for 3400 files that is a very good rank

i owe my good holiday rank to my old istock bestsellers. they paved the way.

at least one advantage from the good old days.

and a visible reminder of what is possible if customers like your files. now I just need useful content for the rest of the year.

That is really an excellent result, probably much higher than what I will have when I drop back down in January.

Now, where will you be when you have 12k files?

I like the idea of combining a genre with different professions. That will give you a lot of niches.

I was looking at country specific "flavors or vibes". For instance in the US christmas is often a very colorful, loud happy party family and friends gathering.

In Germany it is a lot more subdued often very, very private close family only celebration. Even non religious people might go to church on christmas eve, the evening is more quiet with classical music, colors are matte, lots of handmade wood things, matte candles and lights. Nothing flashy, pink or plastic.

A lot of traditions, decoration elements made by grandparents hanging in your tree...etc...

Now if you want to do christmas Italian style that will look very different again, Swedish style, Portugal, Spain, Romania, Poland...Europe has a lot of regionally different christmas traditions.

And the buyer will recognize if you thought of them when you created the content.

Many yearly events, when you look at the details, are celebrated quite differently with the specific local customs.

Same for weddings, birthdays, funerals, memorials, end of school celebrations etc...there are a lot of local variations.

Of course it is best to be from the region to do it well, but with good research I might also be able to generate content for a christmas celebration for the Philippines.

When looking for niches I usually just do soup...x million files, german soup or soup germany...much less...often less than 10 000.

And then you start to see gaps everywhere and realize our job is endless and all the duplicates of duplicates don't matter.

Once you really start to look for niches and gaps you realize how empty the agencies really are.

..... I strongly bet that the Flux Model will improve further enormously, because it's licensed now by Elon Musk / X.

Already moving back down. Still it was fun.

pos 692, files 7200

eta

christmas has a high sales potential but you can still make money with other festivals. you just need a lot more and very diverse files. if you have 1000 christmas images, then you will probably need 5000 easter images to get a similar result.

Quote from: Andrej.S. on December 02, 2024, 17:43

..... I strongly bet that the Flux Model will improve further enormously, because it's licensed now by Elon Musk / X.

sure - just look how he's improved twitter!

Good points, but there's a big problem with the data feeding AI�it's drying up, and worse, it's getting polluted with AI-generated content. Real, high-quality data takes decades to build, and we�re running out of fresh, untouched sources.

Sure, AI is useful, but the idea of generating real-time AI images or videos in 3-5 years feels like an overestimation. There isn�t even an AI yet that can produce high-quality 3D models with proper topology, let alone handle physics-based simulations like volcano eruptions accurately. Those still rely on real CPU-heavy mathematics, not AI magic.

The tech is moving fast, but it�s not that fast.

Quote from: Andrej.S. on December 09, 2024, 07:50

Quote from: bennu99 on December 09, 2024, 07:07

Good points, but there's a big problem with the data feeding AI�it's drying up, and worse, it's getting polluted with AI-generated content. Real, high-quality data takes decades to build, and we�re running out of fresh, untouched sources.

Sure, AI is useful, but the idea of generating real-time AI images or videos in 3-5 years feels like an overestimation. There isn�t even an AI yet that can produce high-quality 3D models with proper topology, let alone handle physics-based simulations like volcano eruptions accurately. Those still rely on real CPU-heavy mathematics, not AI magic.

The tech is moving fast, but it�s not that fast.

Yes, it's true that AI image models have likely exhausted the market-leading stock agencies for model training.

However, there are still terabytes of unused image material available. Just think of photo communities like Flickr. Then there are social networks like Facebook, Instagram, etc.

There are libraries with historical photos.
You can also extract images from videos.

And there are very powerful LLMs (Large Language Models) for generating image descriptions, which are necessary for training image models.

Regarding performance, I can only refer to the Flux Fast model.
Even now, the model can generate images almost in real-time, I think 0.1 to 0.2 seconds with 4 steps.

Yes, Flux Developer or Flux Ultra takes a bit longer (on high-performance servers, around 6 to 10 seconds with 28 steps and 2K resolution).

But you can see where the development is headed. With new chip technology for AI models, you could also generate Flux Ultra with 4K resolution in under 1 second. And that's almost realtime.

This still feels like a bubble or a vicious circle to me because:

a) While it�s true that there are still some untapped data sources, most of the high-quality, profitable data has already been mined. What remains is often lower quality, AI-contaminated, or lacks the depth needed for true advancements.

b) Even with more capable models now than a year ago, we still see fundamental issues like six-fingered hands or, in videos, unnatural zoom effects and uncontrollable movement. These flaws highlight that AI, while useful, has serious limitations.

I believe the AI craze will start to fade as people realize it can�t fully replace real mathematics or deterministic systems. GPUs, by design, are non-deterministic, which is why AI will probably never truly understand physics or consistently generate anatomically accurate results like proper hands. It�s incredibly useful in many areas, but we�re far from it living up to the current hype.