Midjourney - AI art
- Thread starter tang0
- Start date
You have to remember that much has been erased and cancelled from history.
This prompt has been blocked. Our system automatically flagged this prompt because it may conflict with our content policy. More policy violations may lead to automatic suspension of your access.
If you think this is a mistake, please report it to help us improve.
I've been warned because I wanted to use something that was known in recent history
This is why local AI generation will take off in the next 12 months. Already you can download the old Stable Defusion dataset that's been trained on and start generating images using the resources of your desktop. As you can imagine it's a hack of a lot slower but you can prompt it without restriction. It's the old SD so you'll get wonky fingers etc but it's only a matter of time before someone will arrange the dataset and AI algorithm of the latest ones.
Looks like @Pawnless Endgame is already doing a form of it
Last edited:
Soldato
A future London see's the return of Earths President from his meetings at the far reaches of the Galaxy.
Soldato
A future London see's the return of Earths President from his meetings at the far reaches of the Galaxy.
Good. The UK is still at the centre of the universe. Colonising space.
Last edited:
Soldato
This is why local AI generation will take off in the next 12 months. Already you can download the old Stable Defusion dataset that's been trained on and start generating images using the resources of your desktop. As you can imagine it's a hack of a lot slower but you can prompt it without restriction. It's the old SD so you'll get wonky fingers etc but it's only a matter of time before someone will arrange the dataset and AI algorithm of the latest ones.
Looks like @Pawnless Endgame is already doing a form of it
Yeah I got my Automatic 1111 working with Stable Diffusion 1.5 and 2.1 at the moment. My RTX 3070 can output an image in 15 seconds at 640x960 then upsizing them takes another 15 seconds or so per image.
I'm also trying SDXL with Automatic 1111 which is a lot slower, more like 2m 15s per image and they're coming out blocky / corrupted, so something I'm not doing right there.
Edit: my processor is an i7-11700K. I'm not sure if I can also utilise that or if it's GPU time only.
Last edited:
That's faster than I expected from a 3070. I would really like to know what the limiting factor is in graphics cards in terms of this. Memory bandwidth, VRAM or CUDA cores. Do you know any websites that tests cards for this? I think Pugit systems does sometimes but it's not very definitive.
I'm hoping I may get a 4080 for December
Soldato
@robgmun - I use Passmark (computer program, not website) to bench my 3070. It's been overclocked, so its Passmark score (25474) makes it more like a 3080 than a 3070. I don't think it goes into deeper details like CUDA cores. Just gives you a 2D score and a 3D score, and the Passmark website rates the GPUs by their 3D scorees. Passmark can also run performance tests for CPU and hard disk.
Soldato
Good thread on NG on AI video game art
www.neogaf.com
Dream up concept art for a new video game in DALL-E 3
DALL-E 3 is available for anyone to use through Bing Image Creator: https://www.bing.com/images/create?FORM=GENILP It can handle detailed, complex prompts and can generate text, too. What games will your imagination dream up? Try specifying different art styles and describe things in detail...
Caporegime
pikachu standing on a pile or corpses with an evil grin set in a dark future
The hand and fingers problem seems baked into the original training dataset that generators like SD and Midjourney etc are all based on. Whilst Midjourney have made their own improvements and SD has embeddings which can help, it's still very hard to get truly convincing hands.
Part of the problem is that AI art generators fundamentally aren't able to count (AI is a misnomer really, but that's another discussion). So you can't reliably prompt it to generate an image with X of one item, although you might get lucky, especially with smaller values. But if you try X of one item and Y of different item, anything could happen.
Same with text. Individual letters of the alphabet will probably work. A short word might work but will probably require workarounds. A sentence definitely won't as AI generators have no linguistic capabilities.
Soldato
Okies, here's one of mine after a few days of dossing with Automatic 1111. I think it's vitally important to state all of the parameters, in particular the 'seed' number. This is a unique identifier for the image, a bit like a decryption key. I don't choose this number by the way because it's random when the image is made, but without this number, the image can't be reproduced.
1st box: yellow, orange, pink, purple, autumn, clear sky, sunset
2nd box: text, blurry
Checkpoint: sd_xl_base_1.0.safetensors
Seed: 3868097332
Steps: 25
Sampler: Euler a
CFG scale: 7
Size: 768 x 768
Upsize by: 4
Upsizer: 4X-Ultrasharp
1st box: yellow, orange, pink, purple, autumn, clear sky, sunset
2nd box: text, blurry
Checkpoint: sd_xl_base_1.0.safetensors
Seed: 3868097332
Steps: 25
Sampler: Euler a
CFG scale: 7
Size: 768 x 768
Upsize by: 4
Upsizer: 4X-Ultrasharp
Soldato
Soldato
Bring them lights to me.
\
\
Soldato
@aVdub - the Oppenheimer picture is missing Barbie 
Soldato
@aVdub - the Oppenheimer picture is missing Barbie
Would be a great one to do of Oppenheimer accepting delivery of three swastika stamped atomic bombs, but alas I would probably lose my account and be cancelled
OH And Barbie would be driving the truck back.
Last edited:
Soldato
I see you have spotted what keeps Johnny alive
I was slightly disappointed with result of 'johnny 5 from short circuit disco dancing in a nightclub with multicoloured lights everywhere. photorealistic art'
Soldato
@aVdub - I saw up-thread that @adam cool dude tried to use an AI tool on Bing to draw Hitler and he said it got blocked. I didn't realise that Bing could do AI but I know it's primarily a search engine. On the regular search engine itself though (www.bing.com) I only had to type in Adolf and it auto-suggested Hitler, then it came up with pictures of Hitler no probs. So I wonder why it's ok to search for him, but it's not ok to AI him up?
Conversely, I just used Automatic 1111 to draw a cartoon version of Hitler and it came out fine (not blocked). It didn't do the Nazi logo on his hat/cap quite right though. It made it look more like a hash (#) symbol rotated 45 degrees LOL
Conversely, I just used Automatic 1111 to draw a cartoon version of Hitler and it came out fine (not blocked). It didn't do the Nazi logo on his hat/cap quite right though. It made it look more like a hash (#) symbol rotated 45 degrees LOL
The Bing image creator is using dall-e. Give it a go.
www.bing.com
Bing Image Creator
Free, AI-powered text-to-image generator transforms your words into stunning visuals in seconds. Perfect for quick and easy image creation. Unleash your creativity with Image Creator in Bing!