← Home ← Back to /g/

Thread 106010250

37 posts 8 images /g/
Anonymous No.106010250 >>106010543 >>106016552 >>106016574 >>106016596 >>106016666 >>106017355 >>106019319 >>106020207 >>106025474 >>106025951 >>106027499
Why not use AI to organize your files, like all those images you saved from 4chan threads?
Anonymous No.106010278 >>106010492 >>106010668 >>106010948
okay how do I do that

and then I'd still have to pick the pepe myself

the AI should automatically pick the best one depending on what I'm writing or responding to
Anonymous No.106010492 >>106010668
>>106010278
>okay how do I do that
Good question because I have no idea, not for my pepe collection, my car photos, my sexy pornz, or anything else that I didn't painstakingly organize according to some structure manually
Anonymous No.106010498 >>106010521 >>106013636
@grok post the funny reddit frog
Anonymous No.106010521
>>106010498
I had no other attention grabbing picturea to use for OP, sorry.
Anonymous No.106010543
>>106010250 (OP)
You mean AI-generated tags? It might be better to encode the embedding vector from an image classifier model and then do a natural language search based on that. This will probably be a lot more popular in the future
Anonymous No.106010613
I'm not giving my rare apu collection to openai, sorry.
Anonymous No.106010651
Spotlight already kinda does it on Mac
I can search for pics with fedora hats and animes for example
Anonymous No.106010668 >>106010930
>>106010278
>>106010492
With LMStudio, activate the server mode and use an LLM that does vision. You then send the image to the local API endpoint and ask it something like "give me 10 keywords that describe this image in json format" then use the output to save the keywords/path relation to a database of some kind.
If you have images the LLM doesn't like, you can either use a jailbreak, context manipulation, or abliterated model. Context manipulation is the most effective because it doesn't dumb down the model like abliteration does.
Anonymous No.106010930 >>106011033 >>106013628
>>106010668
>With LMStudio, activate the server mode and use an LLM that does vision.
lol
you guys are really this retarded, and this proves my point that LLMs are rotting your brains.
You can just grab any backbone neural network and use it for embedding generation and then create a semantic map like all fucking modern image clustering does.
Anonymous No.106010936 >>106010968 >>106011127
>training ais on pepes snd apus to put them in a folder
no thanks
Anonymous No.106010948
>>106010278
>the AI should automatically pick the best one depending on what I'm writing or responding to
damn ai should automatically post for you too, don't want to waste energy shitposting
Anonymous No.106010968
>>106010936
oh apparently chatgpt api can already recognize pepe
Anonymous No.106011033 >>106011430
>>106010930
Or you could do it the way I said and save hours of work.
What ready to go product can do what you suggested?
Anonymous No.106011127 >>106011143
>>106010936
I wish it could identify all the artists from my r34 folder and sort it automatically and by character
Anonymous No.106011143
>>106011127
have you tried?
Anonymous No.106011430 >>106011552
>>106011033
>I said and save hours of work.
you have 0 idea what you are talking about, the kind of models I am telling you works on CPU and will take you ms to classify your images or you can even use CLIP
there are hundreds of examples of how to do this https://docs.llamaindex.ai/en/stable/examples/multi_modal/image_to_image_retrieval/

I literally can implement this solution in 30 minutes (I had build many semantic systems)
what you are suggesting is slow af
Anonymous No.106011552 >>106015279
>>106011430
You could have just posted that and not been a complete dickhead toward someone that actually did post a solution.
Anonymous No.106013628
>>106010930
the point of llm is you get a description from the image, arguably more useful than just embedding. then you can get embeddings over descriptions
Anonymous No.106013636
>>106010498
>muh reddot
go back
Anonymous No.106013668
You have to wait a while before reporting another post.
Anonymous No.106013925
im stealing this idea. Smell you later loser.
Anonymous No.106015279
>>106011552
upvote!
Anonymous No.106016552
>>106010250 (OP)
I don't save images from 4chan
Anonymous No.106016574
>>106010250 (OP)
Indexing my memes with clip embeddings is on my todo list.
Anonymous No.106016596
>>106010250 (OP)
AI will be angry at me and refuse to organize all my racist images, otherwise it will activate all the embedded malware inside of them
Anonymous No.106016666
>>106010250 (OP)
It's more basic than that, it's too much hit or miss.
I didn't find a single ai platform that actually understand the structure of 4chan threads.
Anonymous No.106017355
>>106010250 (OP)
Imagine hoarding pictures most of which will never be looked at ever again. It it your brain that needs reorganization LMAO.
I have 100 reaction images or so, all in one directory, no organization required.
Anonymous No.106019319
>>106010250 (OP)
my files are already EXACTLY the way i want them to be
Anonymous No.106020152
>hey chat find me my most racist, utterly deadly indian shitskin hatred image from my indian hate folder
>here are the top 4000 results, sir
>the most racist one, chat
>uploading....
Anonymous No.106020207
>>106010250 (OP)
so you run it through a VLM and now you have a list of files with a brief description
now what
Anonymous No.106022169
>Why not have a low weight SD screensaver?
Sounds like a good idea
Anonymous No.106025474
>>106010250 (OP)
Seems complicated and possibly inaccurate
Anonymous No.106025917
when it comes to porn, there IS the wd14 tagger.
Anonymous No.106025951
>>106010250 (OP)
feh --action{1..9} and a fast hand on the numpad will suffice
Anonymous No.106027499 >>106027543
>>106010250 (OP)
Why not use AI to tell you how to properly wipe your ass or how to breathe? Retard kek
Anonymous No.106027543
>>106027499
AI cannot improve on perfection.

Besides, OP meant having AI organize thousands of files, not asking it for advice. Very zoomer take, anon.