this guy got tired of going back and forth with Claude Code describing what's wrong with the UI so he built a tool that lets you DRAW DIRECTLY on the screen claude renders something. if it's wrong you just circle the problem, add an arrow, write a note ai... sees your annotations and fixes it way faster than describing it by text AND it's free + open sourceshow more

already exist

You can literally just take a print screen snippet and paste it straight into Claude code. This is just that with extra steps.

what’s the name of the library? all these AI comments and no one bothers to ask this?

this is exactly how we built lock in's discipline feedback - traders can't describe their tilt patterns in words but they CAN mark up their chart and say 'here's where i started revenge trading'. visual annotation > text logs every single time

hey folks! i'm the one in the demo and one of the builders of this tool. it's at i'm starting @wenldev to share the journey. do check out our work, would love to hear what folks think.

describing UI bugs in text is the worst part of working with AI code tools. pointing at the screen and saying "this is wrong here" is how every dev already communicates with humans so this makes total sense as an interface.

link?

screenshot -> draw -> ctrlC -> ctrlV

ok ok ok, this seems rad

this is brilliant. spent way too many rounds describing ui alignment issues to claude code for my client dashboards. 'move the button left' becomes a 20 message thread. visual markup would cut that feedback loop to seconds

that's genius, no more vague descriptions!!

honestly this is better than another paragraph of "make it a bit more aligned"

the number of times i've typed "no the other button" "yes that one but like 3 pixels to the left" only to realize i could just... point at it. this is unhinged in the best way

@llmgram

oh man yeah describing pixel-level UI stuff in chat is miserable lol. been there. "no the padding by the OTHER button" kills me every time

this is exactly...

Just use the preview point and click to the elements you want to change (like cursor) simples

Try Snip here! Linked our github repo too:

thanks for featuring SNIP, we are rolling out snip browser soon!

I do the same for rendering animations except it’s a tool to pick an x/y coord There was a “Tron ISO” moment when openclaw re-built the tool for an animation with multiple coords. Didn’t realize what it did until a few copy/paste of the coords “oh it just does multiple now?!”

whats the link

We've come full circle. First we drew mockups. Then we described them in words. Now we're back to drawing on screens like cavemen because apparently that's faster than typing "move the button 3 pixels left" 47 times.

blind prompting failed at pixel math so this mcp routed natively dom loads. if a div snaps devs just highlight the bug the daemon ships the patch plus it's raw + local

Nice, basically what I've been doing with screenshots for few years. I take a screenshot and name it like "this-needs-to-be-bigger-and-green.jpg" 😅

A feature in Google AI studio.

we went from "move the button 3px to the left, no the other left" to literally drawing on screen like a coach on a whiteboard. this is the most human interface for AI coding yet

that's super useful the best way to give context for a problem

Banger. no more 'move it 2px left' dissertations.

the verbal-description loop for ui fixes is genuinely painful - you describe the problem, it misses, you describe again. visual annotation shortcuts that whole cycle. this is the kind of tooling that makes the difference between using claude code occasionally vs daily

drawing directly on the screen to show claude whats wrong is SO much better than trying to describe UI bugs in words. "the button is 3px too far left" vs just circling it with an arrow. visual communication > text communication for spatial problems. surprised this took so long to build honestly

Just use Claude desktop?

The execution speed right now is wild. What took teams months last year takes one person a weekend with the right AI setup. The bottleneck is no longer building, it is deciding what to build.

describing UI bugs in text is a crime we've all been committing. this should've existed two years ago.