Hello,

GPT-4 allows you to create a custom version of chatGPT for a specific purpose. I created a custom chatGPT called "Stock Footage Keyword Export". In this chatGPT, I uploaded all the guidelines from various agencies in pdf format, which relate to best practices in keywording, writing titles and descriptions, etc.

I instructed the chatGPT to provide me with 50 keywords and a title for every photo I upload. When I upload a photo from a screen grab, the chatGPT gives me a basic title with many relevant and less relevant keywords. I choose the most relevant keywords and still use other tools such as Wikipedia or SS Keyword tool to find more keywords.

I've attached a screenshot of an example for your reference.

To customize your GPT, I recommend using this comprehensive metadata guide from Adobe Stock https://adobestock.adobe.com/Metadata-Field-Guide.html

Cheers,

Daniel

Thanks for the idea. Have seen today some user on reddit, who linked to some self built key word tools with ChatGPT4.
I am currently considering other areas of possible applications and may take a subscription in future.

Does AI use stop words or ignore them? Does punctuation have any value? Why do I see prompts that have things like f11:10 on the end? What does that mean?

I'd like to know what actually matters, instead of just throwing out words and having some random seed start to make a image that's using some of that and ignoring other parts.

Anyone have a guide?

I've tried a number of times, just for an example, to get SDXL to make "A cow jumping over the Moon" Never one, it puts in a cow, on the moon, over a moon background, all kinds of things, but not a cow, above the moon or jumping over it. DALL-E does better but low quality output.

There is basically a prompt syntax for stable diffusion.

- Type of shot (portrait, full body, from behind, macro, etc.)
- Description of the subject
- Possibly control of the depth of field by means of aperture specification or e.g. "bokeh", "shallow depth of field". Mostly with f1.8 or f2.0. Focal length of objects can also be specified, but in my opinion it has no great influence, rather directly control the portrait type such as Extreme Close Up, Close Up, Medium Shot, Full Shot, etc.
- Additional information e.g. on foreground, background, colors, (rarely also composition, image harmony)
- For portraits, additional information on the facial expressions / gestures of the person (smiling, laughing, sad, angry, etc.)

Weighting works via bracketing.
Square brackets [] reduce the weighting, round brackets () increase the weighting.

Commas and periods play a role in separating information and content. Commas are light and periods are strong separators. However, this is actually very similar to normal punctuation when writing texts.
The same applies to the negative prompts.

With LORAS (fine adjustment models), additional weighting can be applied, e.g. with<lora:sdxl_photorealistic_slider_v1-0:1> the number after :

In addition, the images can be further fine-tuned via the number of steps (iterations) and the guidance scale.
With higher settings, the images generally become more detailed, but the prompts must be formulated very precisely.

Basically, I don't think much of controlling the type of image in the standard model by specifying "photorealistic", "artistic", "surreal", etc.
Instead, you should use this in combination with specially optimized models (see https://civitai.com/models).

I believe that open source Stable Diffusion can beat Midjourney in the long term because of the huge developer community and extreme flexibility.
But it's currently really pain in the ass to find right settings, which deliver good results.

Oh ok, now, thanks to you, I feel like a millionaire! 
as a traditional photographer, of course.

I'm glad to have cheered you up ;-)

Thanks everybody. I've been shooting stock photography for some 15 years now, the main micros, Alamy and Getty Images. I just got really bored with it a few years ago and disgusted at how my income was falling every year. So during the pandemic I started an Etsy store selling fine art prints. Not a huge money make but picking up some slack from lower earning from stock. I recently started a second Etsy store selling "art" (using that term very loosely) from generated AI, MidJourney. Travel posters and other stuff. Those AI images are selling OK, but Etsy isn't a get rich quick scheme for most although some of the stores selling AI art have impressive sales in a short time. Anyhoo, I've been pondering AI stock and thought I'd get some feedback. Thanks again.

I had also thought about an Etsy store, especially I saw that baby posters and other baby print products seem to be quite lucrative. At least the margin seems to be not bad at all.

I've even set up a store and uploaded some designs, but I'm still looking for POD providers that I can integrate.

You do research over all agencies on subject themes that you understand very well.

If for instance, you have a beautiful garden and love gardening and especially caring for roses, then you can do test searches over all agencies what kind of content is available, and what kind of very useful content is missing? If you understand a subject well, you see the missing content very, very easily.

Or you look for a certain region. It can be your home region or maybe your favorite tourist spot you visit every year.

what kind of food images from the region are missing? What kind of typical festivals have not been documented? What kind of favorite resting places of a hike has nobody uploaded�etc..

You are a medical doctor or nurse? Check medical content and you will notice one million things missing�

You are a chef or love to cook? What kind of new trendy food styles has nobody done yet?

You are a carpenter, a body builder, a librarion, an IT guy�look at your genre, what is missing - everywhere?

Often you can reinvent popular genre by just doing something a little different.

The buyers who understand the genre will notice and be grateful.

etc�

The gaps in the agency collections are gigantic, because most people just sort by downloads and then copy the top 20 files as identical as possible.

The libraries are endlessly bloated with duplicates.

Pick a theme you understand well and love and shoot away.

In theory this sounds great and makes sense.
But in praxis your are limited currently by the AI models (you can't generate accurate some topics like medicine stuff, or places it takes to long to find a random seed, which generates a satisfying result).

I also think that most buyers aren't thankful for great content. They just want a fast and cheap result like fastfood.

For example, I am currently trying to generate authentic culture-specific portraits.
The repetitive is sometimes even desired.

I would say this is still the lowest beginner level and can be extended.

I have given ChatGPT some examples of how the prompt should be constructed with a description of the person, culture and landscape in the background.

These are two following generated prompts and generated images in Stable Diffusion.
(raw images are from the first run and must either be adjusted in Stable Difussion by variations or manually retouched after upscaling).

Keywords:
african female, authentic dressed, backdrop sunset savannah

Prompt:
Detailed close up of an African woman adorned with a traditional headwrap, symbolizing humble beauty against the backdrop of an African savanna with a breathtaking sunset. The scene captures the essence of cultural richness and natural splendor, celebrating the woman's grace and connection to her heritage.

Image:
https://ibb.co/brF4dxz

Keywords:
spanish female, authentic dressed, dancing flamenco, backdrop madrid sunset

Prompt:
Detailed Close Up of an elegant, traditionally dressed Spanish woman passionately dancing Flamenco in front of madrid city against the backdrop of a breathtaking sunset. The scene captures the cultural richness and emotive beauty of Flamenco, blending the artistry of dance with the captivating hues of the evening sky.

Image:
https://ibb.co/ZYf4dhW

I currently need about 1 day to generate a certain set-up (approx. 30 to 40 images), sort them out manually and retouch them. That leaves me with about 10 to 15.
I try to generate daily up to 5 max 8 set-ups, so about approx. 50 to max. 120 final images a day. In the meantime I prepare the following prompts.
Currently it is still a very manual and multi tasking task.

But I currently earn a maximum of 10 USD / month with it.
I still have almost 800 images in review on Adobe ...

The problem with hands is, not all hands are flat out, four fingers spread, and a thumb. Some are holding things, some are pointing, some are shaking hands or holding hands. At that point and add variations of how many angles or fingers showing, the AI doesn't understand what a hand consists of, since there's no way to know the exact situation, and how many fingers are supposed to be there? We get the monsters.

Hands are an extremely complex human organ. Extremely flexible, linked to non-verbal language that goes "hand in hand" with facial expressions and body gestures.

AI would therefore have to be taught anatomy, gestures combined with language and emotions.
This is roughly what a human learns over the course of decades through day-to-day communication and social context.

But for simple static AI image generation, you probably only need gigabytes of hand images. But for AI videos ... this might be a very long way.

lol, I saw today that even Yuri offers AI-generated images on freepik.
https://de.freepik.com/search?ai=only&author=61840215&authorSlug=YuriArcursPeopleimages&format=author&last_filter=ai&last_value=only

The funniest thing about it is that he generates just the identical content stuff as thousands of other AI contributors.
There are even better contributors on Adobe.

I wonder if he will still doing classical set productions or switch completely to AI for micro stock production.

Quote from: cobalt on November 19, 2023, 09:34

Yuri gave great examples of how he works, how they do their research and how extensive and authentic the post production is. For instance with tech images that show overlayed computer code, they actually have code written by real software people.

When they do team sports images, they actually design the team clothes specifically for the shooting, they work with a lot of authentic locations etc�it is very, very impressive how much attention to detail goes into his work.

I wonder what his annual production costs were to generate all his microstock photos and clips?

I would say that the costs were astronomically high, but they were worth it in the good old days.

I once found a blog article from the well-known German microstock photographer Robert Kneschke from 2008, in which he writes that he had a seminar with him in cooperation with Fotolia.
The statement from Yuri was that he had earned between 25 and 50k (!) with a single photo shoot with about 100 photos.

In 2013, Kneschke had written another article in which the information was that Yuri had hired over 100 employees with a huge photo studio at the time and already needed 30 months to refinance his photo shoots. That was the reason why he opened his own agency back then.

But in any case, no aspiring amateur would make it that far today up to this pro level because the earnings are only a fraction of what they were back then.

While I don't agree with SuperPhoto on everything (some of it sounds like a conspiracy to me), I have to agree with him that the AI image generators actually use real images to create new ones. This is done by morphing from many learned images.

In 99% of cases you don't notice it, especially with very generic subjects, of which an extremely large number have been used. Here, the morphing makes it unrecognizable and makes you believe that the AI is creating completely new images on its own.

But if you pick out special subjects, e.g. photorealistic wildlife shots of animals, of which there were obviously not that many used for the training, it sometimes happens (at least with Stable Diffusion) that you discover blurred copyright labels in the corner of the image.
So if the AI would generate complete new images than copyright symbols should not appear.
The AI image generators currently have no ability to abstract like a human and create new things by itself. It's only morphing of pictures.

@SuperPhoto
I would argue that a conspiracy on a political level happens not in every corner of the world. I don't know what it's like in the USA. I live in Germany and can't see any political leadership, strategy or conspiracy at all. The country is only living off the substance of the last few decades and is slowly going down the drain.

To be honest, our politicians are completely overwhelmed by such a complex topic.
Checking the input for copyright infringements is totally pointless. It can be hidden.

What matters is the output. Because this determines whether copyright or other infringements occur.
This is exactly where regulation needs to go deep.

It would make sense to regulate the developers in the prompt input. e.g. works of art protected by copyright may not be imitated by the artists through the input, political persons may not be led to deepfakes, etc.

Quote from: stoker2014 on January 21, 2024, 06:41

Where is the representative from shutterstock here on the forum? Maybe we should invite them. Let them tell you why video sales and video income began to decrease.

They closed their forums and made it clear they don�t want to talk to us.

They used to have great team, not anymore.

In their financial statements they keep promoting how they are turning the company to be a �data licensing company� and are already making 20 million a quarter with ai licensing versus 80 million for stock sales which have not increased.

Interesting, Shutterstock signed a 6-year training data agreement with OpenAI.
Since OpenAI has also signed a partner ship with newspaper publishers, such as Axel Springer, it is very likely that OpenAI want to play it safe and have alternate data resources in case they lose the lawsuit against New York Times and some artists.

Interesting that they are obviously using new reasons for rejection, but again they are written in a very generic way.

How many images have been rejected?
Do you have metadata with keywords in it that could be misinterpreted as generative content?

I would possibly edit the image(s) (increase saturation, etc.) and re-upload again. In the past, I've often had success with images being then accepted.

But it's probably best to ask the question in the Discord community.

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.

Just googled the agency. Seems to be a small german agency.
Don't know what to think about them.

They have sent out a lot of copyright warnings.
I think that's their primary source of income and not the actual sale of images.

I would rather recommend the german agency https://www.imago-images.de/. I often see local news in Germany licensing images from them.