[ben_w]

> How many people are just using ChatGPT or Gemini to get the image edits they want instead of reaching for Photoshop? I don't know if this is showing up in Adobe's subscription numbers, yet, but I expect it will eventually.

As per https://xkcd.com/1015/, I suspect many people are doing this and the artists hate all the examples even more than the average consumer who simply treats it as a sign of low-budget work.

My own experiments with ChatGPT's image system is that while I have some pictures I'm very happy with, I also have a surprisingly hard time getting it to follow direction, e.g. shadow direction being inconsistent between foreground and background, making anthropomorphic animals look like they were meant to be (more Elder Scrolls' Khajiit and less big-eyed cartoony fursuits) etc. Stable Diffusion is much easier to deal with in that regard, but then it can't do text and has a much higher frequency of body-horror.

Doing things right is expensive, and most people have no budget. But my guess is people without budget were probably the ones who previously downloaded random pictures off the internet and used them without checking: https://www.theguardian.com/media/2009/jun/11/smith-family-p...

That or perhaps even use pirated versions of Photoshop.

[ACCount37]

The recent ChatGPT Images 2.0 has awe-inspiring image editing and composition capabilities. I can totally see what people mean when they say "Photoshop killer".

[vunderba]

Yup. You still have to do some manual work if you want to create something like a comic and make sure each panel is internally consistent with the world. Heck, even OpenAI’s own cookbook demonstration of the dog comic was a complete mess. I ended up having to make a lot of manual edits just to fix it.

I also think most people fall into the Fiverr camp: they need something that’s “good enough,” and don't have a particularly critical eye for detail - in which case even one-shotting with the latest models, gpt-image-2 or nb2 can get you there.

[koonsolo]

Same experience on my end. It's crazy what the latest version can pull off, with reference images, text, etc.

[rimliu]

so both of you do not understand what Photoshop is really used for.

[ACCount37]

Do you? I somehow doubt it.

Can you provide an example of something "Photoshop is really used for" that ChatGPT Images 2.0 is completely incapable of doing?

[arm32]

- Keeping a human face exactly the same while removing imperfections and being able to make small, purely non-destructive edits in series.

- Export in very high quality without estimative upscaling.

[ACCount37]

Are you up to date on frontier capabilities?

ChatGPT's image pipeline used to completely suck at sequential edits - i.e. "compounding piss filter" and more. The new Images 2.0 one is a generational leap over that. Have you tried it? Do you still have issues?

Not sure what top export resolution is in the system now, but they did bump it up some.

What remains is no true native CMYK? But that is a somewhat niche capability. Not sure if you can partially compensate in "pre" by prompting for print-ready images to keep gamuts constrained and get reasonable outcomes from a naive RGB->CMYK conversion.

[koonsolo]

I used to hire people to do photoshop for me. So yes, I kind of get the idea.

I don't hire them anymore. Can you guess why?

[fragmede]

So tell us instead of being vague And condescending?