For those who don’t follow this stuff Flux is a new text to image model that was just realeased and is making big waves in the AI space. It seems to have some understanding of who Twilight sparkle is, although it doesn’t get her cutie mark right.
>
This is without any finetuning to specifically train it on ponies which is incredible. I have doubts as to whether a pony model will be trained on it though, since the licencing is questionalable (not as bad as SD3) and there are some technical difficulties and unknowns in doing so.
>
It was truely incredible how easy it was to get workable images out of Flux. It took me literally 20mins to make this meme. It was literally like how people who don’t understand AI, think it is.
Of course, if falls apart when asking it to do any other characters than the mane six, and simple poses and shots. And while it is great, It won’t be replacing Ponyxl for ponies any time soon
>
>
>
>
>
>
>
I’m a Button Fixer. I fix Buttons <3
I’m not exactly a fan of AI images on account of their “soullessness” but I can’t help but be impressed (and sbit unnerved) at how good it’s getting at it.
@Background Pony #6231
Well, it’s a horrible mess for me. I’m not a blender or unreal engine material artist to dig through those nodes. I like the simplicity of A1111’s webui that made for not-so-advanced users.
@Background Pony #6231
Really? I don’t think ComfyUI is that bad. Is reminds me of Scratch from when I was a kid. It’s fairly logical to build stuff and once it’s working, you don’t have to touch it. It’s just understanding someone elses workflow that you’ve downloaded that sucks. I use Kohya ss for lora training and I find it worse to use to be honest.
I would like to try that model once it’s available on A1111 WebUI ;)
On the less related matter: f#ck the guy who decided to name that effin’ clusterf#ck “ComfyUI” - there’s nothing comfy about that garbage ¬_¬
True
Yeah I totally understand the simplicity. I pretty much only used Apple products until I got into Ai
Well, it’s a horrible mess for me. I’m not a blender or unreal engine material artist to dig through those nodes. I like the simplicity of A1111’s webui that made for not-so-advanced users.
There is a limit, 2 AI-generated images per 24 hours (Rule #6).
Real artists deserve more love :)
Really? I don’t think ComfyUI is that bad. Is reminds me of Scratch from when I was a kid. It’s fairly logical to build stuff and once it’s working, you don’t have to touch it. It’s just understanding someone elses workflow that you’ve downloaded that sucks. I use Kohya ss for lora training and I find it worse to use to be honest.
Well, the basic ComfyUI workflow isn’t too bad, but more advanced ones look like spaghetti.