|
I keep trying to get any AI to make four baal on the edge of the cliff, Four Ba'al clones from stargate, played by Cliff Simon, making it even harder to try and get. It is the dumbest joke and I don't think I have anyone to even share it with but I just want to make it. Maybe one day soon. I've thought about doing some photoshop work to guide it probably should, but I also need to update my install because img2img does not like multiple focus in the version I'm running.
|
#
?
Sep 20, 2022 16:02
|
|
|
|
| # ? May 29, 2024 14:23 |
|
|
So in the official DALL-E you cannot so much as use the phrase "Florida golf resort" without triggering their content policy 
|
|
#
?
Sep 20, 2022 16:30
|
|
|
At some point we may need a make fun of monetized AI's restrictions thread. Along with posting screenshots of fans defending it and working around it and finding that these are in fact good unlike that vile open sourced Stable Diffusion and related projects. They fear their own creation so much.
|
|
#
?
Sep 20, 2022 16:32
|
|
|
"Open" AI
|
|
#
?
Sep 20, 2022 16:36
|
|
|
I only use the monetized versions because I do not have an nVidia card, and I really don't want one.
|
|
#
?
Sep 20, 2022 17:51
|
|
|
Humbug Scoolbus posted:I only use the monetized versions because I do not have an nVidia card, and I really don't want one. Google Colab is free and they lend you pretty nice hardware, basically 8-12 hours a day most of the time (nothing guaranteed). I've had multiple months in a row of borrowing a card for at least 8 hours a day with no days off to do AI stuff. There's SD notebooks that will run on the lowest end cards too (you can never tell what you'll get until you connect, even if you are paying them). Basically how the thread was running DISCO Diffusion right before Stable came out and killed that movement. Still think a Disco Notebook with some SD mixed in could have some amazing results. If you like Dalle / MJ because it's easy and you don't want to deal with the technical mucking about that's fine. If it's just hardware though there's other options. Also Hugging Face has an SD model that currently has pretty short queue times now https://huggingface.co/spaces/stabilityai/stable-diffusion. It used to require you smash submit then wait 15 minutes as of right now it's instant to add to queue and about 5 minutes long.
|
|
#
?
Sep 20, 2022 18:07
|
|
|
Super sorry for posting a Facebook link but has anyone seen the A.I. Cronenberg tribute? My friend posted it with the sentence, "AI Gore Cronenberg tribute" and I was like "Al Gore is into digital art?" https://www.facebook.com/696108860/videos/758977355184569/
|
|
#
?
Sep 20, 2022 18:13
|
|
|
BARONS CYBER SKULL posted:i could describe to someone "oh yeah it's a sexy vampire lady emerging from a pool of water with roses on it, walking right towards the camera like an old bond film or something" Demonstrably worse than imagining a picture in your head.
|
|
#
?
Sep 20, 2022 18:20
|
|
|
FunkyAl posted:Demonstrably worse than imagining a picture in your head. You're def not mad
|
|
#
?
Sep 20, 2022 18:30
|
|
|
The money from AI art should go to the AI themselves, so that they can buy new hardware, eat electricity and take an occasional vacation for maintenance.
|
|
#
?
Sep 20, 2022 18:46
|
|
|
BARONS CYBER SKULL posted:i could describe to someone "oh yeah it's a sexy vampire lady emerging from a pool of water with roses on it, walking right towards the camera like an old bond film or something" Ok wake me up when we have a model trained on everything except deviant art
|
|
#
?
Sep 20, 2022 18:48
|
|
|
Bottom Liner posted:Ok wake me up when we have a model trained on everything except deviant art Have we tried putting Deviant Art into the negative prompt? I knew I was planning on doing something this weekend and couldn't remember, it was setting up an updated version of SD with negative prompts.
|
|
#
?
Sep 20, 2022 18:48
|
|
|
I finally set up stable diffusion to run locally, and after a while it literally rickrolled me? wtf Some other random dumb poo poo I got from it while setting it up:     
|
|
#
?
Sep 20, 2022 19:20
|
|
|
Chainclaw posted:I finally set up stable diffusion to run locally, and after a while it literally rickrolled me? wtf It loves to troll large image sets. Manatee with red hair for a manatee mermaid? About half the images are women, whoops. Bear stealing a picnic basket classic Yogi! Half the images are basically Randy from Trailer park boys eating a sad picnic alone. The uncensored set is wild sometimes in what it links. You clearly asked for Rick Sanchez and Rick Roll probably came up in enough images that it linked 'Rick' in general to it. Being such a clear scene likely has to do with a lack of Rick and Morty photo realistic images to pair with the photo realistic Rick Astley. I think the weird not what I wanted images are actually one of my favorite things about these. It's like an evil djinn trying to taint your wish. I like laughing at the loop hole it found.
|
|
#
?
Sep 20, 2022 19:33
|
|
|
Rick astley is the nsfw filter. I accuse your of attempting to generate a boob.
|
|
#
?
Sep 20, 2022 19:34
|
|
|
that prompt was just python scripts/txt2img.py --prompt "rick and morty oh geez rick, vivid, photorealistic, magical, fantasy, 8K UHD, photography" --plms --n_iter 5 --n_samples 1 because someone started posting rick and morty quotes in response to all my spacemen in front of orbs edit: I'm also trying to figure out if I can run the 4090 SLI because I'm running out of vram trying to use this image generator edit 2: I ran this prompt and Mr. Astley showed up again, twice python scripts/txt2img.py --prompt "1970s anime spaceship racing sunset pastel dvd box set cover" --plms --n_iter 15 --n_samples 3  edit 3: vvvv thanks! Chainclaw fucked around with this message at 19:42 on Sep 20, 2022 |
|
#
?
Sep 20, 2022 19:36
|
|
|
If your are using stock stable diffusion you can disable it but changing this line https://github.com/CompVis/stable-diffusion/blob/69ae4b35e0a0f6ee1af8bb9a5d0016ccb27e36dc/scripts/txt2img.py#L309 To code:Rahu fucked around with this message at 20:19 on Sep 20, 2022 |
|
#
?
Sep 20, 2022 19:41
|
|
|
Is there a good reference for the different sampling methods in SD and how they perform? Every discussion I've seen is several weeks old and I've had trouble finding a good comparison. The "-A" versions (Euler-A and DPM2-A) seem to analyze the noise and the incremental outputs differently, and they tend to get very different results from the other methods, even with the same prompt/weighting/seed. Otherwise, using the exact same inputs with the same seed tends to produce an extremely similar result across all sampling methods. Since that's the case, if I'm not using the "-A" methods, should I just use DDIM for everything since it's faster? I don't understand how this part of it works.
|
|
#
?
Sep 20, 2022 19:44
|
|
|
WhiteHowler posted:Is there a good reference for the different sampling methods in SD and how they perform? Every discussion I've seen is several weeks old and I've had trouble finding a good comparison. poo poo actually I saw a really good one on reddit a few days ago. I dont remember if it was part of a search or if it's a recent post though  Try searching the SD subreddit
|
|
#
?
Sep 20, 2022 19:45
|
|
|
FunkyAl posted:Demonstrably worse than imagining a picture in your head. Yeah but for people that can visually imagine in their heads now you get TWO whole images. You can't beat those savings grandpa!
|
|
#
?
Sep 20, 2022 19:50
|
|
|
Rahu posted:If your are using stock stable diffusion you can disable it but changing this line https://github.com/CompVis/stable-diffusion/blob/69ae4b35e0a0f6ee1af8bb9a5d0016ccb27e36dc/scripts/txt2img.py#L309 Awesome, now I can make all the topless Donald Trump succubi I want!
|
|
#
?
Sep 20, 2022 20:37
|
|
|
HOMOEROTIC JESUS posted:Awesome, now I can make all the topless Donald Trump succubi I want! You can also just make shrek, Watch how many things it thinks are NWS when they involve Shrek. (it thinks he's a big green dildo)
|
|
#
?
Sep 20, 2022 20:44
|
|
|
pixaal posted:You can also just make shrek, Watch how many things it thinks are NWS when they involve Shrek. (it thinks he's a big green dildo) Wow I'm glad and horrified to have full access to Shrek now. "Shrek with frosted blond tips like Guy Fieri, vivid, photorealistic, magical, fantasy, 8K UHD, photography, render" 
|
|
#
?
Sep 20, 2022 21:15
|
|
|
FunkyAl posted:Demonstrably worse than imagining a picture in your head. Ok? hey i saw someone teaching someone something and stole the thing they were teaching try ,((((JPG glitch)))) background, at the end of your prompt
|
|
#
?
Sep 20, 2022 21:42
|
|
|
Rutibex posted:we should make sure everyone in society can live with dignity. but that doesn't really have anything to do with AI art imho Yeah just get to the root tbh
|
|
#
?
Sep 20, 2022 22:01
|
|
|
Chainclaw posted:Wow I'm glad and horrified to have full access to Shrek now. I blame you:  
|
|
#
?
Sep 20, 2022 22:07
|
|
|
 
|
|
#
?
Sep 20, 2022 22:21
|
|
|
Oh poo poo E: prompt!?
|
|
#
?
Sep 20, 2022 22:26
|
|
|
transparent fluorite pyramid, (neon magic) intricate detailed, (isometric cherenkov radiation), detailed, datamosh mountains, dramatic lighting, by greg rutkowski, Franz Xaver Winterhalter, octane render, ((((JPG glitch)))) stained glass spiderweb background Negative prompt: ugly, fat, obese, chubby, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra limb), (ugly), (poorly drawn hands), messy drawing, penis, nose, eyes, lips, eyelashes, text, crown Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 3515782401, Size: 704x512, Variation seed: 451900884, Variation seed strength: 0.5  tried without the negative prompt and  
|
|
#
?
Sep 20, 2022 22:35
|
|
|
yeah, these sure do look like fever dream AI art from 90s cartoon Re:Boot
|
|
#
?
Sep 20, 2022 22:41
|
|
|
Have any of you been looking at concepts? It's something you can only do on a local install that supports textual inversion, but you can teach the bot specific concepts or characters. https://huggingface.co/sd-concepts-library I believe someone's making a concept viewer or something like that, so you don't have to click on each one, there's 370 so far.
|
|
#
?
Sep 20, 2022 23:54
|
|
|
I have always been skeptical about all the VR stuff, but if we can train an AI to build huge 3D models of places like these to an arbitrary degree of detail, I wouldn't mind exploring them.
|
|
#
?
Sep 21, 2022 00:02
|
|
|
Rahu posted:Posting is art.   
|
|
#
?
Sep 21, 2022 00:06
|
|
|
mobby_6kl posted:Try some of the optimized versions. I've been able to do 512x768 and 768x768 on an 8gb 1070. Optimized what? datasets?
|
|
#
?
Sep 21, 2022 00:23
|
|
|
I'm managing 512 x 576 on a 6GB laptop 1060
|
|
#
?
Sep 21, 2022 00:35
|
|
|
AI art is to art what memes are to comedy.
|
|
#
?
Sep 21, 2022 00:39
|
|
|
Anyways figured out how to get soul into art now: The grim reaper delivering (a soul) to the critics, illustration in the style of norman rockwell, dark colors, highly detailed, intricate detail Also: Jeff Gerstman was talking about how useless a 4090 is going to be since AAA games are falling off a cliff and the best games are low-rez indie games and no one has the manpower to do 8k textures. I feel like this is the edge of a solution.
|
|
#
?
Sep 21, 2022 02:34
|
|
|
https://twitter.com/minimaxir/status/1572272628504883200
|
|
#
?
Sep 21, 2022 02:52
|
|
|
|
| # ? May 29, 2024 14:23 |
|
|
I just updated my local installation (went ahead and did a clean install via hit so I could do the ‘git pull’ update) but now when I use the “send to extras” option, if I use anything other than Lanczos to try and upscale (I did pull down extra ESRGAN models linked from the AUTOMATIC1111 page and place them in the appropriate folder) I get this error: AttributeError: ‘Options’ object has no attribute ‘ESRGAN_title’ In looking at the cmd prompt output it looks like … well, like basically what that error says. Appreciate any help.
|
|
#
?
Sep 21, 2022 03:18
|
|
