For v0 apps
Your v0 app looks great.
Does it still look great tomorrow?
v0 generates beautiful UI in seconds. Lastest locks it down with deterministic visual regression tests - so the next prompt doesn't quietly break the last one. See why v0 authentication keeps silently breaking.
Why Not Just Ask the AI?
You can ask v0 to test what it built. Here's why that breaks.
v0 is great at generating UI. It's not great at guarding UI. Every new prompt regenerates markup, shuffles Tailwind classes, and quietly nudges layouts. Two approaches - and which one actually holds.
Ask the AI to test your v0 app.
The LLM runs the flow in a chat, declares it looks fine, and moves on. Different path every run. No artifact, no history, no screenshot diff. And every single re-run burns tokens - on top of the tokens v0 already cost you.
- Non-deterministic - different selectors, different results each run
- Tokens compound - every preview deploy is another bill
- No visual layer - can't see the 4px shift that broke the hero
- The model grades its own homework
Use Lastest. Generate once. Replay forever.
Point Lastest at your Vercel preview URL, click through the flow once, hit stop. AI generates deterministic Playwright code with multi-selector fallback, captures a visual baseline, and replays the exact same thing on every push - zero AI calls, zero token cost.
- Deterministic replays - same code, same screenshots, every run
- Zero tokens after generation - run thousands of times for $0
- 3 visual diff engines - Pixelmatch, SSIM, Butteraugli
- Self-healing selectors - 7-layer fallback survives v0 re-renders
- Wires into Vercel preview deploys - test every PR automatically
Built for v0 Workflows
Prompt the UI. Guard the UI.
Two features that matter most when you're shipping v0 apps to Vercel on every prompt.
01 · Record
Click through your v0 app. That's the test.
Open the recorder, paste your Vercel preview URL, walk through the flow you care about, hit stop. Lastest captures every click, type, and navigation and generates deterministic Playwright code - no selectors to maintain, no Playwright docs to read.
- Works with any Vercel preview or production URL
- Captures multi-step screenshots per test run
- No access to your v0 source required - tests the running app
02 · Visual diff
Did the last v0 prompt quietly break the layout?
Every v0 prompt generates new markup. New class names. Different wrappers. DOM tests pass, and the hero section is still 12px off. Lastest compares screenshots with three diff engines (Pixelmatch, SSIM, Butteraugli) so visual drift shows up where you can actually see it - before it hits production.
- Text-region-aware diffs reduce false positives from dynamic content
- Branch-to-branch compare for Vercel preview PR review
- Approve or reject each change with a one-click slider
Common questions for v0 apps
- Why do v0 prompts keep silently breaking my UI?
- v0 regenerates whole route trees on each prompt. Tailwind classes shift, IDs rename, layouts nudge. The page still renders, but a button now lives in a different place or a card no longer matches the design. Lastest captures a visual baseline; the next prompt either matches it or surfaces a screenshot diff for review.
- Can I just ask v0 to test what it built?
- You can ask. The LLM clicks a happy path once, declares success, and moves on. Different path next run, no artifact, no visual baseline. Tokens compound. Lastest records the flow once, replays the same flow forever, and grades the result against a fixed baseline you reviewed.
- Will Lastest work with Vercel preview deploys?
- Yes. Point the recorder at the preview URL on PR open, fail the PR on a non-trivial diff. The GitHub Action
las-team/lastest/action@mainwires this in with one step. Approved baselines replay deterministically with $0 token cost forever. - What happens when v0 renames every selector?
- The recorder captures seven selector layers (data-testid, id, role, aria-label, visible text, CSS, and OCR image search). When v0 swaps Tailwind classes or wrappers, the selector still finds the element. When the visible UI shifts, the visual diff catches it.
- How is this different from Chromatic or Percy for v0?
- Chromatic needs Storybook (v0 doesn’t use one). Percy and Applitools charge per screenshot and ship your UI to their servers. Lastest tests the running v0 app at any URL, ships three diff engines, and is free, open source, and self-hosted.
- What does it cost?
- Zero. Lastest is open source (FSL-1.1-ALv2) and self-hosted via Docker. Generate the regression suite once, replay on every push for $0. No per-screenshot, per-seat, or token cost on replay.
Add visual tests
to your v0 app.
In five minutes.
No cloud account. No API key. No procurement. Point it at your Vercel preview URL and go. Open source, FSL-1.1-ALv2 licensed, $0 forever.
★ Star & Install from GitHub →# 1. clone
$ git clone https://github.com/las-team/lastest
# 2. start
$ cd lastest && docker-compose up -d
# 3. point at your v0 preview URL
$ open http://localhost:3000
✓ View the source