Hey folks

I was wondering how much time it would take to recreate the basic functionality of VisualMind/PhotoTag and it turns out that not so much!

Here's what I did: https://xpiksapp.com/blog/build-ai-keyword-tool/

I know some people on the forum created a custom ChatGPT agent (requires ChatGPT Pro subscription), but my way is only relying on API key which is super cheap. Has anybody done anything similar?

We launched the generation of descriptions in IMS Vision! The neural network generates not only what it sees, but also what idea/concept is expressed in the file: https://imstocker.com/en/blog/152-ims-vision-desc-gen

P.S. There is also a coupon code now

Quote from: korner83 on January 21, 2024, 05:53

I like the Studio App, only bug I have found is when you already have meta data (keywords etc.) in the image file then I can't submit it into the IMS Vision. "No matching file selected." error appears on top of the screen in red. But the image is selected, I also tried with multiple images, but it is the same error. I can only use the IMS Vision if the keywords are not populated in the image.

It would be grate if I could still send it to IMS Vision and merge the keywords.

Also auto category function would be nice.

Quote from: Imstocker on January 20, 2024, 15:14

Quote from: korner83 on January 19, 2024, 10:54

Can it also generate Title and description?

Quote from: Imstocker on January 19, 2024, 00:23

Try: https://vision.imstocker.com/

It can be also used with IMS Studio that gives ability to save result of recognition directly inside the files. It very convenient for batch processing. Alternatively, you can export metadata to popular CSV formats

We are working on that. For now it generates only keywords

There is option in settings "Keyword suggesting > Do not send files to IMS Vision if they already have keywords". But if you deactivate it, your keywords would be replaced by new ones. I think we can add such function to not replace but to supplement keywords.

Quote from: korner83 on January 19, 2024, 10:42

Is https://dam.everypixel.com working for you? I always get a generic error and can't rid of the pop-up window which want me to create a new project.

Both the upload and AI keywording worked.
But I have not tried the release to the stock agencies.

Would be great if one could use 3 GB for free.

Quote from: stoker2014 on November 16, 2023, 09:45

To effectively generate titles and tags, this chat must be able to receive batches of 30-100 photos (in low resolution) and produce a list with the name of each file, title, description, and tags as an xls file. And all this must be done quickly.

If you upload a "Iron Man" image it maybe uses the word "Marvel" or similar? Need to check everything after. Do not want my account blocked. Dangerous game

To effectively generate titles and tags, this chat must be able to receive batches of 30-100 photos (in low resolution) and produce a list with the name of each file, title, description, and tags as an xls file. And all this must be done quickly.

Lol, no - I know what I am talking about, and haha, no, there is no "ptsd" associated with it... But haha, thank-you - that's kind of funny. Maybe with all the hard work & years to comprehend it - maybe there was some 'ptsd'...

Anyhoo. "It" is not "intelligent". It is a sophisticated computer algorithm. The same way you would not call a calculator intelligent - just able to process numbers faster than you most likely do in your head. I guess for you it passed something called the "turing test" (you should look that up). And it's not the worst comparison, (not even poor at all) - lol - it is a very apt description. It parses natural language 'better' than google (I wouldn't say perfectly), it re-arranges content and applies a randomization element to it to seem 'unique'.

And lol - the examples I presented aren't actually aren't "theories", there is ample documentation, evidence, and it just being the plain truth to support that. It is important to know the difference between a "conspiracy theory" and an actual "conspiracy". But - that word/phrase in itself ("conspiracy theory") - is a triggering mechanism designed to make people automatically dismiss a subject they really don't know anything about, nor seem to have the inclination to look into. It is a psychological tactic that many have been programmed with (much like a computer) to attempt to lift one's own opinion of themselves while attempting to dismiss an argument with no actual thought or coherent argument. You really should look into that, but I do know sometimes it can be hard to get past one's own personal bias.

That component actually very relevant and on-topic, because it shows this so-called "ai" fails the "turing test". (Not that that is the only prequisite, but it is a very popular test within the "ai" community to see if a machine effectively simulates intelligence).

The people that "own" the machine (small group of very rich psychopaths via 'investment' firms, etc) want to basically use it to "control" people's thoughts/emotions, but different topic. The "content" you have access to ('subset of data') is not the same content "they" have access to.

Back to the "ai".

a) It is part of the 2023 narrative (storyline) that has been HEAVILY pushed by the media. "ChatGPT" is nothing new. It, and systems like it - have been available for the last 20 years. This goes for midjourney as well - I personally used systems similar (albeit not as fast) many many years ago.
Based on what I've read/seen - it seems the big co-ordinated push was to try and 'scare' people, make them think there is a 'thinking' machine (it is not), and try and push them into welfare.
b) It is not a "thinking" machine. It is essentially a pattern re-arranger, that uses 'weights' to determine associations (context) while doing natural language parsing.

Getting back haha to I suppose the real question - which I guess you got a bit off topic when I was answering someone else... (do you have any experience prior to 2023 with "ai" systems?)

The "keywording" mechanism works by essentially finding associations with words (look back into Princeton's (circa 2000) project called 'wordnet' for a basis).

"It" (the "ai") scraped (stole) other people's content (and then subsequently many agencies "paid" you for those same keyword associations with that content to hand over to the "ai" scraping sites, with no opt-out - so they knew what they did was wrong)... anyways. It then created models of these associations.

So when you type in something like "toaster" - it searches through its computer model to figure out what weights (i.e., "strength") of associated words with that based on the particular context - and then gives you a list of associated keywords.

It's like you looking up other people's keywords for the word "toaster", noticing which ones appear most frequently (i.e., assigning "weights") - and then decided that those are the most 'relevant' keywords. This computer system just simply does it a lot faster than you counting all the words by hand.

pps, if you want to have some fun to see just how "intelligent" it (chatgpt) really is...

type in a sentence questioning any of the "official" narratives... i.e, "climate change" being caused by airplanes spraying chemicals in the sky and deliberately lit "forest fires", the "carbon footprint" is designed to restrict people's travel under the guise of 'saving the plane-t", ... the "covidnarrative" nonsense (i.e., never was, still isn't "a contagiousvirus", the jabs in fact are poison, designed to cause everything from cancer to infertility to more), the 'illness' ppl think they experienced was in fact from poisons on the mathsks, the cleaning products, the isolation, that (most) were emotionally manipulated into wearing/doing - all designed to poison themselves via a shot)... that "electric" vehicles are designed to ween people off "oil" for control (restrict travel, try to make vehicle ownership prohibitive)... that the "ai" narrative (which most people were CLUELESS about until January 2023 when BOTH CNN/FOX "news" started HEAVILY promoting that into the 'public consciousness) - the "ai" narrative is designed to try and make people afraid so they are happy to collect a welfare check and be told what to do & how to live their life, otherwise they 'wont get food', etc... question ANYTHING like that...

if it rejects the premise (which it does, it becomes dismissive/name calling, a tactic designed to avoid answering the question), provide proof/evidence... it has been "programmed" to reject certain information...

the "knowledge" it has is what it has scraped (stolen) off the internet, and other sources, and of course what it has been programmed to output/topics it has been told/trained are 'off limits', etc...

pps - the guy in the video when he talked about "it" coding, "it" didn't code - it basically searched online coding sites, & re-arranged certain aspects of code (and it usually does it incorrectly)...

"it" is essentially a more advanced "google" search engine that mimics aspects of human intelligence, but it is certainly not "intelligent".

when "you" use it, "you" are training it to become a better SIMULATION...

Quote from: stoker2014 on November 15, 2023, 02:45

Quote from: korner83 on November 13, 2023, 10:03

Hello guys,

I have created a GPT to generate meta data for our stock images:
https://chat.openai.com/g/g-RlTKUHzlr-image-metadata-guru

You can upload your image and it will give you Title, Keywords in the order of importance, description, category and it will also notify you if there is any visible trademark or logo related issue with the image. It is not supporting editorial content.

You will need a subscription to use it. Let me know what you think about it.

On what basis are keywords created? Where does GPT find them?

ChatGPT (as well as other so-called "ai") basically "stole" other people's works, to make inferences/associations with certain things...

Going back to 2000 with something called WordNet (from Princeton) to now, easiest way of explaining it, is:
a) the "ai" has created computer models of certain "concepts" (more specifically, certain patterns)
b) it has assigned 'weights'/associations with certain patterns
c) when you input an image/a sentence/etc - it searches for patterns that have greater weights (associations)

So chatGPT "found" the keywords from the 15+ years of "keyword tagging" that regular people have done, ripped it off - and then associates it with certain image patterns you input.