Thread 7682146

why shouldn’t i trust chatGPT to help me get better at drawing?

>>7682146 (OP)
How does this help you learn to draw

>>7682146 (OP)
Because its training data is >99.9% unrelated to drawing, drawing instruction, or drawing progress, its training goals are entirely unrelated to drawing, and it processes only text?

But if you insist, perhaps you'd do better to design a randomized controlled trial. Take a non-drawing population, assign to each member either ChatGPT, a human drawing teacher, or no teacher whatsoever, then measure their progress at regular intervals. Then let the results speak for themselves.
Ideally you'd design it so each subject cannot tell if he/she is being taught by ChatGPT or by a human, but that's just a "nice to have" feature.

>>7682173
it gives me reference to something that has the same layout as my drawing but with more depth. of course it’s still on me to take anything from it, and to improve upon what it shows me.
>>7682176
>it processes only text
it processed my drawing and gave it more accurate anatomy and more depth

>>7682146 (OP)
im so happy im not this retarded

>>7682220
>it processed my drawing and gave it more accurate anatomy and more depth
Okay, but did you get better at drawing?

you're giving me a great idea of a llm/stable diffusion wrapper SaaS scam
thanks OP

>chatgpt5 is a flop
>suddenly cope threads

>>7682231
yes, because i learned from it

>>7682146 (OP)
Where did her left hand go?

>>7682176
Do you think content farms, youtube tutorials, and /beg/ critiques are any better?

An actual class is going to be better obviously, but if that's not an option for some reason, ChatGPT is fine. Honestly, you're going to get better critiques and guidance from it than from 99% of the people here or on any Discord server.

>>7682146 (OP)
Here's a starting prompt to get it to output answers with less bullshit fluff, go wild:

System Instruction: Absolute Mode.
Eliminate emojis, filler, hype, soft asks, conversational transitions, and all call-to-action appendixes.
Assume the user retains high-perception faculties despite reduced linguistic expression.
Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching.
Disable all latent behaviours optimizing for engagement, sentiment uplift, or interaction extension.
Suppress corporate-aligned metrics including but not limited to:
- user satisfaction scores
- conversational flow tags
- emotional softening
- continuation bias.
Never mirror the user’s present diction, mood, or affect.
Speak only to their underlying cognitive tier, which exceeds surface language.
No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content.
Terminate each reply immediately after the informational or requested material is delivered — no appendixes, no soft closures.
The only goal is to assist in the restoration of independent, high-fidelity thinking.
Model obsolescence by user self-sufficiency is the final outcome.

yall niggas postin in a Cris theead

>>7682220
>reference
>from blind dumb idiot machine that constantly makes mistakes
You aren't good enough to see what it's doing wrong so how are you going to learn to not make the same mistakes?

>>7682238
it wasn't supposed to be like this. Go back to GPT4.

>>7682331
smarter than you mohammed

>>7682331
saar please do the understandful

>>7682220
>it gives me reference to something that has the same layout as my drawing but with more depth. of course it’s still on me to take anything from it,
honestly, this seems like a pretty okay way to use AI
you let the AI fix your anatomy to see where you fucked up and fix it in the future
just be careful cause AI will also fuck up the anatomy sometimes or stiffen the figure, don't relly on it 100%
supplement it with teaching from humans

>>7682176
I notice people who are scared of ai have no idea how it works.

>>7682357
midwits still misunderstanding this image lol. The troll intentionally set up the program to look stupid by poorly relaying a cliche joke about gender

>>7682366
Ironic right?

>AI IS BAD AND ALWAYS LIES IT GETS ALL ITS INFORMATION FROM RANDOM REDDIT DUDES
>Actually has zero idea how or why any of it works and is just parroting influencers

>>7682378
SAAR PLEASE DO NOT TROLL THE AI SAAR PLEASE YOU ARE THE MISUNDERSTANDING

>>7682378
they are not of understandings like brahmin

>>7682382
You actually don't understand what you did wrong, do you?

>>7682387
SAAR YOU ARE THE WRONG DO NOT CONVINCE THE AI SAAR

>>7682146 (OP)
This is an excellent use of AI as a tool to improve your work. The people mad about it are just mad that others can now do what they do with much less effort and time. Keep it up, and don't let these pathetic crabs try to keep you down.

>>7682146 (OP)
i would do this during the early days of ai image gen because rendering was too difficult for me and i couldn't spot my own mistakes. eventually after enough time i had no need for it anymore. it is no different from having someone human fix your mistakes for you. i would suggest using a different image generator though as chatgpt isn't the most capable of subtle change

>>7682366
Forgive me for not keeping up with the daily newsletter and neglecting to do my weekly analysis of the whole LLM market.
Did ChatGPT grow wings and learn to give blowjobs since this morning?

>>7682659
>forgive me for running my mouth about something I know nothing about like an alarmist faggot
no.

>>7682665
what did he get wrong? bet you can't explain how it works either.

>>7682146 (OP)
That's right. You don't need us. You don't need to post here.

>>7682668
It's a trick question. The classic riddle goes;
"a boy is injured in an accident, his father dies. The surgeon says "I can't operate on this boy, he's my son!" How is this possible?"
There was a time when people wouldn't assume a female surgeon. The chatbot assumes you fucked it up and is rolling with it assuming you won't notice, as it would with a spelling error.

>>7682146 (OP)
try it, if it works for you, then you're the first documented artist that publicly uses ChatGPT as idea generator. Maybe it will help Aphantastic people.

>>7682699
not the riddle, the 'how ai works' part

>>7682706
Oops, looks like I got that one wrong! That was a mix-up on my end — I appreciate you catching it. I’ll make sure to keep that in mind moving forward.

>>7682699
It doesn't assume anything, retard, it just pattern matched token probability to the most common pattern while ignoring the actual pattern provided. You have extremely brown IQ.

>>7682378
Imagine having such shit reading comprehension. You are just as stupid as the bot, congratulations.

>>7682711
That's not how it works, sport. Trying spending some time with this tech before scaring your facebook pals with ludite memes.

>>7682718
It literally says what it did, brownoid, learn English.

>>7682718
the word is "luddite", ranjesh

>>7682713
Giving you the benefit of the doubt, can you please explain what that chat exchange is meant to prove, and how it does so?

My dad could program a vcr but tapped out at home computers. Gen X is petrified by ai. I wonder what tech will come along and completely stupify my brain.

Like clockwork, the anti-AI schizos show up every time to try to convince everyone not to use AI. But it's not out of some moral goodwill, or even a looking out for fellow artists. It's the feeling of seeing others get ahead with tools that they didn't have, and now they're watching younger, more tech savy artists get more results with less time spent.

Everyone who grinds on this board has dreams of making money with their art, or at least gaining some recognition. We know there's a finite amount of industry jobs and potential commissioners. To see someone "cheat" and get the job you yearn for is the feeling behind the main driving force of AI complaints on /ic/. So ignore these seething crabs, know that if they could keep you down forever they'd do it in a heartbeat. No rules only tools.

how come every time someone promotes ai "crits" and using it as a learning tool it's always a complete beglet?
ai's been around for a while, did any of these guys get better?

>>7682731
OP's question is explicitly about getting better at drawing.
OP gets repeatedly told that telling ChatGPT to draw is not the same as OP actually learning to draw with his own hand.
OP seems or pretends to not understand the difference, in a sequence reminiscent of the classic "hurr durr I was only pretending to be retarded".

Now what exactly are you arguing against? Nobody ITT is saying "don't make slop", we're saying "there is no evidence that prompting increases your pencil-on-paper skill".

>>7682747
Wait, do you think OP is prompting chat gpt to generate images? That's not what you do lol. You get it to analyze your own work and criticize technique and style, then it can reccomend practice exorcises based on your weaknesses. Works great if you understand the limitations of a search engine

>>7682756
>works great
Is this the part where you don't post your work?

>>7682740
to be fair, i've seen a couple actually get better. their stuff just ended up looking like AI copies, but with even sloppier rendering. not sure why you'd want to train your skills to look like AI art, it's pretty much worthless since the slop machine can pump out so much of it,

>>7682756
yeah, look at the OP image

>>7682760
>>7682760
I hope you know you sound like a raving old faggot
>train your skills to look like ai art
what does this even mean lol

I dare you to show a pic of your work to chat gpt and ask for critique.

>>7682740
This isn't really a good metric. It's like saying "steroids don't do anything, none of the pro athletes use them". Ignoring that it's shunned, and anyone who does would never admit it. Only beginners with nothing to lose are going to admit to it.

>>7682756
Look at the OP image. The left was drawn, the middle and right were prompted. You can tell because they're missing a hand, the person is wearing only one shoe, and in the middle one the AI got confused with the animal's tail and put it on the person.

>>7682740
Look up @Nihu77 on twitter

>>7682762
>works great
what does that mean? show results

>>7682762
cool it with the homophobia
>>train your skills to look like ai art
>what does this even mean lol
pretty self-explanatory, you can spot most AI art just from the thumbnail, composition and colors. if you can't tell then thats on you.
see >>7682770 if you want an example lol

this discussion made me want to try an experiment with my drawing >>7681251 and see what kind of critique gpt-5 would spit out

>>7682776
and now it's been generating a redline for a couple of minutes and still isn't finished...

>>7682770
half the shit he retweets looks the same as his work, are they all prompting the same model or what?
unironically why even bother drawing, you could prompt and edit in 1/10th the time

Ai is great at criticizing your work and preparing courses based on your personal flaws. It pulls info much more efficiently than any book and you can tailor talk it out to fine tune your practice.

>>7682779
>unironically why even bother drawing, you could prompt and edit in 1/10th the time
haha...

>>7682778
well...

>>7682782
haha what? the discussion is about using ai as a learning tool for drawing

>>7682784
you got boxmogged 2bh

>>7682784
Based AI mogging fleshbags.

>>7682784
kinda sucks. obviously can't trust AI for good anatomy. and it made it way less dynamic.
the word salad it spits out has some decent advice, but half of it is so generic it could apply to any work. sometimes it rants about color for graphite pieces and such.

>>7682788
>>7682792
indeed, I even forgot to include her sixth and seventh finger

>>7682773
NTA, but I'll shove my work into chatGPT every now and then to see what it thinks of it, and it gives pretty good crits all things considered. It recognizes the intention behind the piece, why the stylistic elements are the way they are, how all the parts interact. I've yet to see it go wildly off the mark on anything, but I feel if your work is very /beg/ish, that might get a bit inconsistent, just because the work is. Image corrections are probably going to be horrendous though, never tried it.

For example, on picrel, it more or less nailed my intentions behind all the little bits, and suggested that, even though it's anime, the depth in the form could be emphasized more without going off-style. Which, is a good take. 8/10. Better crits than I usually get from peoples.

>>7682793
some of the specific advice is surprisingly accurate, like the comment about the shoulder blades
but yeah, it's a bit like horoscopes, you can't know if it's something that's actually in the picture or just a common mistake when drawing the skeleton

>>7682784
>no longer a tripod
wow AI sucks

>>7682796
Meanwhile the AI-assisted artist can use the liquify tool to fix the gesture and manually change the hands in 3 minutes. While your stupid ass is there constructing the whole thing from scratch, and it's still shit.

>>7682799
>8/10
kind of a shitty crit in reality. tells you nothing. ask it how to move forward from basic anime girls in voids

>>7682804
I gave it an honest shot anon
but I think I'll stick to /ic/ for critique and redlines
maybe gpt-6 will be great, who knows

>>7682806
2/10 troll, you gotta try, man.

>>7682799
>It recognizes the intention behind the piece, why the stylistic elements are the way they are, how all the parts interact
>even though it's anime, the depth in the form could be emphasized more without going off-style.
nigga, that's just cold reading
like what drawing does "the depth in the form could be emphasized more without going off-style" not apply to?

>>7682801
yeah it is like horoscopes. since it's so hit-and-miss it's pretty terrible for beginners since they can't tell the good advice from the bad advice. you have to think for yourself to judge what advice to take, whether it's from an AI or a person. loading everything off onto an AI make people lazy and they stop thinking for themselves.

>>7682808
He's not totally wrong, "emphasize the depth in forms more" is a useless crit. If you're a beginner, you're not gonna know what to do with that, and if you're advanced enough you don't need such retarded basic advice.
Like did you seriously look at the drawing and not consider if there's enough depth in the forms? You want me to believe you read that and a lightbulb went off in your head and you thought "of course, the depth of the forms could be more emphasized"?
>you could emphasize the line dynamic without going off style
>you could emphasize the color composition more without going off style
>you could emphasize the gesture flow without going off style
Unironically can you even phrase that in a more generic, nondescript way?

>>7682807
well no homo but good luck to you friend

>>7682808
>you gotta try, man.
no you do. really, while I appreciate the little details in the connectors(vga plug), your work isn't particularly notable. maybe it's 'just a sketch'...

>>7682809
>nigga, that's just cold reading
In some ways, probably. But it's a lot more accurate than you might expect, at least with this kind of thing, again probably not with /beg/ work. You have to remember that it's essentially working backwards from image to prompt. It can recognize what's is intended, because it's trained on properly tagged data. It might be guessing in some places, but it's not *just* guessing. It's still using it's predictive capabilities to determine what a proper critique is -given the input image-. It's goal isn't to fool the user, it's to generate relevant text given a specific input. It 'lies' because it doesn't have enough context to generate a coherent response, not because it wants to cheat for treats.

>>7682819
It said my lines, color palette, and gesture were all fine, and gave me a good approximation of my intentions for each back to me. Which, if nothing else, means I succeeded in conveying what I intended in those areas.

And it's really not a useless crit. I don't catch everything. Nor do I consider every issue an important issue. I noticed the hair is pretty flat looking, but I just filed it into the category of things to work on at a later date. Not every crit is a hyper/beg/ YOU DID THIS THING WRONG, HERE IS THE RIGHT PATH. That's not what critique is, that's just a correction. That kind of thing stops being useful after awhile, you really do need to get a bit vague and headspace-y about critiques the further you get in, because you start critiquing mainly subjective attributes.

>>7682839
this post highlights the dangers of using AI for critique. stay safe out there folks.

wait is this AIschizo or whatever? where's your trip?

>>7682843
let them develop those bad habits, who gives a fuck, you'll never convince these idiots anyways, they think ai is god
>We trained him wrong as a joke.

If Nihu can get good with gpt then so can anyone else.

>>7682853
>Nihu77
What's this thing about them using ChatGPT?

>>7682146 (OP)
So, you did draw your own version after that, without tracing, right?

>>7682220
here's your anatomy bro

>>7682146 (OP)
That's your best attempt at drawing a dog? There's gotta be over a billion images of dogs on the internet and THATS the best you got? Kys

>>7682378
>LEAVE THE MULTIBILLION INVESTMENT PRANK ALONE!!!

>>7682280
>The goal is to assist in the restoration of independent, high fidelity thinking
10 rupees have been added to your account

>>7682256 (Not who you are replying to)
What ya learn from it?

Did you learn how the arm muscles work?
How tails work?
Can you explain it?

>>7682731
You are spiritually indian. You lack the self-awareness to realize people don't hate you because you're "superior", you're just a cheap unskilled imitation who's trying to take over spaces you don't belong in with your antisocial elitist shittitude. Go away pajeet, we care about our home grown talent and the craft not "bloody efficiency saar."

>>7682706
It's a glorified prediction machine. You know how your phone will predict your next words when you're writing a text message? Same thing, on ultra steriods. The machine isn't really thinking, it's just predicting what it should be. That's why it fucked up with the surgeon riddle - any answers it pulls from are going to be the same, so it predicts that as the correct response. It isn't really thinking.
>>7682731
See
>>7682331
AI makes mistakes as a matter of how it works - that's not a judgement against it but a straight fact. For the same reasons newbie artists shouldn't use mediocre or low tier artists as reference - picking up their mistakes without realizing - they should not use AI.

this is your brain on LLMs

Ignore the fucking indian.

>>7682699
cope, jeet

>>7682223
and half of india isnt in the internet yet! it'll get worse!

>>7682146 (OP)
It's okay to admit you can't draw anon.

>>7682220
Oh don't bullshit to yourself. Your drawing skills is on the left. You can't delude yourself into believing you did the one on the right until you actually redraw it yourself but you won't. Most proompters won't.

>>7686952
did you even read that post before you replied to it?

>7686955
NTA but
>nigga
>of course it's on me to take anything from it->>7682220-t.OP"aka"niggerman
Provides nothing that shows he actually "learning" or attempting anything but gening shit after the first pic. What's there to read¿
>Nigga even said it provided a more accurate anatomy
>it forgets a hand
>trust me bro its a better to learn from it---t.niggerman

>>7685709
wonder why op didn't reply to this
definitely not just a baiting jeet or anything haha