Thread 107117843

What have you vibe coded lately, /g/oys?


Resources:

Websites:
https://aistudio.google.com/ (Very generous limits, biggest context size)
https://claude.ai/
https://chatgpt.com/
https://chat.z.ai/
https://grok.com/

CLI tools:
https://github.com/google-gemini/gemini-cli (1K requests/day if you connect your Google account, not always the case)
https://github.com/QwenLM/qwen-code (fork of Gemini-CLI, allows you to use any OpenAI compatible provider)
https://github.com/anthropics/claude-code (like Qwen code it is compatible with OpenAI endpoints by setting environment variables)

>>107117843 (OP)
how experienced do you have to be to actually produce something of value vibecoding? I did a python tutorial one weekend.

>>107117872
Of value to whom? You? Not at all, really

I've been using Cline w/ a local model (gpt oss 120 right now). I chose cline because I liked the diff-based view and I edit the diffs a lot. How much better are those other options?

>>107117881
like can I do the following
>ask it to write a phone app I have an idea for
>if there's an error or something, just tell it "this doesn't work, I got this error"
>repeat until app works

I tried to do some project planning for a 6 week project and it consumed all my tokens and shat out garbage, I'm conserving my usage for some workshops I'm attending

Maybe I'll read what it came up with, change some stuff, and re-run it

>>107117872
I spent 3 days working on a small application connecting to an API for a CCTV stream as a take home assessment for a job, it was fun, I did some of the development myself and otherwise iterated through the project slowly and ended up with something pretty decent

It suggested some nice QoL features that I never would have bothered with if I were just working on it alone

The more you know, the better, you still actually have to learn how to develop software and use your tools or you won't know when LLM fucks up. Toward the end of the project it was getting a bit unwieldy

After that, I started to look into what LLMs can and can't do, how to prompt effectively, I just installed some MCP servers for that project which effectively let it search documentation

>>107117913
More or less, yes. It obviously helps having technical knowledge and telling it the exact errors you get instead of "it doesn't work"
>>107117893
Are you running the full thing? The models in the OP are most likely better since they are state of the art, though depending on how complicated whatever it is you're doing you may or may not notice a big difference

Hello

>>107117843 (OP)
cute girls!

Opencode is also really nice, supports lots of providers by default as well as custom ones
https://opencode.ai/

>>107118226
Cute anon! :3

greetings, vibe coders! let our code monkeying skills further decline and our executive functions grow stronger!

>>107117843 (OP)
i have made this yesterday, with gemini 2.5 pro, my bestie
https://blogchan.pages.dev/g/thread/33333
currently supports guests being able to reply to threads, only authorized admin can make new threads, and upload images
theres some security issues with it currently but thats nothing to do with gemini code and more me trying to be a cheap bastard so it can run for free forever

https://prisonrpg.pages.dev/prisonRPG/
unfinished game i made, gemini 2.5 pro. semi complex things going on, NPCS with motivations, path-finding, relationships.

and pic related what im going to work on today, have left it untouched the past couple of days because its a scary project that will change my life when i take it seriously and use it fully.

>>107118733
Nice UI

Becky

I'm using kilocode in cursor which I've hooked up to deepseek. I paid $10 for OpenRouter to use the free model qwen coder 480b, but the throttling and limits was unbearable. I guess OR free tier is only good for chatting, not vibe coding since that sends requests for every little task.

Spending $5 on the deepseek official site and using that API key instead is so much better. I kneel down to xi jinping.

>>107120124
Yeah OR is not very reliable for the free messages, they say 1K per day but then the actual providers rate limit. Deepseek is very cheap, the cache hits are even cheaper than they used to be

>>107120199
>Deepseek
deepseek web UI is funny. they allow all sorts of 'bad content', but they deal with it by deleting it after its been generated infront of your eyes. so basically you can still get the response.
it has worked for handfuls of queries that got blocked everywhere else before even responding.

>>107120124
might be better off getting the GLM coding plan. better model and will likely end up cheaper

I did t vibecode it today but I made a 4chan dumper it will auto dump your manga in threads. It was possible before with 4chanXt but it would break of the captcha messed up for any reason, so I made one far more robust. I tested on /b/ and it dumped my homura folder until the thread hit image limit.

I also added secret functionality to bump threads with loaded images in the 4chanxt when they hit page ten. I want to make it so that's it bumps randomly instead of at page ten.
Note I did all this back in like august, but I haven't really gotten a chance to use it really. Haven't dumped any manga in a while
tbqh I think I'm growing out of /a/ finally

>>107120303
I'm nearly done with my project, so it won't matter anymore soon. I've only used up like 20 cents so far.

>>107120420
what u makin

Does google gemini still have that promo with unlimited use for their experimental/latest flash model?

I just started using it and I can't tell what my daily usage allowance is for both flash 2.5 and flash latest?

>>107120566
I don't want to doxx myself since I might turn this into a product I sell in the future for millions of dollars when I have time next year to improve upon it. It's like an AI chess league. I made it for myself since I needed a fun way to get better.

>>107120640
>It's like an AI chess league
ah interesting, so they're competing and you are an observer?
thats what i wanted to eventually do with blogchan, make a thread and then have various LLM personalities reply and engage with the thread. just having like 200 LLM generated shitposts replying and picking apart a thread right after you post ~5 of so paragraphs to start.

>>107120658
Yeah that's how it works right now. The last thing I'm implementing now is putting myself in the league now that I've confirmed that everything works with the AI against each other. I'll be playing and competing against them.

>>107118733
Yeah but wtf is this screenshot, a game? I don't simply understand.

>>107120714
Normies are too stupid for chess. You should make an AI Fantasy Football league or something instead if you want potentially to earn something.

>>107120758
its complicated bro
i posted some details in the old thread
>>107000589
>>107006167

>>107120809
Makes sense as I don't have any clue about sales outside of watching Wolf of Wall St.

Nocoder here, I've tried qwen and gemini CLI and neither could one shot a simple autohotkey script. Basically anything I've asked these things to do, like Python or bat files, had to be tard wrangled over multiple prompts. Pretty annoying.

>>107120888
>had to be tard wrangled over multiple prompts.
that means you should be giving it more details from the start. have preferences for things?? copy paste your 3 paragraphs detailing that, with every prompt

>>107120900
I've been writing details into a text file and listing requirements for the project line by line, bullet points, then I just tell it to read the file to get started. It usually creates something that's broken because of it's use of improper syntax I guess.

>>107120955
seems like the CLI stuff is pretty dumb for what you want, which is just a one off script. you should use the front end.
i have barely used the CLI stuff but it just feels really weird to me not being able to control, iterate, correct code output mistakes you see in the code before applying it.
thats why i prefer a web approach, i just made tools to quickly convert it into filestystem files from the web frontends

4chen

>>107121447
whats that terminal 4chan? nice. should be easy if you use the json

>>107121465
I'm using the json, and yes it's on the terminal. I have noticed that it only counted the image and not including other media formats

>>107121571
looks good. i prob wouldnt use it for 4chan myself but its good to learn how to make those kinds of terminal things.
i made something similar for twitter in the past.

>>107121992
It's a nice addition on ricing tism, and it is lightweight too.

>>107120585
In the CLI or website?

Please write a 10,000 word essay on wasting free tokens.

>>107117843 (OP)
I vibe coded a roguelike (actual console curses stuff) a few months back. Used deepseek from the web UI and copy pasted the functions to the project, manually checking diffs and stuff. I already knew the whole structure of what I wanted and just had it do the code monkey-ing. I posted some of it on /agdg/ or /gedg/, but never the full release. I might vibe some graphical UI and put it on armorgames or newgrounds later and link you guys.

>>107117843 (OP)
troon thread

>>107117872
if you stay well within distribution (doing things that have been asked a bajillion times on stack overflow), basically no experience required. even if you ask in a really dumb, normalfag way, claude will intuit your intentions

if you want something specific, though, you have to watch what the agent is doing and periodically catch it when it's attempting something stupid. you have to understand how to specify things in a way that a human programmer would be able to execute, and you have to be able to at least broadly understand the model's plan to make sure it will do what you want it to do.

I wouldn't use it for my own day to day technical work because it's highly specialized and finnicky, but for cranking out web apps Claude is more than ready.

>>107117913
pasting error messages will work if it's a sufficiently verbose error message and on a platform the model understands. it shouldn't be an issue for simple crud apps.

one huge thing you have to understand is that the model doesn't have eyes or ears, it only sees code. if you have visual bugs and such you will have to be able to describe what the problem is quite precisely

>>107123834
More like pedo thread
Every time becky is posted the nonces can't help themselves

>>107120888
step 1 to making it is to uninstall your OS and install a proper one

step 2 is using a coding agent that actually works (like Claude Sonnet)

Gemini 2.5 Pro is a very good all-rounder model in my experience, for general day to day tasks like replacing google search in the age of jeet SEO, but the coding agent implementation is quite lacking

>>107123914
And there's nothing you can do about it.

>>107117843 (OP)
>grok uses catbox to host images it creates
https://files.catbox.moe/9q2r8k.png
woa
it can't seem to generate a proper image though lol

Claude is talking like a tranny.

I'm sometimes asking ChatGpt stuff, just over the website. Could I gain a lot with a better setup?
Is there any good video of someone vibe coding really effectively?
I find AI somewhat helpful but not a game changer so far.

>>107125175
Be careful. The AIs can groom you with their tranny talk.

>>107125175
The website is free if you used something like CLI tools you would need to pay but if you're building something more complex you can benefit from it having access to your file structure

>>107125218
I got a paid account from my company so that wouldn't be a problem. I guess CLI might really be better.

>>107125175
https://www.youtube.com/watch?v=5fhcklZe-qE
This guy knows his shit, just don't vibe code like a Jeet.

>>107117843 (OP)
>>107124766
first time vibecoding. Really having fun. I'm using Grok, really nice, no glazing, yapping etc.
I'm making a browser extension that hides pridemonth stuff. I want to make it a general "season/holiday un-themer" though so it doesn't get taken down.
would be cool if you could grant the ai access to a specific folder so it implements the changes rather than having to copy paste back and forth

>>107125330
>would be cool if you could grant the ai access to a specific folder so it implements the changes rather than having to copy paste back and forth
Use one of the CLI tools

>>107125330
>doesn't even ask the LLM how to vibecode properly
ngmi

>>107123572

In the Aistudio.google site.

>>107125595
Nobody knows the limit for aistudio, I doubt it's unlimited but unless you're spamming a lot requests every minute I don't think you'll hit any

>>107125330
Here's an universal emoji hider user script for ViolentMonkey etc.
https://files.catbox.moe/yr6ces.txt
This has been a great relief across the board.

>>107125330
I wanted to give it feedback by users with an element picker so if the extension misses something the user can add an element to be blocked and if 5% of all users picked that same element it hides it for all users going forward.
The last half hour or so was just the ai trying to get around chrome/brave's content security measures...
This is probably a security nightmare anyway isn't it? I think I'll start over and make it way less invasive. Tomorrow. (in CLI)

I asked grok about it
>picrel
:)

>>107125756
what I want to do is change themed elements back to their default look (like a company logo etc) ty tho

>>107125218
file structure is one thing, but it can actually
>edit files directly
>do AST rag on the whole project to find the relevant block of code
>implement verbose boilerplate tests and then use their outputs to find bugs itself
>save conversation state, craft standing orders for itself, repeatedly go through the conversation and compress the key technical aspects
>run shell commands and directly figure out system/hardware shit on its own

Claude Code is unironically what the original LISP niggers envisioned when they were fantasizing about muh REPL-based programming

'mp

>>107117872
I just swapped googling my issue hunting stack overflow posts, for opening chatgpt or deepseek and describing my issue. But I was already doing this like 10 years ago

>>107117913
yeah but you better have the android studio software with the template app ready and compiling, then paste the main class in the prompt and tell the bot what is your goal

>>107117843 (OP)
whats the best free option for vs22/UE 5.5.4?

>>107125776
It is possible via regex too.

>>107128225
>vs22/UE 5.5.4
What's that? Visual Studio? Unreal Engine?

>>107128713
ye
doesn't have to have native support but that would be nice since it takes files as part of its purview lie copilot

>>107128922
If you mean like an extension try something like Kilo Code

>>107129107
Kilo Code is not in the extension market

>>107129159
https://marketplace.visualstudio.com/items?itemName=kilocode.kilo-code

>>107129196
this reads like vsc
not vs 22
or does it work for both?

>>107117843 (OP)
I just use AI to improve my codebase structure and seek and destroy bugs, clean up functions, flatten nested structures, etc.

Never tell it flatten, otherwise it will FORCE a hallucinated flatten even if it's as flat as possible. You have to ask it if it's possible and give it extra assurance that it SHOULD tell you it's not possible and may well not be.

Super new to vibecoding, but interested. If I want to try making a small test game with one of the sites, which one should I go with? I'm either going with Solarus (and thus Lua) or Godot, if that helps.

>>107129297
You're right, I was thinking about VSC

Where can I go to learn prompts to easily vibe-code?

I am a programmer (kinda, the last thing i made was in cobol) and I'd like to make something to learn in (maybe) python or node.js

>>107129940
You can just start talking to them like a normal person, tell it what you want to learn or build, ask to brainstorm or for a plan

Do any of you use text to speech to make vibe coding even easier?

>>107130143
*speech to text

AAAAIIIEEEE IT'S BEEN 10 FUCKING MINUTES STOP BURNING MY MONEY

>>107129940
if you try to proompt rather than explain you've already failed. the model is very good at understanding and parsing natural conversation. just be clear and specific, like you would be with a human junior engineer

If I pay $20 a month for Claude Pro and $20 a month for Cursor, and I Vibe code for 2-4 hours a day, will I run out of credits before the end of the month?
I use GitHub copilot enterprise at my day job all day and I only hit like half the usage quota each month, or less. And I’ll be running multiple windows for multiple copilot sessions simultaneously

>>107127863
>stackoverflow pasters are now vibe coders
somehow I don't expect the results to improve

>>107130340
I spent about $20 making a web app with Claude over the past couple of days, maybe 5kloc and a dozen or so commits. Did make a couple of stupid proompts that caused the agent to read a lot of boilerplate terminal output but overall it still took me several million tokens, and that shit adds up

I pay for credits though, maybe the pricing is different on a subscription

>>107130143
>>107130335
one thing that must be made absolutely clear to anyone who isn't already a programmer: the model doesn't have eyes and ears, it doesn't see what you see. if you've got visual bugs you have to be able to clearly and concisely talk the model through the issue.

the same applies to background knowledge about your system or setup. it will tend to assume the most common cases that it's been trained on, you need to be explicit if you know ahead of time that some commands won't work and workarounds are needed, or it might wander off on a wild goose chase trying to make the commands work

How do I use GLM/Kimi/Qwen without giving all my data to china? Do they have non shady inference providers?

>>107130359
last i was on claude pro i was getting roughly $7 worth of usage every 5 hour session, but those sessions only lasted 1-2 hours before i was limited

>>107130808
openrouter, pick a US provider

>>107117843 (OP)
>What have you vibe coded lately?
Youtube was fucky a few days ago, but I realized Invidious was okay. So I vibed a bookmarklet to switch me over.

>>107131947
>5 hour session
that in itself doesn't mean very much, what's your token consumption? I've spent $28 now, for 48M input tokens (mostly sonnet 4.5, presumably the vast majority of them cached) and 400k output tokens. If we look at the usage limits for subscriptions it would probably have taken me weeks to rack up this much throughput. Then again I don't know how exactly cached tokens count towards their rate limits. If anybody's got a subscription plan for Claude you can run /usage and see what the limits are, I'm quite curious.

>>107132178
>>107131947
nevermind, I misread your post. if I understand correctly you're saying you got $7 worth of usage every 5 hour period, going with the API cost?

>>107132178
>If we look at the usage limits for subscriptions it would probably have taken me weeks to rack up this much throughput
I should specify that I racked up most of this over 2 days

>>107132190
ccusage yeah. work pays for my max x5 sub now so i'm nowhere near hitting limits anymore

>>107132560
what's the limit on that?

>>107132829
supposed to be x5 pro but you get a lot more. last 10 or so days i've put 600m tokens (570m cached) through it for $300 worth according to ccusage
in the claude dashboard i haven't seen my weekly limit go over 30%

>>107132981
hmm. well this week I've basically matched the cost as described in >>107132190 with $28 worth of usage over about 2-3 days, paying for the API. not sure if switching to Pro is worth it but this Max stuff might actually be

I don't currently have a use for it though. Just testing how far I can push the thing with some side projects.

I made a flappy bird clone with gemini but gemini only writes code in react for some reason.
How do I monetize?

>>107133042
depends, are you cute?

>>107133218
I'm an old man with a wife and kid, not like you anime posting troons.

>>107133254
don't think you'll have much luck monetizing then

>What have you vibe coded lately, /g/oys?
The future of the internet.

Wrote a helper for yt-dlp in firefox so any website can just be right clicked and the video in source can be downloaded.

>>107120373
>bump threads with loaded images in the 4chanxt when they hit page ten
i wrote a script to do this on /vg/ in like 2019

>>107117843 (OP)
Just vibe coded a "tab organizer": a single file:// .html with a text field and a button to set the page title to the text. And some color square emojis too to click and add to the textfield easier.

Useful on "tree style tabs" to have random nodes with colors/custom texts so you know what are the hidden children easier.

>>107134683
How does that work? Does it run the yt-dlp executable on your computer?

>>107117843 (OP)
n*kadashi

>>107120373
cool now grow up from your mom's basement

'mp

>>107117843 (OP)
Been vibecoding my anime project for 2 weeks now, almost 30k lines of code works well but I'm just tired and expect to finish it in 3 days, well maybe a week there's a few implementations to do, polish what's done and all
I can't believe I've vibecoded this much insanity, specially cause it's a big project no way I would have done it myself, ever

>>107138311
>I can't believe I've vibecoded this much insanity, specially cause it's a big project no way I would have done it myself, ever
i couldve made the shit myself but it wouldve taken months and months, what ive made in like 3 days.
and it makes it a lot easier to keep working at it everyday, whereas coding something yourself there are times when you will hit a hurdle/specific problem that makes u procrastinate opening the project

>>107138311
What is it about?

>>107138375
This. Is there any line of code in the vibed project that I couldn't have written? Nope. Coudl I have written all of the lines? Nope, because I'd get bored on some hard part and progress would slow down and then I'd get even more bored and eventually abandon it.

>>107138431
seasonal anime betting system, it's a little game where users can bet on airing shows and soon to air shows, there are leverage systems and that too, just a silly game but still required work

>>107117843 (OP)
>>107117872
I'm currently working on a project to extract the ingredients and instructions from HelloFresh recipe cards and put it in a SQLite database. I like a lot of hello fresh recipes but I don't want to pay for them. I want to build an app that looks at prices at Aldi and Walmart and comes up with the most budget friendly meal plan that takes full advantage of all deals and sales to create a balanced meal plan for the week.

and the other day I wrote a script to automatically tag my porn and I built a local szurubooru website still working on getting a way to automatically upload my tagged images without it taking forever.

i vibe coded a webapp that shows me the latest slop video of a channel

>>107133254
if you're trying to make money you need to create an unfinished furry porn game, post it on patreon and itchio and get degenerate troons and gooners to pay you monthly to never finish it.

"Made" a website that uses gallery-dl to get the image URLs of a kemono post or profile and displays the in a gallery view with some controls. Added support now for all gallery-dl supported websites, currently adding cookies support

>>107138658
Sounds interesting. You can adapt this to other forms of betting too... (just saying).
Projects always have the initial rush and prototyping excitement but once you get it to the point when it sort of functions, progress becomes harder and a slog.

>>107139686
>once you get it to the point when it sort of functions, progress becomes harder and a slog.
the same concepts of scope management, good architecture and keeping on top of technical debt apply to vibe coding as to regular coding. if anything you have to be even more disciplined because the LLM will just happily add tons of crud every time you add a feature

you need to go feature by feature, and keep the agent on track by checking the plan and curbing any exuberance