MicrostockGroup Sponsors


Author Topic: Maigen - Automatic metadata generator  (Read 707 times)

0 Members and 1 Guest are viewing this topic.

« on: July 05, 2024, 00:34 »
0
Hello everyone

I'm developing software to automatically generate metadata using the gemini API, it's called Maigen.

You can try it here
Download : https://drive.google.com/file/d/1wCP4IO9rTWPBSH71JBBW3SFBlYetoluG/view?pli=1

Trial key : MAIGEN_TRIAL_IKN48

Currently only available for Windows
Tested on windows 11

Watch the demo video here: https://youtu.be/X69rVdqpddw

New features:
- Ability to export metadata to CSV file.
- Ability to select Shutterstock and Adobestock auto categories.
- Ability to filter the content list.
- Ability to run the generate process simultaneously.
- Ability to automatically rename files based on generated titles.
- Ability to stop the generate process.
- Ability to limit the titles and keywords to be generated.
- Ability to display detailed status of the generate process.
- Ability to fully automate the generate process. auto pilot.
- Some additional menus for help.

If you find any problems or have any suggestions, please let me know.

Thank you.
« Last Edit: Yesterday at 22:04 by edihasrin »


« Reply #1 on: July 05, 2024, 02:29 »
0
What will be the price? Just wondering.

« Reply #2 on: July 05, 2024, 03:07 »
0
Is this free? How do you get API from Gemini?

And what is the privacy?

« Reply #3 on: July 06, 2024, 00:59 »
0
What will be the price? Just wondering.

23
« Last Edit: July 06, 2024, 03:18 by edihasrin »

« Reply #4 on: July 06, 2024, 02:06 »
+1
Automatic metadata generator with modern and advanced features

Hello there.
Modernity is always relative.
In the 70's, my parents bought and used this as a "modern" tool:


Today, I use this,
Just because, it does the job better and quicker  ;):

And you know what? I save money, still today.

Your tool is unable to achieve what I do in my workflow for 19 years now. But I don't and will never produce AI images, maybe the reason.   ;)
But maybe later, I will buy a humanoid robot to push the button to start my computer.  :D
« Last Edit: July 06, 2024, 03:01 by DiscreetDuck »

« Reply #5 on: July 06, 2024, 03:11 »
0
Thank you very much for the feedback, I will fix it.
« Last Edit: July 06, 2024, 03:20 by edihasrin »

« Reply #6 on: July 06, 2024, 06:55 »
0
What will be the price? Just wondering.

23

What is this 23? And what kind of answer is this?

« Reply #7 on: July 06, 2024, 08:12 »
0
The free Google gemini API I have tested generates title, keyword, and category metadata for up to 500 images in 1 day. maybe some accounts or some countries are still limited. for the price is $23/month. If your gemini API is hit by a limit, you can request an increase in RPM (Request per minute) here https://ai.google.dev/gemini-api/docs/quota

« Reply #8 on: July 06, 2024, 08:39 »
+1
While I haven't tried out your tool (it may be very good, so good job/kudos first of all putting it together! :)), I've found many "ai" keyword generation tools have issues (especially the ones based on OpenAI)...

While "most" of the time they provide good results, probably (deliberately) by design OpenAI sticks in a bunch of garbage/irrelevant crap, such that you still need to manually review everything... (even if you've refined the prompt over and over and over and over again)... openAI is very annoying that way...

So - if you had say the picture of the cat with the can opener as above, "most" of the time it might generate decent results, but then it will stick something marginally relevant (or completely irrelevant) like "zoo, ferocious, apple, cheesewiz, salami, tuna, creative, commons" and you are like "what is that?"... or... it will do something like (for keywords) EVERY now and then "cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat, cat", or... just be totally way off base like "astronaut, television, commercial, telephone", and you are like 'huh'?

So... not sure if you have done extensive testing, but something important to look at, you'll probably need to do many iterations of testing before you get ideal results, and even then, the chatGPT/openai (which are based off of MASSIVE theft) - will give really crappy results...

Just an FYI.

But anyways - definitely good job on putting a tool together! kudos for that! :)

« Reply #9 on: July 06, 2024, 08:44 »
0
PPS - looked at your video, cool.

a) What programming language did you write it in, & how backwards compatible is it (i.e., which windows version)?
b) From a UI standpoint - I'd remove the "warning sign" that keeps popping up about "too many AI generations! wait a few seconds!". Once or twice - okay - but when it pops up multiple times - a user would find that extremely annoying. From a programming perspective - you should be handling all of that behind the scenes (i.e., automatically adjust the timing/API requests/etc) so the user doesn't have to see that... Otherwise - they think they are doing something "wrong", can "panic", and/or just find it quite annoying especially when it is in a red alert like that...

Overall, good job though!

« Reply #10 on: July 06, 2024, 09:30 »
0
Thank you very much for your advice.

I don't know the flow of other AI tools if they only use prompts like "Create a title and keywords for this image".

but for myself my benchmark is the adobestock metadata guide, in the prompt I provide several example images from their documentation here https://helpx.adobe.com/stock/contributor/help/titles-and-keyword.html and some in PDF files, titles and keywords provided in their documentation.

I then emphasized to focus on choosing metadata that is likely to be used in search engines, with the 10 most important keywords should be at the beginning and in the title, then some other commands.

For now I'm satisfied with the results, but sometimes if the image is difficult to understand the context, the results can be confusing.
« Last Edit: July 06, 2024, 09:46 by edihasrin »

« Reply #11 on: July 06, 2024, 10:30 »
0
Thank you very much for your advice.

I don't know the flow of other AI tools if they only use prompts like "Create a title and keywords for this image".

but for myself my benchmark is the adobestock metadata guide, in the prompt I provide several example images from their documentation here https://helpx.adobe.com/stock/contributor/help/titles-and-keyword.html and some in PDF files, titles and keywords provided in their documentation.

I then emphasized to focus on choosing metadata that is likely to be used in search engines, with the 10 most important keywords should be at the beginning and in the title, then some other commands.

For now I'm satisfied with the results, but sometimes if the image is difficult to understand the context, the results can be confusing.

No problem.

Just letting you know - since the "AI" tools aren't actually thinking machines, (and simply sophisticated theft + pattern arrangement) - many times they will give very bad results - so you need to check for that. I'm not familiar with googles "ai" - but it was probably developed similar to the openai (google usually steals many other people's ideas, then passes it off as "search functionality" - differnet topic though)... but anyways - the "ai" systems will stick in "wierd" results... so just make sure you check for that.

Cheers!


« Reply #12 on: July 06, 2024, 11:51 »
0
Hi there! thanks a lot for giving an option to try.

Ive entered the trial key - but it says:

Provide API key

after pressing the Generate button.

I am on PC, Win 11 Home

« Reply #13 on: July 06, 2024, 13:13 »
0
Hi there! thanks a lot for giving an option to try.

Ive entered the trial key - but it says:

Provide API key

after pressing the Generate button.

I am on PC, Win 11 Home

you can make it here

https://aistudio.google.com/app/apikey

« Reply #14 on: July 06, 2024, 15:40 »
0
.... but anyways - the "ai" systems will stick in "wierd" results... so just make sure you check for that.

Cheers!

ignoring your usual off-topic diatribes, just how can his app check for 'weird' results w/o a human manually checking each result?

you say you tried the app.  then you fly a tangential argument about theoretical errors  - what 'very bad results' did you find with THIS app?

it's up to the user to ensure this tool is producing results they consider accurate
« Last Edit: Yesterday at 13:35 by cascoly »

« Reply #15 on: July 06, 2024, 17:17 »
0
impressive - was able to run w/o reading doc

no false ID  - from first test:

correctly found: 
-- Close-up of Hazelnut Tree Catkins in Spring Bloom
-- Scenic view of Alhambra Palace towers and gardens against mountains and blue sky
-- Blooming rosemary sprig. Fragrant purple flowers and green leaves.
-- Steel bridge over a canyon with a river below.


***** problem
while the CSV says -- 'Scenic view of Alhambra Palace towers and gardens against mountains and blue sky'

 the data written to the image is different!
it writes "Scenic View of Granada with Alhambra Palace" - ie, adds Granada but deletes part of description; plus title and description are identical which is an error for DT

another example - metadata in image "Fishing boat deck with colorful buoys and ropes."  but for CSV it says "lobster buoys ..." when they're crab floats (a minor error)
.................
when it stops with the msg "reached trial limit', it doesn't let me save the CSV of what it had captioned

*** Results
Of my first 20 images, 2 could be submitted with no changes. the others need further editing - mostly the location or latin name.

Suggestions:

... add to user account:
   api key
   save a copyright notice & write to image
...  let me define location for a batch
... allow a byline for a batch (for SS editorial) eg "SEATTLE - JUL 6, 2024 - "


« Reply #16 on: July 06, 2024, 17:19 »
0
my concerns:

--need to be able to process more images in trial
-- i keep a xlsx with more detailed caption, etc, so will need to copy from the CSV  to that file

So I'll need to assess how much time i'll save and what the costs are. 

I don't add metadata at a constant rate, especially due to frequent travel, other tasks, etc, so a monthly fee may not be useful - would suggest either a price for a week's use or a much reduced annual fee paid up front.

« Reply #17 on: July 09, 2024, 13:49 »
0
Generation & CSV

when you try to generate more images than the trial limit - cant save CSV for those it had done - ? reached trial limit but doesn't say so until i try to start over with a new selection

CSV important for updating xlsx caption file, otherwise have to capture 1 by 1 from individual images.  caption file is important as i generate only 1 of a series.  since it doesn't provide actual castle names, location, 'AI generated'  etc in most cases, desc & tags need to be updated before applying to rest of a series.

Useful Concepts added:
-"Businessmen in suits struggle through ocean waves, depicting challenge, teamwork, and resilience in the face of adversity."
-"Contemplation of Mortality. Scholar Studies Amidst Symbols of Death and Knowledge"  - hamlet & Yorick skull

Suggestion:
.. allow me to enter additional desc & tags words i want to add to all images of a batch
... .. allow entry of byline (for SS editorial) -- "SEATTLE - Jul 7, 2024"


Problems:
... when you run 2 batches, the second CSV overwrites the first
 
... too many capitals
Scenic View of Arch Bridge Over Canyon River

... set advanced 35 tags, 200 char in advanced, but not saved

minor errors:
"Stealth Fighter Jet on Aircraft Carrier Deck" is actually a Turkish drone - image includes Turkish flag
 "sizzling steak " is a pot roast braising in a cast iron Dutch oven

« Reply #18 on: Yesterday at 14:02 »
0

I'm developing software to automatically generate metadata using the gemini API, it's called Maigen.

You can try it here
Download : https://t.ly/KQvGs

Trial key : MAIGEN_TRIAL_IKN48
Thank you.

...

no longer works - says 'link  has expired'


 

Related Topics

  Subject / Started by Replies Last post
4 Replies
14512 Views
Last post January 02, 2007, 00:52
by scrappinstacy
Automatic DMCA generator

Started by Microbius Image Sleuth

6 Replies
5922 Views
Last post August 09, 2011, 19:56
by TheSmilingAssassin
2 Replies
12914 Views
Last post June 05, 2014, 09:43
by suzbah
3 Replies
6424 Views
Last post April 24, 2017, 15:54
by probe.node
2 Replies
240 Views
Last post March 13, 2024, 08:22
by waitingonthestuff

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors