How do you get 4K out of any Ai engine! Chat gets me 1K and as far as I know, that’s all you get - anything else is upscaling. But you can only go so far with a 1K image. Majestic.ai does great with trees, grass, and bushes, but destroys anything else that isn’t organic. Am I missing something?
Gigapixel AI
Gigapixel AI (Topaz AI)
To get up to 4K resolution, I use Gemini, but with the Google AI Studio. It is some kind of Advanced Environment for AI usage that needs some linked payment methods and costs - besid the already necessary Pro membership - also a few coins for each generation. But, you can choose 4K, without watermark.
For what? Upscaling your renders or generating from scratch? Do you mean magnific.ai? I use magnific for render upscale. Although haven’t for a while since they like moved to a new web of web or something. For editing renders nano banana pro is excelent, but some websites don’t offer 4k resolution, like open.ai
But if you use venice.ai it has an edit button, you can then choose nano banana pro, and get generated 4k images, or even textures.
I can upscale in PS, who purchased Topaz. I also have Topaz and I use Majestic. I am talking ChatGPT generated images. To upscale you need something good to start with. ChatGPT and the others produce 1K images and upscaling makes them look like oil paintings. I was told most models were trained on low-resolution web graphics, so you need a new model, like Flux 2.
you can use nano banana in photoshop too. But I found it changes things slightly like your perspective, objects etc. Flux 2 does a better job, but some times…None of them offer perfect solution imho. I used this one called Upscayl, but they all suffer from similar issues.
We all see tons of Ai renderings online, most done with ChatGPT or Gemini, and all have the same limitations 1K. Are these even real projects? All I am seeing good use of Ai for is ideas, enhancing organic things like trees and grass, and low-resolution web stuff.
when we were experimenting with ai in comfyui, we came to the conclusion that all of them re-generate the image for upscaling (if its ai based) there by re-imagining the image it self so it’s always different. True upscale like topaz has always those artifacts that you talk about. Right now i think its a trade off, some people make simple renders and upscale them their clients understand that it limits the look/control in favor of cost and they are ok with it.
It is good to know that I am not alone :). My boss believes AI is the future, which I know is part of the future, but there are serious limitations. I mean, scrolling through Instagram you might be impressed, but zooming in past a post stamp they fall apart quickly. I am sure 99% are not real projects, so no real client, and no real expectations. - it is all pretend.
I make one 3000px X 1688px image using Ai before.
I think it is upscaling the 1K image. It is all a little confusing to me.
Don’t quote me, but as far as I gleaned when this all began is that 1k was the largest and anything else is further upscaling.
I don’t think it’s a limitation per se, only that the 1k is achievable, resource-wise, now, with larger base output being an even more gargantuan drain on resources than it already is.
Maybe…
I have studies ChatGPT for two months now, and from what I can tell, you are stuck at 1K. Majestic.ai does great upscaling 2K back to 4K, but you can’t do much with 1K. I read that Flux is the best model right now, and can handle 4K, but it failed to do much on my local stable diffusion.
if you use ChatGPT’s API the resolutions are much larger but you have to pay per image.
There are specific upscale models that you can use for this purpose in ComfyUI. The workflows are pretty well simplified now, basically drag/drop and render.
I will look into, thanks!