Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

BONECRUSHER said:
This is likely only true for RDNA3 and RDNA 4 maybe. The cocksuckers wanted to drop all support for RDNA1 and RDNA 2 until the public outcry got really loud a few weeks ago. My 6600 XT didn't work at all last time I checked, other than Vulkan.
Click to expand...
Do you mean because they released some game optimisations only for RDNA3 and RDNA4 a couple of weeks back? I don't think they ever said they were stopping doing driver updates for previous architectures.

As to ROCm, yes, it's HPC software I think first designed for the CDNA architecture, that they're back-porting support to consumer chipsets for. It's available for RDNA4 and also RDNA3. But I don't know how far back their going to go for cards that simply aren't as puissant in hardware terms and a minority of a minority for people wanting to use ROCm (firstly it's more professionals, then of the consumer stuff, you're talking cards three generations ago that wouldn't be good with it anyway). I believe that you can get older ROCm versions on RDNA2, it's not listed for v6+ currently. I would just stay with Vulkan on those. Sorry.
 
Overly Serious said:
I don't think they ever said they were stopping doing driver updates for previous architectures.
Click to expand...
1762783256226.png
Overly Serious said:
I would just stay with Vulkan on those
Click to expand...
Nah, fuck that, it's better to dual boot with Linux at that point for full speed and compatibility.
 
BONECRUSHER said:
View attachment 8150600
Click to expand...
I don't really follow the gaming side of things but that sounds like what I said, I think? That they said they weren't doing game optimisations for the older cards anymore. You said that they were dropping all support for RDNA1&2 and I didn't think that sounded right. And looking at that statement it isn't. They're still supported.

BONECRUSHER said:
Nah, fuck that, it's better to dual boot with Linux at that point for full speed and compatibility.
Click to expand...
Well, what I said was with the new ROCm versions which are native on Windows, there's a lot less need to. If you don't have those new ROCm versions then... yes?
 
 
It's definitely doing a solid job so far. I didn't really understand everyone slobbering over GPT 4o Ghiblification, especially because of the piss filter, but these results feel a lot more up to par. The price has at least quadrupled (probably why you cannot access it for free, and why gens included with AI Plus/Pro are watermarked), but it's still worth the price considering just how much it can do. That said, how long is this gonna last before it gets filtered to absolute shit?
fd040746-4f25-4d5a-8860-6df32cd43aa0.jpg
 
I do enjoy my investment into a 3090. It may not be the fastest, or the most power efficient, or the most optimal purchase when it comes to GPU's, and it may be missing a bunch of fancy newer Nvidia features, but I can game and run some AI on the side without spending a fortune on a 4090 or a 5090 for the VRAM capacity. Most importantly: it's mine, it's offline, and I only pay in electricity to run AI.
 
Slav Power said:
I do enjoy my investment into a 3090. It may not be the fastest, or the most power efficient, or the most optimal purchase when it comes to GPU's, and it may be missing a bunch of fancy newer Nvidia features, but I can game and run some AI on the side without spending a fortune on a 4090 or a 5090 for the VRAM capacity. Most importantly: it's mine, it's offline, and I only pay in electricity to run AI.
Click to expand...
I wish I had just jumped into a 3090 from the start instead of mixing 3060s and 3080s from eBay. I got there anyway eventually.
 
Z Image Turbo was released yesterday. It's a new Chinese model trained with 6 billion parameters (for comparison, Flux 2 has 32 billion). Uses a small Qwen LLM for text encoding, allowing natural language prompts with surprisingly good understanding. Text, prompt adherence, aesthetics, hands, everything is really, really good for a model that will run easily on 16GB VRAM. It doesn't know what a kiwi bird is though.
ComfyUI_01703_.png
HuggingFace model page
ComfyUI workflow

A base (i.e. non-turbo) and edit model are still to be released.
 
I haven't really dabbled with this stuff since my early technical interest playing around with SDXL and earlier, though I did download and run Flux locally for a bit of a play. I'm thinking of renting a little cloud compute power and spending a morning mucking around with this again and seeing how far it's come on.

What would be the most interesting models to play with - focus mostly realistic photos (I'm not talking about porn) and maybe some supernatural elements - I thought I might make illustrations for some RPG stuff, character portraits, things like that. But also just for fun to push the boundaries of what these models are capable of now. Would be fun to do some hyper-realistic character portraits that were short animations as well.

Also am I write in surmising that LoRAs and fine-tunes are less of a thing now? All the big discussion now seems to be around huge models like Flux 2 and such and I don't think there's the same community modding/LoRA/fine-tune sort of scene around those due to size? But maybe I've just not been looking.
 
You must log in or register to reply here.
Back
Top Bottom