode when model identity could affect the judgment.
Save the useful directions. Keep the candidates that answer the research question or raise a useful new question. Record why each one is worth carrying forward.
Turn findings into the next design step. Use the strongest observations to guide another round, a prototype, or a discussion with the design team.

Tools that can do it

Facts are published here for only one tool, so the comparison is short. For any other option, check its own page for the current screen formats it accepts, how it generates candidates, how it compares outputs, what it exports, and what it costs.

### RēDesign

RēDesign is made by the team that publishes this site. It sends one UI screenshot to several AI models at the same time and shows the HTML redesigns they return in a gallery next to the original.^[1] In one run you can ask each model for a different number of variants.^[1] It comes with prompt presets such as faithful refresh, bold reimagine and minimalist, and you can also write your own prompts.^[1]

For comparison work, it shows you two outputs with the model names hidden, and each pick you make updates a ranking of the models.^[1] You can download a whole run as a zip, or download a DESIGN.md file for the redesign you picked so a coding agent can rebuild that look.^[1] With one click it checks an output's text contrast against WCAG AA, using the rendered pixels.^[1]

It shows an estimated cost before you start a run and tracks what each run costs.^[1] It works out which provider each pasted API key belongs to and rotates through several keys per provider, pausing keys that keep failing.^[1] It checks each saved redesign on your machine against a fixed list of generic AI styling patterns and shows the findings as a badge on the card.^[1] It locates and crops logos, product photos, avatars and illustrations so redesigns can embed the real brand marks.^[1]

Everything in the app can also be done from the command line, and an AI agent can control it too.^[1] You can line up several batches, start them together and drag waiting batches to change their order.^[1] RēDesign is free; you only pay the AI providers whose API keys you use.^[1] The packaged app is for Windows only; running from source works on Windows, macOS or Linux with Bun 1.2 or newer.^[1] On Windows you download redesign-windows-x64.exe from GitHub Releases.^[1]

Common mistakes to avoid

Starting without a research question. A visually different screen is easier to discuss when the reason for generating it is clear.
Changing too many things at once. If several design ideas change together, it becomes harder to connect a finding to a specific direction.
Replacing the original screen. Keep the current screen available so every candidate has the same reference point.
Treating generated HTML as the final design. Use candidates to explore directions, then review the underlying design decisions against the research goal.
Ignoring accessibility during comparison. Check important accessibility requirements as candidates are reviewed, not only after one direction has already been chosen.
Letting model identity influence the comparison. When a tool supports hidden model names, compare the outputs before looking at which model produced each one.
Forgetting the cost before a large run. Check the estimated cost before generating candidates and review the recorded cost afterward when the tool provides those details.

Frequently asked questions

Can RēDesign generate several redesign candidates from one screen?

Yes. RēDesign sends one UI screenshot to several AI models at the same time and shows the HTML redesigns they return in a gallery next to the original.^[1] In one run, each model can also be asked for a different number of variants.^[1] This lets a researcher compare several directions while keeping the same current screen as the starting point.

Can RēDesign compare redesigns without showing model names?

RēDesign shows two outputs with the model names hidden, and each pick updates a ranking of the models.^[1] This gives a researcher a way to compare outputs before seeing which model produced them. The comparison can therefore focus on the candidate shown on screen rather than on a model name known in advance.

Can RēDesign use different prompts for redesign research?

RēDesign comes with prompt presets such as faithful refresh, bold reimagine and minimalist, and it also lets you write your own prompts.^[1] A researcher can therefore frame a run around the question being explored, while keeping the current screen as the source for the redesign candidates.

Can RēDesign check accessibility in redesign candidates?

RēDesign can check an output's text contrast against WCAG AA with one click, using the rendered pixels.^[1] This can be used as one review step after candidates are generated. Other accessibility requirements should still be checked separately when they matter to the research question or the design constraints.

How much does RēDesign cost to use for redesign candidates?

RēDesign is free; the only payments are to the AI providers whose API keys are used.^[1] RēDesign also shows an estimated cost before a run starts and tracks what each run costs.^[1] Check those estimates before generating candidates when the research process involves several models or batches.

RēDesign is free; you only pay the AI providers whose API keys you use. ^[1]

Try RēDesign

Sources

RēDesign's own description of itself (the product of this site's publisher)