Thing is, image generators might be better than when SD came out years ago, there's been zero progress on actually meaningfully integrating AI into artists' workflow.

The problem is AI assisted art does not yet exist the same way you can get an LLM to help you finish a project you started by hand, where you can arbitrarily decide how much control you want to exercise over the output.

▲

dragonwriter 3 hours ago | parent | next [-]

> Thing is, image generators might be better than when SD came out years ago, there's been zero progress on actually meaningfully integrating AI into artists' workflow.

That’s an interesting opinion, but I have no idea what, other than ignorance, it is based on.

> The problem is AI assisted art does not yet exist the same way you can get an LLM to help you finish a project you started by hand, where you can arbitrarily decide how much control you want to exercise over the output.

Except it exactly exists in that way, and (while the tooling was more clumsy than today) it has at least since SD1.5 was SOTA.

If you are judging image generation by the native web interfaces provided as the public face of the big hosted models, you really have no understanding of what is out there, and if you aren't doing that, I don't know where your claims could come from.

	▲	torginus an hour ago \| parent \| next [-]
		Except none of this stuff is good enough to be used in production. Yes, I know of Lora-s, embeddings, Controlnet and ComfyUI, dozens of text-to-3d models etc. I'd say this is at the level of ChatGPT 4 - interesting and shows promise, but only usable with major concessions. You know why people complain about AI art and not AI code? Because AI art is quite noticeable. It still doesn't fit into any real artists workflow (someone who is skilled at drawing, and wants to accelerate their workflow without major loss of quality). The burden of proof lies on you, I suggest you include it next time instead of insulting the person who asks a question.
	▲	numpad0 2 hours ago \| parent \| prev [-]
		Nothing had changed on that front. No one's doing non-literal semantic deconstruction and reconstruction of art. This is perhaps due to irresponsible and thoughtless atmospheres surrounding AI image generators and its core developer circles selectively filtering out those with relevant backgrounds. Maybe some Google products are a bit of an exception, but they're default 5-10 years ahead until it stops updating 15 years before going open source, so that's besides the point. Other than that, it's all techbros making tangential changes and claiming changes were made. Changes maybe, improvements, not really happening.

▲

cthalupa 3 hours ago | parent | prev [-]

> there's been zero progress on actually meaningfully integrating AI into artists' workflow.

This simply isn't true. An easy way someone who does most of their art more "traditionally" can use Krita to very much just use AI to help finish a project.

https://www.youtube.com/watch?v=Ly6USRwTHe0

ComfyUI was also designed from the ground up to follow the same sort of node-based workflow that many CG artists are used to with things like Houdini, Blender, Substance Designer, etc., and includes all sorts of ways you can arbitrarily control where and how much the AI is used for.