- Joined
- Dec 6, 2020
Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump
- Thread starter Near
- Start date
-
🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
- Joined
- Sep 9, 2016
Hey now, I need no beaver now. Not when I have Jesus Koist, Superstar!
edit: You know what. I never put up these cute koi-horsies I made a few days ago.
Last edited:
- Joined
- Mar 4, 2019
- Joined
- Jul 4, 2022
- Joined
- Aug 28, 2019
I know this isn't the request thread, but those Koi look great. Could you make some koi-dragons for 2024?
- Joined
- Sep 9, 2016
On vacation til mid January with just phone access. You can make them tourself on bingai.
Prompt should be something like:
"Chinese New Year dragons. They are koi fish. Koi fish scales. Chinese dragon body. Sanke koi."
You have to specify koi fish instead of just koi because Microdick autochanges koi to coy when alone and you just end up with dragons giving you bedroom eyes.
- Joined
- Dec 6, 2020
Tried a bit of Bing AI today. Little ways off from New Years, but look at this kiwi in anticipation of 2024.
- Joined
- Aug 17, 2018
Vaccination has never looked so classy.
- Joined
- Mar 31, 2020
Cola is popular even with ice bears
And the Anime version:
And the Anime version:
- Joined
- Sep 9, 2016
I just tried my example prompt and it worked pretty great. These are literally the first four results.
Edit: But I can do better. Modifying the prompt to read "Chinese New Year dragons. They are koi fish. Koi fish tails. Koi fish wiskers. Chinese dragon claws. Chinese dragon bodies. Sanke koi." gives this even better quartet.
Excellent, now all I need to add is that they are barristas in two piece surfing gear and all will be complete.
Last edited:
- Joined
- Sep 16, 2019
So I managed to get ComfyUI running on a 7800xt GPU on Linux Mint. I'm getting around 2-3 it/s for a 768x768 image which is a major improvement over my 1050ti. That would generate an image of the same size at 5 s/it. I know the 7800xt doesn't have official ROCm support and AMD lags behind Nvidia for Stable Diffusion. However, according to the SD benchmarking website (just type in 7800xt), I could be getting around 10 it/s on the same model I'm using with my 7800xt. The other users are also running it on Linux so it's not an issue with me missing DirectML or Microsoft Olive. I'm running the latest Torch and ROCm as well. The only real difference I can see is that they are running Automatic1111/SD.Next and they're running it on the 6.2.0 Linux Kernel while I'm on the 6.5.0 version. Is this a limitation of ComfyUI vs. Automatic1111/SD.Next or am I missing some launch commands for ComfyUI? It's hard to find info about running SD on my card so any help would be appreciated.
- Joined
- Nov 21, 2020
The kernel should make no difference. I'd try A1111/SD.next and see if that makes a difference. I assume your torch and rocm versions match.
In my world the best way to troubleshoot performance is to start as close to the reported system as possible and see how that looks.
I should also probably try SD.next and see how my AMD compares to the benchmarks. It seems slow, but I can't find any good SDXL expected benchmarks for it.
- Joined
- Sep 16, 2019
SD.Next seems to give me similar performance and my benchmark is showing 5.76/3.5/3.79 it/s. A repeat shows 5.87/7.14/8.1 it/s with differing optimizations. I'm going to guess the differences are from that and me using the nightly version of Torch. It's hard to find, but I found some users on Github that said they had performance drops with newer versions of Torch on different RDNA3 gpus. I'm probably just going to have to wait until the 7800xt is officially supported. The 7900xtx and 7900xt recently got support and they seem to have consistent performance, so I hope that filters down soon.
Last edited:
- Joined
- Mar 29, 2014
- Joined
- Nov 21, 2020
- Joined
- Dec 6, 2020
No. It mostly depends on what model/checkpoint you're using. If it's something like Midjourney or DALL-E, the colours used tend to be saturated, lending to a dreamy sort of feel. With something like Stable Diffusion that has a plethora of checkpoints to choose from, you can go from flat 2D animation to 3d photorealism.Does AI-made art always have that surrealism feel?
Last edited:
- Joined
- Mar 29, 2014
I still see AI-generated art as always having more or less of this surreal feel to it, whether or not it's photorealistic.
- Joined
- Dec 6, 2020
- Joined
- Sep 9, 2016
Starting prompt:
"a man made of dough, in a chef's hat. He is laughing and saying, "hee hee""
Best redult:
But I can do better.
Prompt:
"a man made of dough, in a chef's hat. He is laughing and saying, "hee hee". Realistic."
Final form.
Prompt:
"A man made of dough, in a chef's hat. Crawling over a gutter on the ground. He is laughing and saying, "hee hee". Realistic."
Have I accidentally found a way to trick BingAI into making fleshy asses?
"a man made of dough, in a chef's hat. He is laughing and saying, "hee hee""
Best redult:
But I can do better.
Prompt:
"a man made of dough, in a chef's hat. He is laughing and saying, "hee hee". Realistic."
Final form.
Prompt:
"A man made of dough, in a chef's hat. Crawling over a gutter on the ground. He is laughing and saying, "hee hee". Realistic."
Have I accidentally found a way to trick BingAI into making fleshy asses?
- Joined
- Aug 28, 2019
