More interested to see how well it preforms as a base when adding works on top. As one knows, most checkpoints, loras, embeddings and whatnot have stretched the original 1.4/1.5 models to their limits. Finally having a fresh base which preforms nicely as a baseline would be curious to see.
I’m pretty excited for the official SDXL 1.0 release tbh
Can you run this through discord? Or do you need a dedicated machine for it. I have no SD experience
Has a SDXL trail for registered users right now.
That’s awesome. Access is a win for everyone!
Looks great! Just waiting for ANYTHING to update so they’ll support SDXL, I know ComfyUI supports it now but I can wait a bit.
for the default 1024x1024 images that SDXL produces, how much VRAM is needed? Is 12GB sufficient?
staff said 8 gb will be enough
That would be very cool. I am eying to buy a ARC A770 16GB for ML applications, but if I can keep it to 8/12GB VRAM a lot of second hand GPUs would also work.
Don’t get an arc. ML stuff mostly relies on CUDA which is only supported by Nvidia cards.
Apparently Intel’s oneAPI is catching up quickly to CUDA, at least compared to AMD’s ROCm mess.
Maybe, but I would be worried that I couldn’t always run any application I wanted, or be able to try out brand new stuff because I have to wait on support to be added.
microsoft and amd joined hands in making Ai chips there is hope for AMD
yeah staff said he finetuned it in 3090
I’m not sure I’d say “better”, but from the few test generations I did, it seems like it’s finally catching up.
MID just gives you visually pleasing images whateever the prompt is but sdxl have better balance of aesthetic pleasing & following prompt accuratly
As a noob, I’ve only been using artbot for Ai generation. When I choose the xdl beta model, I get weirdly abstract images. Looks cool but super abstract!
tell me what you want i will try it.
I’m just screwing around honestly. But if you’re keen to try a prompt. I’m looking for “An angry baboon throwing a hand grenade”
I’ve been trying different styles, kinda like cyberpunk, anime style. But open to anything.
This gets me thinking, perhaps we should make a weekly thread where everyone posts their own creations from the same prompt? Taking creative liberty as the see fit. It’ll be cool to see what everyone comes up with.
That is absolutely awesome! Better than anything I’ve managed!
idk why its not making hand granade but it still better than (delibrate ) best 1.5 model
Same prompt
Can you run this through discord? Or do you need a dedicated machine for it. I have no SD experience
currently you use this in discord bot and there is a leaked model that you use in stablehorde or you can download it run it . official model will be released in mid july .
if you are newbie you can use discord bot its similar to midjourney
