Quote from: Unamas on January 16, 2024, 13:16

Quote from: blvdone on January 13, 2024, 09:28

it took 10 years to get 6,500 photos on Adobe Stock, but took only 3 months to add 3,000 AI generated photos on Adobe Stock

I'm sure you spend less time and money on generating AI photos than the real life photos

Much much less.  And I don't need to spend any money to create images in all over the world, or beyond earth too.  So, it's a game changer.  I don't even need to hire/pay my stock models anymore.  That's bad for them.

This is really crazy for people photography.

But I'm pretty sure that for experienced professionals, in the near future, DSLR processors with AI will soon be announced that will perform real-time image enhancements and, for example, pore reduction, digital make-up, etc. for portraits.
In combination with the high native resolution, they would still have a chance on the market.

Otherwise they will be left behind extremely quickly.

Quote from: Andrej.S. on January 17, 2024, 18:53

I found this contributor at Adobe today:
https://stock.adobe.com/ch_de/contributor/209153539/oleg?load_type=author&prev_url=detail

He has almost 315k (!) generated images. What the heck?
But at least he has a good, probably manual quality control.

But still, how do you manage to generate so many images in such a short time? It can only be a huge team, probably from Eastern Europe (Oleg?)
I wouldn't be surprised if they start completely plagiarizing other agencies with Img2Img and Inpaint at some point in the future.

Extremely demotivating.

lol, wow! does he have a special pass on how much he can upload? because I'd say he's uploading 30-40k/month then...

but eh, don't worry about it... focus on yourself. imagine there is just one GIGANTIC contributor that has 35 MILLION "ai" images, because in essence that is kind of what it is like.

lol, yeah since the mass is generated through Midjourney and Stable Diffusion, you're right, it's like one just contributor with million of images. For Adobe it makes probably no difference anyway.

But what I don't understand is why no internal Adobe reviewer have noticed this?
What is the point of such contributors uploading between 2 and 3 thousand slightly different images for a single topic and all of them being accepted? That is mass spamming.
So it is very likely that Adobe simply closes its eyes to contributors who have a high acceptance rate and generate enough new revenue through their mass.

So this is the new work ethic and the modern zeitgeist?
The winner in the end is the one who can just generate the most automatically with a technical advantage without any skills?

I am eagerly awaiting the coming future, in which AI will play just as big a role in other industries. No, actually I don't anymore.

@ spike
I had tried to use Stable Diffusion locally, but my hardware is not sufficient for this, so I am currently using it via an online provider.

What I am currently very interested in is upscaling with Stable Diffusion with Ultrasharp Model in combination with Controlnet tiles. Unfortunately, my VRAM is not sufficient here either and new equipment is currently not profitable due to very low revenues.

Is there a good online provider that charges by time and allows the above-mentioned upscaling combination?

Thanks in advance!

// Update:
Tried today www.thinkdiffusion.com
The provider offers an online Automatic1111 UI with all current available advanced upscale models (e.g. the above mentioned Ultrasharp and Controlnet).
But quite expensive with 0,99 USD / hour, since the 4x upscaling takes about 10 minutes. So you get around 5 - 6 images / hour rescaled.

Also you have to play a lot around to get good results. With the first free trial (17 minutes on rapid server) I got only disappointing results.

Currently I upscale with chainner and the 4xSSDIRDAT model with better results than with 4xUltrasharp on Upscayl or Topaz Photo AI / Gigapixel but for sure a little bit worse than Midjourney or probably with SD Controlnet upscale.
With 4GB VRAM NVIDIA it takes local on Windows just around 6 minutes.

I have had contact via Discord with a contributor who has a bot running 24/7, generating, key-wording and uploading thousands of images. He was trying to find ways to open multiple accounts. As an Adobe Community Expert I advised him of the rules around multiple accounts (they are allowed for different file types, but cannot be for the sake of bypassing upload limits). He then offered to give me 1000 images which I declined!

That's the problem when you open the gates completely.
There are always black sheep who want to exploit it completely to the detriment of everyone involved.
I wouldn't be surprised if Adobe soon introduces an AI image upload limit for all contributors.
I don't understand why they don't adopt Dreamstime's system anyway. Here, the upload limit is set based on the acceptance ratio.
And for new users, I think the upload is limited for only a few hundred images.
This would make the business unprofitable for such people, who want to exploit the system.

Do you know any AI tools that are helpful to reduce the rejection rate of photos and increase sales?

So far I have found the following online tool from everypixel:

https://labs.everypixel.com/demo
https://labs.everypixel.com/ugc-photo-scoring
https://aesthetics.everypixel.com/

There are two models:
One model is said to have been developed with high-resolution stock photos, the other with Instagram photos and additionally with manual evaluation criteria from professional reviewers.

It is supposed to evaluate also the technical details of photos, but my impression is that it primarily evaluates the aesthetics such as composition, color harmony, etc. of images.
Often I personally agree with it's generated ratings, so I would say it's not completely garbage.

However I am currently trying to use it to preselect the AI-generated images from my batches that get the highest ratings.

Yep, noticed it since yesterday.
There were for sure "power users" who used batch processes to automatically generate several thousand images, upscaled and then simply uploaded them.

I think it will become really difficult for actual photo amateurs to improve their photo skills by uploading to stock agencies.

The quality coming in with ai content, even from complete art amateurs, is just so much better.

Real commercial photography will become a domain of professional photographers, like real oil paintings only done by actual masters of the craft.

But I think the value of editorial photography will go up, as will photography with real models and real people, but only if it is done in a highly localized environment or subjects that need to be correct for the genre - medical, engineering etc...also gardening with real flowers and not ai mixed hybrids or real animals and underwater life that is genuine and not ai created.

If you keep this in mind, there is a huge field of content that can be worked on, especially if you add latin names to flora and fauna that should be even more valuable than now.

In my opinion beginner or amateur photographers have no chance at all anymore. They are so far behind that they no longer even have the opportunity to learn what is in demand on the picture market through feedback from sales.
They should just do it for hobby or upload on Flickr, etc., otherwise they will only get frustrated.

I also think that many will switch to editorial photography and the value will rise up when the quality is high since the flooding deep fakes.
Just as authentic shots in special workshops, factories, hospitals, scientific institutions, where not everyone has access.

With AI images, I think the biggest differentiating feature will be the aesthetics and striking features of the images. Anyone can generate images. But few have experience of what really looks aesthetic, how color harmony, composition, etc. play a role.
Often simple images are also bought because they visually match other content (color of the website, subject, etc.) or they are just striking enough for click baiting in (social) online media.

Quote from: blvdone on January 14, 2024, 23:04

The fact that she went from making about $100/month before uploading AI images to $1,000/month in 6 months is mind blowing.  Will she make $10,000/month another 6 months later?  Very unlikely.  But I think she can reach $3,000/month in a year or two if she keeps adding quality AI images like she's been doing.  This kind of stuff never happened.  It's the only positive I've seen in the last 5 years in microstock industry for contributors.

https://youtu.be/F6fSIbMpri8?si=Zj8ag7628oTYYE8w

True, there are probably more santa claus images doing better than hers. My point was - for "santa claus" (while of course not every image is a santa claus), of the 425,000 results - basically you'd need to be in the top 10 or top 20 to see any 'significant' financial results.

Not sure why "her" santa claus was chosen over 425,000 other santa clauses (if I knew that, I could consistently produce a lot of different images that do well) - but I'd say it is a combination of luck, timing, happening to stumble upon the right keywords/etc... in other words - not necessariliy easy to duplicate. (I made some santa claus images that I thought were quite good, and for the 3-4 hours involved in coming up with concepts, making them, titling, keywording, upscaling, editing, cropping, fixing, categorizing, submitting, - I think I made... let me check... looks like about $5 (about $1/hour)... Of course, $5 is better than nothing, and of course there is the possibility for future residual income - so I am of course grateful for that... but... there are certainly a lot of santa clauses...

Plus - when she (and not just her) makes videos of "OMFG! look how I made SOOOO much money doing VERY little work, like and subscribe my video channel so I can make MORE money!!!" - it gets lazy people coming out of the woodwork to try and get rich quick...

If things stayed the same - then yes, maybe she'd get to $3k/month... I highly doubt it though - because quite possibly there will be 10x the contributors a year from now "inspired" by her get rich quick videos, + also competing platforms/sites/tech/etc...

It's "possible", but not sure if it is likely...

But again, go for it if you wish. I find it interesting, just saying it may not necessarily be 'easy' or the 'get rich quick' path you are looking for. Certainly is possible though, obviously other people have done it.

You're right in that respect.
You need almost everything to be successful today.
Timing, market analysis, good keywording, skill and luck and perhaps even online promotion / advertising via Instagram, etc.

If you follow her channel, you'll see that she really invests a lot of time. She either works as a graphic designer or is employed part-time, where she can do this even work at (face time job) or has time from midday until the evening.

If you do this alongside your main job, you have to concentrate extremely hard on the essentials. You hardly have time to "practice" what sells well on the market.
I'm currently first trying to read up on the technical side of things to get the most out of it (chaiNNer, ControlNet Tiles, etc.) and then find market niches or at least to pick out motifs that are in demand and cover a very wide range of buyers.

@ blvdone
Out of interest, do you use Topaz Gigapixel AI or Topaz Photo AI?

I felt like I got better results with the last one.
The files are also processed more slowly and the file size is larger at the end.

@all
Have anyone tested https://magnific.ai/?
Looks promising, but probably the example images are just cherry picked.
Also extremely expensive as subscription.

I'm probably will playing more around with Upscayl since I have seen you can import and own use models from github:
https://github.com/upscayl/upscayl/wiki/%F0%9F%96%A5%EF%B8%8F-Model-Conversion%E2%80%89%E2%80%93%E2%80%89Get-more-models!

Thank you for the linked video!

These should now be very small details that can lead to a rejection. From my point of view, extensive retouching makes no economic sense.
It's probably better from the start to focus on motifs that deliver a good result directly, which can be retouched with minimal effort.

For example, I imagine interiors to be also a very difficult subject.

@
I wouldn't bet on it in the long term.
Nvidia is a heavyweight cooperation partner that brings a lot of technical and theoretical know-how in the software/hardware industry and model development of AI.
NVIDIA Picasso for example seems also to generate 3D models and has a web integration of user applications like photoshop, etc.
https://www.nvidia.com/de-de/gpu-cloud/picasso/

If they manage to generate the 3D objects photorealistically, I see even more potential here, because you can generate entire scenes in which you can adjust every detail as desired (think of product visualizations, for example, in which you can rotate the objects or move the light source as desired).

//Edit:
It is interesting to note that Cuebric has been developed with Nvidia for products for film production.
Looks extremely promising there.
https://cuebric.com/disguise

@ cobalt

As you already have more experience, have you already successfully reuploaded images, which were rejected on the first?
Have you just downscaled them or run denoise, etc. procedures?

I might try uploading smaller batches with differently edited images soon, from which I can learn what is paid more attention to. However, there is probably still the problem of the dispersion of reviewers.

If Adobe had a detailed guide with sample images to learn from, you could build automated actions for mass editing in Photoshop. I mean their current guidelines are just to generic written.

https://helpx.adobe.com/stock/contributor/help/generative-ai-content.html
https://helpx.adobe.com/stock/contributor/help/photography-illustrations.html
https://helpx.adobe.com/stock/contributor/help/the-review-process.html

Interesting is that they seem in theory also evaluate the commercial value, title and keyword accuracy, aesthetic properties, uniqueness, what would be really great, but in practice only technical reasons are mentioned.

I would say when your photos/videos sell, those will move up the keyword search ladder.  So, if an image/clip sell a few times, it's likely to sell well for at least like 3 years unless it's timely subject.  So, I think my AI photo sales will only get better as many of them make first sales and show up higher in search.  I'm not done adding AI images although the pace has slowed down now.  Probably by the end of this year, majority of my photo sales will be from AI images.

Yes, it definitely makes sense to reduce the pace. Otherwise, after a few months of piecework, you'll start to just hate it.
I wonder if it makes sense instead to invest some more time creating Lightroom or Photoshop presets that overlay the Midjourney look and create more natural colors for faking real photos, which can stand looking unique out from the mass. 

Is there really no adjustment prompt or filter for creating more raw looking images, which you would postprocess in the further step?

Too bad that Adobe doesn't differentiate between the reasons for rejection.

Another trigger for strictness could also be that the reviewers are now only looking for the quality of Midjourney V6, which was released at the end of December and seems to have a considerable jump in quality:
https://mid-journey.ai/midjourney-v6-release/

Which AI engine do you use?
I'm too stingy for 48 to 60 bucks Midjourney's pro plan with stealth mode. I currently use a cheap provider with stable diffusion engine and various models/loras, who offers a private mode for 15 bucks / month without generative limitations.