MicrostockGroup
Microstock Photography Forum - General => General Stock Discussion => Topic started by: waitingonthestuff on February 19, 2024, 12:22
-
I have to say I am tempted to look at the AI generated Tit/edescription and keywording that some sites use as it would save so much time.
Does anyone on here go down that route? I just find manually doing (even with something like Chat GPT) it is so, so slow.
-
AI generated keywords are not great, saving time on meta data is the wrong way to go...
I've tested AI generated keywords with around 1000 images before developing a custom solution that is way better (based on real stock sites keywords data)
you should try it:
https://upstock.guru/keyword_checker
Just make sure the titles contain the top keywords you generate for best results.
-
Thanks. Will have a look
-
i've stared using ChatGPT for keywording with great results - it gives many relevant tags i wouldn't have thought of. I enter my image description, then edit the results
-
Neither. AI sucks for keywording, usually gives two-word phrases as keywords, and is more or less useless.
I also don't do it manually.
-
I write my own keywords, with help from Shutterstock's tool. But if using ChatGPT, you can just ask it to give you single word keywords, comma separated, all on one line. If you don't ask it to do that it will usually give you two word phrases with no commas, and each on a new line.
-
I use this, for me there's no match: www.udream.ai (http://www.udream.ai)
It's an AI keyworder tool. I've tried many but this one is on another level. What impressed me is that it can understand the concepts within the images.
It also writes a minimum 5 words title (for SS), different and longer description (for DT) and puts the first 5 keywords in relevance order (for AS).
The first 50 credits are free so it's easy to test (1 credit = 1 image).
It's not instantaneous, but man it's good (around 10 seconds per image).
Much cheaper than other crappy keyworders also, about 2 cents per image with the smallest credit pack.
Basically I only use this now, it speeds up things so much.
-
Pixify AI Keywording Tool: Available at https://pixify.io/ai-keywording-tool, (https://pixify.io/ai-keywording-tool,) this tool offers the capability to create keywords, titles, and descriptions for stock photos rapidly. It employs image recognition to extract keywords from uploaded images and then uses those keywords to generate titles and descriptions.Their price is also very competitive (1 file unit means max 50Mb, so for photos it's 1:1)
Xpiks: They offer an auto-keywording plugin that seems quite efficient in generating relevant keywords for images.
Imstocker Vision: Their tool (https://vision.imstocker.com/en/keywords (https://vision.imstocker.com/en/keywords)) is another interesting option in the realm of automated keywording and metadata generation. Currently the title and subscription is not supported
Excire Foto: They have a tool (https://excire.com/en/excire-foto/ (https://excire.com/en/excire-foto/)) that focuses on intelligent photo management, including AI-powered keywording. Currently the title and subscription is not supported, also the keywords are very limited basic words only
IMatch: Another potential competitor worth mentioning is IMatch by photools.com (https://www.photools.com/help/imatch/app-auto-tagger.htm (https://www.photools.com/help/imatch/app-auto-tagger.htm)). It also supports 3rd party AI providers.
https://www.descripshot.com (https://www.descripshot.com) Nice results, but not the best value
https://www.phototag.ai/ (https://www.phototag.ai/) Very good quality and value, also supports API
-
I do it manually and very accurately.
-
I do it manually and very accurately.
it's not about accuracy - i can do that - it's that ai brings up further 'accuracies' i never considered
-
Xpiks: They offer an auto-keywording plugin that seems quite efficient in generating relevant keywords for images.
Xpiks offers descriptions too, btw.
In general, I always recommend to take the AI generated ones as a starting point. If you want to improve your sales, treat metadata seriously! It was discussed multiple times in Xpiks blog.
-
Pixify AI Keywording Tool: Available at https://pixify.io/ai-keywording-tool, (https://pixify.io/ai-keywording-tool,) this tool offers the capability to create keywords, titles, and descriptions for stock photos rapidly. It employs image recognition to extract keywords from uploaded images and then uses those keywords to generate titles and descriptions.Their price is also very competitive (1 file unit means max 50Mb, so for photos it's 1:1)
Xpiks: They offer an auto-keywording plugin that seems quite efficient in generating relevant keywords for images.
Imstocker Vision: Their tool (https://vision.imstocker.com/en/keywords (https://vision.imstocker.com/en/keywords)) is another interesting option in the realm of automated keywording and metadata generation. Currently the title and subscription is not supported
Excire Foto: They have a tool (https://excire.com/en/excire-foto/ (https://excire.com/en/excire-foto/)) that focuses on intelligent photo management, including AI-powered keywording. Currently the title and subscription is not supported, also the keywords are very limited basic words only
IMatch: Another potential competitor worth mentioning is IMatch by photools.com (https://www.photools.com/help/imatch/app-auto-tagger.htm (https://www.photools.com/help/imatch/app-auto-tagger.htm)). It also supports 3rd party AI providers.
https://www.descripshot.com (https://www.descripshot.com) Nice results, but not the best value
https://www.phototag.ai/ (https://www.phototag.ai/) Very good quality and value, also supports API
Small update: now IMS Vision support generating titles and description too: https://imstocker.com/en/blog/152-ims-vision-desc-gen (https://imstocker.com/en/blog/152-ims-vision-desc-gen)
-
i've stared using ChatGPT for keywording with great results - it gives many relevant tags i wouldn't have thought of. I enter my image description, then edit the results
I use ChatGPT only for bad content, when it’s a shame to throw it away and can be upload. ChatGPT tags are always bad, but better than nothing.
-
AI generated keywords are not great, saving time on meta data is the wrong way to go...
I've tested AI generated keywords with around 1000 images before developing a custom solution that is way better (based on real stock sites keywords data)
you should try it:
https://upstock.guru/keyword_checker
Just make sure the titles contain the top keywords you generate for best results.
I didn't like your tool. It offers very few images when searching, the actual ones are always not what I’m looking for, and the tags are very weak. It’s better to search for images on stock sites yourself, copy tags from there, and then create a general list.
You need to refine your model. Need more pictures. Focus on shutterstock and display top images or videos from this site. I should be able to select not 2 pictures, but 5 or 10. Therefore, you should display both 20 and 50 pictures.
-
I didn't like your tool. It offers very few images when searching, the actual ones are always not what I’m looking for, and the tags are very weak. It’s better to search for images on stock sites yourself, copy tags from there, and then create a general list.
You need to refine your model. Need more pictures. Focus on shutterstock and display top images or videos from this site. I should be able to select not 2 pictures, but 5 or 10. Therefore, you should display both 20 and 50 pictures.
You have no idea how it works, there is no "model",
if you're not getting enough results, you're not using it correctly, it always returns enough images to select from
try a single keyword and go from there, it basically does what you describe but automatically + refines the results and sorts the keywords by retime keyword data and positions, doing that by hand will take you x10 longer
-
and sorts the keywords by retime keyword data and positions,
I do not need it.
At best, you can choose 1 or 2 images in your model. On average, everything it shows is not suitable. Therefore, your model should display not 10 images, but much more. That's why I'm telling you that you should connect to the search results on shutterstock. In general, the way your model works does not suit me.
-
I do not need it.
At best, you can choose 1 or 2 images in your model. On average, everything it shows is not suitable. Therefore, your model should display not 10 images, but much more. That's why I'm telling you that you should connect to the search results on shutterstock. In general, the way your model works does not suit me.
again with the "model", it doesn't mean what you think it means
you don't have to use it, please don't
-
I do not need it.
At best, you can choose 1 or 2 images in your model. On average, everything it shows is not suitable. Therefore, your model should display not 10 images, but much more. That's why I'm telling you that you should connect to the search results on shutterstock. In general, the way your model works does not suit me.
again with the "model", it doesn't mean what you think it means
you don't have to use it, please don't
So remove this topic from this forum. Your model will not suit stockers.
-
So remove this topic from this forum. Your model will not suit stockers.
Does your username contain the year of your birth date?
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
-
I do not need it.
At best, you can choose 1 or 2 images in your model. On average, everything it shows is not suitable. Therefore, your model should display not 10 images, but much more. That's why I'm telling you that you should connect to the search results on shutterstock. In general, the way your model works does not suit me.
again with the "model", it doesn't mean what you think it means
you don't have to use it, please don't
So remove this topic from this forum. Your model will not suit stockers.
you dont get to decide who can post here!
you've given your opinion without providing any actual DATA to support it
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
-
I have tested using "ai" keyword suggestions, but I've found I have to manually verify all the keywords myself because:
a) sometimes they are totally unrelated/garbage output
b) sometimes they include trademarked terms
c) sometimes they don't produce enough results
d) sometimes the results aren't really what buyers are looking for (i.e., say you had a picture of "oranges", buyers might actually be looking for "produce section of grocery store", not "oranges").
e) And the "ai" (chatgpt) has an annoying thing where - even if you revise the prompt and it "seems" to work - it will randomly change it at a random generation location and mess up the output so you have to revise the prompt again...
So...
"Eh", - "maybe" it saves a little bit of time - but there is a lot of work involved in making sure it is accurate.
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
but location alone can't tell what's in the background,esp'ly for landscapes - eg a view from a pass may be looking in any of 4 directions with different mountains to be named.
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
but location alone can't tell what's in the background,esp'ly for landscapes - eg a view from a pass may be looking in any of 4 directions with different mountains to be named.
The ask was merely to obtain the location from the location metadata in the first place (in lieu of manually as you say you do).
From there, most would expect the location helps set the context for the rest of the keywords with respect to what’s in the picture..
eg. A picture of a big metal tower in Paris, or a big canyon in Arizona, or a large waterfall in Zambia. With location, a good guess could be made as to what each description is referring to.
Another example would be of wildlife where almost all Id forums ask for location (and date) because different species can otherwise look identical.
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
but location alone can't tell what's in the background,esp'ly for landscapes - eg a view from a pass may be looking in any of 4 directions with different mountains to be named.
The ask was merely to obtain the location from the location metadata in the first place (in lieu of manually as you say you do).
From there, most would expect the location helps set the context for the rest of the keywords with respect to what’s in the picture..
eg. A picture of a big metal tower in Paris, or a big canyon in Arizona, or a large waterfall in Zambia. With location, a good guess could be made as to what each description is referring to.
Another example would be of wildlife where almost all Id forums ask for location (and date) because different species can otherwise look identical.
but there can be different species in the same location - eg crows vs ravens, or multiple species of antelope, etc
i'd rather enter the actual information rather than have the AI 'guess' what i'm photographing - if location is 'pisa' it cant know whether image is the tower, the cathedral or a zoom of the coat of arms on a wall, or a detail of columns or sculptures; in Nuremberg several cathedrals are visible from one place. better to use a AI that does an image search in the first place
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
but location alone can't tell what's in the background,esp'ly for landscapes - eg a view from a pass may be looking in any of 4 directions with different mountains to be named.
The ask was merely to obtain the location from the location metadata in the first place (in lieu of manually as you say you do).
From there, most would expect the location helps set the context for the rest of the keywords with respect to what’s in the picture..
eg. A picture of a big metal tower in Paris, or a big canyon in Arizona, or a large waterfall in Zambia. With location, a good guess could be made as to what each description is referring to.
Another example would be of wildlife where almost all Id forums ask for location (and date) because different species can otherwise look identical.
but there can be different species in the same location - eg crows vs ravens, or multiple species of antelope, etc
i'd rather enter the actual information rather than have the AI 'guess' what i'm photographing - if location is 'pisa' it cant know whether image is the tower, the cathedral or a zoom of the coat of arms on a wall, or a detail of columns or sculptures; in Nuremberg several cathedrals are visible from one place. better to use a AI that does an image search in the first place
First, if the image has gps coordinated as implied by @DaLiu comment, then no guessing is needed. Reverse geocode provides a location.
Second, the AI tools I’m referring to take an image as input (for ChatGPT you need a subscription or use a 3rd party tool that uses the paid API). Thus, location would merely aid it in its description and would not dictate it.
-
The problem with all these AI tools, they don't provide the location, keywords and description about location which for me is very important, even tho the test image had the coordinates in the picture, Udream (the only one tested) didn't use any of that info to provide the location, so useless for me.
when i use chatgpt i ask for tags from my description which does include the location
The way I understood the comment, the ask is for the tool to figure out the location by itself which it can do if the image contains the gps metadata (which is more reverse geocoding than AI) or if the image is of a recognizable location.
but location alone can't tell what's in the background,esp'ly for landscapes - eg a view from a pass may be looking in any of 4 directions with different mountains to be named.
The ask was merely to obtain the location from the location metadata in the first place (in lieu of manually as you say you do).
From there, most would expect the location helps set the context for the rest of the keywords with respect to what’s in the picture..
eg. A picture of a big metal tower in Paris, or a big canyon in Arizona, or a large waterfall in Zambia. With location, a good guess could be made as to what each description is referring to.
Another example would be of wildlife where almost all Id forums ask for location (and date) because different species can otherwise look identical.
but there can be different species in the same location - eg crows vs ravens, or multiple species of antelope, etc
i'd rather enter the actual information rather than have the AI 'guess' what i'm photographing - if location is 'pisa' it cant know whether image is the tower, the cathedral or a zoom of the coat of arms on a wall, or a detail of columns or sculptures; in Nuremberg several cathedrals are visible from one place. better to use a AI that does an image search in the first place
...
Second, the AI tools I’m referring to take an image as input (for ChatGPT you need a subscription or use a 3rd party tool that uses the paid API). Thus, location would merely aid it in its description and would not dictate it.
chatgpt DOES not require payment - you can use it for free directly
-
chatgpt DOES not require payment - you can use it for free directly
As I was attempting to clarify, it sounds like your usage is the free tier which does not allow you to upload an image.