Author Topic: AI metadata generation options? (Read 1283 times)

gnirtS · « **on:** April 10, 2025, 22:53 »

OK let's be honest here. I hate keywording. I loathe it with every fibre of my being. It takes me far longer than editing photos of videos. I often leave edited stuff sat for weeks in a folder purely because i can't stand the prospect of keywording it for upload (just as im doing now for example).

Recently it seems to be a massive influx of heavily marketed (lots are spammed everywhere) AI keywording tools. Often these have dubious to no privacy or data security policies and so on.

Are any of these actually any good? Will they produce useful metadata i can use with only minor tweaking then upload? Or are they all producing as the old ones did overly generic results?

Are any worth using? Specifically, i want something to JUST generate keywords (and descriptions maybe). I don't want to use it to upload content. I just want the keywords to stuff into LR or a CSV.

I've had a little play using Google Gemini (not ChatGPT as you can only upload about 3 images a day) and "fed" it on the various submission guides. It produces OK results that still need tweaking but OK as a starting point. Its not capable of searching for similar images on stock sites and scraping keywords to start with though. The main issue i have is amnesia. It'll occasionally forget i told it base verbs, no plurals or return 80 or 20 keywords instead of 30-40 ish and so on. I cant get consistency out of it.

Is it worth me trying to tweak my current Gemini model or trying one of the other services? If so, which one? I want one that isnt spammed with referral codes to every FB group in the world.

Injustice for all · « **Reply #1 on:** April 11, 2025, 03:06 »

I can't help you because I don't use AI to automate the indexing process.

from what I've seen you can get decent results,but generally you still have to review the title and check the keywords,so I find it easier to write everything myself.

gnirtS · « **Reply #2 on:** April 11, 2025, 04:43 »

Quote from: Injustice for all on April 11, 2025, 03:06

I can't help you because I don't use AI to automate the indexing process.

from what I've seen you can get decent results,but generally you still have to review the title and check the keywords,so I find it easier to write everything myself.

Review and adjust is fine. Im trying to avoid the 20-30 mins of searching stock sites, collating keywords, adding them from my own database, removing duplicates and so on and putting them into priority order.
Im trying to streamline a process i absolutely detest.

Injustice for all · « **Reply #3 on:** April 11, 2025, 06:46 »

I simply use content that I already have in my port to copy keywords from content that is similar in subject or concept,then i review them and add some more pertinent to the subject.

not always but generally,when I upload content to AS I upload 2 copies,I send one,and I keep one among the contents to be sent with the indexing already copied from the same content sent,so in case of rejection,if I want to make changes and send again,I just copy the indexing data from the content I left.

If I want to send this content to other agencies in the future,I simply copy the indexing from the content that is in my AS port.

This is because if the content is rejected it is still possible to copy the title but not the keywords.

gnirtS · « **Reply #4 on:** April 11, 2025, 06:54 »

Quote from: Injustice for all on April 11, 2025, 06:46

I simply use content that I already have in my port to copy keywords from content that is similar in subject or concept,then i review them and add some more pertinent to the subject.

not always but generally,when I upload content to AS I upload 2 copies,I send one,and I keep one among the contents to be sent with the indexing already copied from the same content sent,so in case of rejection,if I want to make changes and send again,I just copy the indexing data from the content I left.

If I want to send this content to other agencies in the future,I simply copy the indexing from the content that is in my AS port.

This is because if the content is rejected it is still possible to copy the title but not the keywords.

My system isnt overly different - i keep an xls with the common topics and themes to copy/paste. I dont like to mix it up though so im not reusing them all the time. The issue is when i have a new subject/topic/place i need to start from scratch and it takes ages. Its this im trying to shortcut.

Injustice for all · « **Reply #5 on:** April 11, 2025, 07:01 »

upload an image to SS and use the keyword tool,then simply copy and paste wherever you want.

gnirtS · « **Reply #6 on:** April 11, 2025, 09:09 »

Quote from: Injustice for all on April 11, 2025, 07:01

upload an image to SS and use the keyword tool,then simply copy and paste wherever you want.

Thats a tiny part of whats needed. Get some keywords off that, they're generic. After that need to add customisations to make it unique, adjust priority and so on.

My general process involves the SS keyword generator, AS similar search, then add individual customisations, trim them down then prioritise. Takes ages and hate it. Ive hated it for the best part of 15 years. It makes uploading any new topic offputting.

Andrej.S. · « **Reply #7 on:** April 11, 2025, 13:37 »

Lol, anyone who still writes their own keywords in 2025 is no longer up to date. Especially the good keywords require the most work.

The AI beats 99% of human keyworders. So just use it. I've already mentioned here on the forum one good provider.

Faustvasea · « **Reply #8 on:** April 11, 2025, 14:31 »

I use a tool which I can generate up to 2000 images a day with no cost for images. And different tool for video. Works for me.

I don’t intend to invest too much time into this. Especially if the reward is less than 10 pence

cascoly · « **Reply #9 on:** April 12, 2025, 15:08 »

i have thousands of images, many years old, waiting to be meta'd

i've been using visaulmind.ai which does a great job for about 2-3 cents / image. (you can try it for free on 30(?) images) you can append keywords or add to descriptions before submitting (eg adding a byline for editorial, or location when it isn't obvious. but VM has identified the rhine river by a castle in the background or accurately named the interiors of Istanbul mosques. and Even translates from some greek or latin inscriptions on mosaics. it correctly identifies plants & animals. for simitars they actually write different descriptionsi. in a recent example of a poster from last week's protest "only you can prevent fascist liars ", it understood it as a reference to smokey bear!

i can set up 200-300 images in a few minutes & it runs in background - i then select for various post-processing (several thousand) & resizing. i've added >2500 images to SS, >1000 to AS over last 3 months and i can also update my excel captions sheets by appending the resulting csv (these sheets allow me to locate any of 50K archived images by keyword in at most a few minutes)

for free services like chatgpt, you need to consider the time it takes to prompt or upload & you'd still have to add the result to your image meta.i'm not aware of any free services that process in bulk & update the image meta & provide a csv

gnirtS · « **Reply #10 on:** April 12, 2025, 22:19 »

Quote from: cascoly on April 12, 2025, 15:08

i have thousands of images, many years old, waiting to be meta'd

i've been using visaulmind.ai which does a great job for about 2-3 cents / image. (you can try it for free on 30(?) images) you can append keywords or add to descriptions before submitting (eg adding a byline for editorial, or location when it isn't obvious. but VM has identified the rhine river by a castle in the background or accurately named the interiors of Istanbul mosques. and Even translates from some greek or latin inscriptions on mosaics. it correctly identifies plants & animals. for simitars they actually write different descriptionsi. in a recent example of a poster from last week's protest "only you can prevent fascist liars ", it understood it as a reference to smokey bear!

i can set up 200-300 images in a few minutes & it runs in background - i then select for various post-processing (several thousand) & resizing. i've added >2500 images to SS, >1000 to AS over last 3 months and i can also update my excel captions sheets by appending the resulting csv (these sheets allow me to locate any of 50K archived images by keyword in at most a few minutes)

for free services like chatgpt, you need to consider the time it takes to prompt or upload & you'd still have to add the result to your image meta.i'm not aware of any free services that process in bulk & update the image meta & provide a csv

Cheers i might give them a look. My workflow is similar to yours with a master XLSM of keywords as well i guess.

Im having some luck creating Gems from Gemini (because its free) and trained it with the AS and other docs and some other criteria. Its single image currently but i could Python batch it if needed.

Trying to decide if its worth me continuing with my own attempt or pay someone else to further reduce my income(!).

Ill use the 20 credits to do a compare with it v mine.

Example of mine spitting out (with obvious debug stuff to help me fine tune).

Code: [Select]

[b]**Image Analysis:**[/b]
The image is a close-up underwater photograph featuring a Pink Skunk Clownfish (likely *Amphiprion perideraion*). The fish, characterized by its pinkish-orange body and distinct white stripe behind the eye, is nestled amongst the green, bulbous tentacles of its host sea anemone. The setting is clearly a marine environment, suggesting a coral reef habitat typical for this symbiotic relationship. The composition focuses tightly on the fish's head and upper body, highlighting its interaction with the anemone. Key themes include marine wildlife, symbiosis, underwater ecosystems, and natural habitats.

[b][u]**Keyword Analysis and Grouping:**[/u][/b]

[b][i]* **Top 10 Keywords:** [/i][/b]
The most important keywords capture the specific subject, its scientific name, its symbiotic partner, the general category, the environment, and the core concept. These are: `pink skunk clownfish`, `Amphiprion perideraion`, `sea anemone`, `clownfish`, `underwater`, `symbiosis`, `coral reef`, `marine life`, `wildlife`, `fish`. These are prioritized as they are the most direct and searchable terms related to the image content.

[i][b]* **Descriptive Keywords:**[/b][/i] 
These specify the type of fish, anemone, and general animal category, along with prominent colors and visual attributes. Examples: `anemonefish`, `invertebrate`, `animal`, `green`, `pink`, `orange`, `tropical`, `vibrant`, `colorful`.

[i][b]* **Activity/State Keywords:** [/b][/i]These describe what the subject is doing or the relationship depicted. Examples: `hide`, `shelter`, `symbiotic relationship`.

[i][b]* **Location Keywords:**[/b][/i] These define the setting and environment. Examples: `ocean`, `sea`, `marine environment`, `natural habitat`, `Indo-Pacific`, `home`.

[i][b]* **Conceptual Keywords:**[/b][/i] 
These describe the broader themes, ideas, or feelings the image evokes. Examples: `nature`, `biodiversity`, `ecosystem`, `habitat`, `protection`, `safety`, `marine biology`, `harmony`, `fragility`.

[i][b]* **Technical Keywords:**[/b][/i] These relate to the photographic style and composition. Examples: `close up`, `macro`, `underwater photography`, `natural light`, `nobody`, `no people`. Technical keywords are placed last as they are generally less critical for initial search relevance than subject and concept keywords.

--- ---

Quote

**SUMMARY**

Image analysis: Close-up underwater shot of Pink Skunk Clownfish (Amphiprion perideraion) sheltering in green tentacles of host sea anemone, illustrating symbiosis in coral reef habitat. Focus on marine wildlife and nature.

TITLE 1: Pink Skunk Clownfish in Green Sea Anemone [41]
TITLE 2: Clownfish Amphiprion Perideraion Hides [38]
TITLE 3: Vibrant Anemonefish Among Green Tentacles [42]

DESCRIPTION 1 (Revised): Pink skunk clownfish Amphiprion perideraion nestled safely in green sea anemone. [76]
DESCRIPTION 2 (Revised): Close view pink skunk clownfish refuge in green tentacles host sea anemone. [73]
DESCRIPTION 3 (Revised): Marine wildlife Amphiprion perideraion fish symbiosis green anemone home. [69]

KEYWORDS: pink skunk clownfish, Amphiprion perideraion, sea anemone, clownfish, underwater, symbiosis, coral reef, marine life, wildlife, fish, anemone, nature, habitat, protection, animal, ocean, sea, tropical, vibrant, colorful, green, pink, orange, hide, shelter, invertebrate, anemonefish, biodiversity, ecosystem, home, safety, natural habitat, marine biology, Indo-Pacific, close up, macro, underwater photography, natural light, nobody, no people [40]
--- ---

Do you require any keyword variations yes/no?

cascoly · « **Reply #11 on:** April 13, 2025, 14:40 »

Quote from: gnirtS on April 12, 2025, 22:19

Cheers i might give them a look. My workflow is similar to yours with a master XLSM of keywords as well i guess.

I'm having some luck creating Gems from Gemini (because its free) and trained it with the AS and other docs and some other criteria. Its single image currently but i could Python batch it if needed.

looks like you still need to transfer results to images manually? for 100 images that eats time.

my captions xlsx contains more than keywords -- for each image, i store description and keywords. key to my system is renaming all images to YYMMDD-{seq}, so can quickly find any image by date. details are at
https://cascoly-images.com/building-a-microstock-tracking-system-part-1

VM edits the meta and provides a csv of all images

gnirtS · « **Reply #12 on:** April 14, 2025, 06:47 »

Quote from: cascoly on April 13, 2025, 14:40

Quote from: gnirtS on April 12, 2025, 22:19

Cheers i might give them a look. My workflow is similar to yours with a master XLSM of keywords as well i guess.

I'm having some luck creating Gems from Gemini (because its free) and trained it with the AS and other docs and some other criteria. Its single image currently but i could Python batch it if needed.

looks like you still need to transfer results to images manually? for 100 images that eats time.

my captions xlsx contains more than keywords -- for each image, i store description and keywords. key to my system is renaming all images to YYMMDD-{seq}, so can quickly find any image by date. details are at
https://cascoly-images.com/building-a-microstock-tracking-system-part-1

VM edits the meta and provides a csv of all images

I use a different workflow video v stills.

Video i have a macro enabled master XLSM that i paste in details and it'll spit out a CSV compatible with SS,AS,P5,BB automatically which i then just upload to the service.

Stills are different as i specifically want each image in LR to contain the title, description and the keywords in case i ever need that specific image again (or rarely, start a new stock site membership).

I could get macros to pump external data into LR to populate the list but not sure its worth it.

Injustice for all · « **Reply #13 on:** April 14, 2025, 07:48 »

Quote from: Faustvasea on April 11, 2025, 14:31

I use a tool which I can generate up to 2000 images a day with no cost for images. And different tool for video. Works for me.

I don’t intend to invest too much time into this. Especially if the reward is less than 10 pence

Are you sure these free AI models producing content that is allowed for commercial use?

Faustvasea · « **Reply #14 on:** April 14, 2025, 08:26 »

Quote from: Injustice for all on April 14, 2025, 07:48

Quote from: Faustvasea on April 11, 2025, 14:31
I use a tool which I can generate up to 2000 images a day with no cost for images. And different tool for video. Works for me.

I don’t intend to invest too much time into this. Especially if the reward is less than 10 pence

I use google gemini to generate metadata for my images on that tool. And use different website to generate Metadata for my videos. Works pretty good for me.

Are you sure these free AI models producing content that is allowed for commercial use?

Pathfinder · « **Reply #15 on:** April 15, 2025, 05:48 »

Hi everyone!
We’ve just launched and would love to hear your feedback on the tool’s usefulness and what features you need. We also have a referral reward system for collaborators!
newbielink:https://craftflow.gumroad.com/l/Descriptor-AI-powered-Tool [nonactive]

Descriptor is a web-based tool that allows you to quickly generate metadata in one click. Whether you’re a photographer, illustrator, video creator, or digital artist, this tool helps you save time by automatically generating titles, descriptions, and keywords based on image analysis. No special skills are required—just upload your files, and the AI will do the rest! You can also add custom keywords, choose your preferred file format, and download a ready-to-use CSV file for platforms like Adobe Stock, Canva, and Shutterstock.

P.S. First readers get 100 free credits with the code – enjoy! 🙂
A9X2SXGC6S62LME9

WaterView · « **Reply #16 on:** April 15, 2025, 15:20 »

I have been getting meta data ideas for free from

https://www.pixify.io/studio/free-ai-keywording-tool

I upload a medium sized png thumbnail, usually 5% of actual size of the pic. Rarely it takes a 10% of size of pic for it to clearly see it.
Within a minute or so it brings me back title, description and keywords.
You can choose standard or optimized for Getty/Istock.
Then I customize it to make it even better and more exact.
My fave site so far. Helps save me time and come up with more diverse metadata

MicrostockGroup Sponsors

Author Topic: AI metadata generation options? (Read 1283 times)

gnirtS

Injustice for all

gnirtS

Injustice for all

gnirtS

Injustice for all

gnirtS

Andrej.S.

Faustvasea

cascoly

gnirtS

cascoly

gnirtS

Injustice for all

Faustvasea

Pathfinder

WaterView

Related Topics

Sponsors

Microstock Poll Results

Sponsors