|
Can we, as a self-imposed, unenforced rule, try to add some basic info to the image posts? Just the name (or a short tag) for the engine and the prompt that were used? I think that would be helpful to figure out how certain keywords affect the output.
|
# ? Sep 1, 2022 03:21 |
|
|
# ? May 29, 2024 20:53 |
|
mcbexx posted:Can we, as a self-imposed, unenforced rule, try to add some basic info to the image posts? There are lots of better resources for looking up prompt image pairs for each AI system, but I do think it's generally nice to at least know the prompt just because it's more interesting to look at them with that context.
|
# ? Sep 1, 2022 03:25 |
|
Darmok and Jalad at Tanagra
|
# ? Sep 1, 2022 03:36 |
|
tote up a bags posted:I've had real trouble getting quality with SD, to the point I'm wondering if I messed up the installation or something. I'm on a 2080 so I think it should be fine. The output is deterministic, so just take someone's image with known parameters and see if it looks the same on your end. As for advice, SD likes long descriptions. Add detail. Sometimes adding some frivolous descriptions will help focus the image. Also experiment with just tacking some artists you like onto your prompts. Here's a few examples. I prefer changing the aspect ratio to suit the content, but I kept these 512*512 for simplicity's sake. (remember, all the parameters matter, including dimensions) ________________ closeup on a forgotten pharaoh in an underwater cave on mars, by Mike Winkelmann moebius, tomasz alen kopera Steps: 29, Sampler: DDIM, CFG scale: 11, Seed: 9157 ________________ studio Ghibli, rusting boat converted into crate base in wasteland, fallout concept art, subtle lighting, john berkey, muted colors, symmetrical composition, Mechanical Design, gray blue, syd mead, Ralph McQuarrie artstation Steps: 30, Sampler: DDIM, CFG scale: 11, Seed: 125 ________________ award-winning closeup portrait of a pretty bored 90s woman wearing a hoodie by rebecca guay, at dawn in a bustling metropolis, gerda wegener michael kaluta phil noto Steps: 30, Sampler: DDIM, CFG scale: 12, Seed: 8746 ________________ kodachrome photograph of husky in Siberia, warm lighting Steps: 40, Sampler: DDIM, CFG scale: 9, Seed: 5854 (I don't have much experience with photograph prompts yet) Rinkles fucked around with this message at 03:59 on Sep 1, 2022 |
# ? Sep 1, 2022 03:57 |
|
Somebody do Louis CK bravely masturbating in front of a Russian tank I used all my credits on cute pictures of puppies
|
# ? Sep 1, 2022 03:59 |
|
mobby_6kl posted:Aaaand we're there! don't care. this piece of poo poo (or one that looks just like it, I don't know) sold for like $86 million
|
# ? Sep 1, 2022 04:02 |
|
Rinkles posted:The output is deterministic, so just take someone's image with known parameters and see if it looks the same on your end. OK yep, I got the exact same result, so I guess I'm just bad at prompts lol
|
# ? Sep 1, 2022 05:06 |
|
Tried out some sofubi stuff with the huggingface stable diffusion
|
# ? Sep 1, 2022 05:10 |
|
unzin posted:don't care. this piece of poo poo (or one that looks just like it, I don't know) sold for like $86 million flag that I think the AI is malfunctioning
|
# ? Sep 1, 2022 05:13 |
|
These dudes are cool
|
# ? Sep 1, 2022 05:13 |
|
me shitposting code:
|
# ? Sep 1, 2022 05:52 |
|
Got the e-mail about the new outpainting feature. Seems to work OK. Original: Local Gigapixel upscale: DALL-E outpainting: (Note: RIP Jim)
|
# ? Sep 1, 2022 06:22 |
|
can someone with actual SD please outpaint the Nathan Fielder "friends laughing just out of frame" picture I think someone tried it a couple months ago and it didn't work but surely with SD we can get er done
|
# ? Sep 1, 2022 06:25 |
|
The outpainting feature is actually very buggy right now, most of the time when you put the 1024x1024 square over the image, type in a prompt, and click "generate", the progress bar quits before getting to the end, and nothing happens. I lied about it working OK. I tried the Nathan Fielder friends out of frame pic and couldn't get anything to work. https://www.youtube.com/watch?v=r3-Y31ONZuI
|
# ? Sep 1, 2022 06:45 |
|
tote up a bags posted:OK yep, I got the exact same result, so I guess I'm just bad at prompts lol Everybody is bad at prompts at the start, because the default prompts are like "a bird with a funny hat" when they should be word salad like "an award winning nature photograph of a bird with a funny hat, by the world renowned photographer Annie Leibovitz, sharp focus, golden hour, 4k, highly detailed" Get the subject in there, but to really spice it up you just need to give the model something. e: And there might not even be anything wrong with your prompt, with a different seed it could produce wonderful results Wheany fucked around with this message at 06:50 on Sep 1, 2022 |
# ? Sep 1, 2022 06:47 |
|
AARD VARKMAN posted:can someone with actual SD please outpaint the Nathan Fielder "friends laughing just out of frame" picture someone did manage to do it with dalle. afaik, there actually aren't great tools for doing this kind of thing with SD. not without affecting the source image. Tunicate posted:Just don't include the person in frame. Easy! Bula Vinaka posted:Got the e-mail about the new outpainting feature. tbh, that's not great.
|
# ? Sep 1, 2022 06:49 |
|
SD @ home, "a dachshund in an mech suit, 4K, award winning, highly detailed, artstation", CFG=12:
|
# ? Sep 1, 2022 08:24 |
|
Original from SD: Outpainting from Dall-E 2:
|
# ? Sep 1, 2022 09:01 |
|
BARONS CYBER SKULL posted:Original from SD: I think this is Dall-e doing work-to-rule
|
# ? Sep 1, 2022 09:45 |
|
couple of fun ones (all 512x512, k_euler_a, 50 steps), grabbed prompts from https://lexica.art/'s new grid mode and messed around a bit: a golden retriever wearing sunglasses, gta5 box art, borderlands style, celshading, trending on artstation, by rhads, andreas rocha, rossdraws, makoto shinkai, laurie greasley, lois van baarle, ilya kuvshinov and greg rutkowski beautiful ghoul, bloodborne style, celshading, trending on artstation, by rhads, andreas rocha, rossdraws, makoto shinkai, laurie greasley, lois van baarle, ilya kuvshinov and greg rutkowski floating cupcake, bright blue sky, clear skies, bright colors, bright tone, beautiful, scary, epic, cinematic, god lighting, back lighting, sun rays, lens flare, clean lines, ultra detailed, fantasy, concept art, unreal engine (Love how this one picked up some watermark from somewhere) a white stoat happily jumping around , gta5 box art, borderlands style, celshading, trending on artstation, by rhads, andreas rocha, rossdraws, makoto shinkai, laurie greasley, lois van baarle, ilya kuvshinov and greg rutkowski new av material scattered pencils on a desk, fantasy game, front shot, highly detailed, illustration, fantasy art, in the style of greg rutkowski, epic, fantasy, intricate, hyper detailed, artstation, concept art, smooth, sharp focus, ray tracing this is what happens when you set CFG to 7 apparently? It's nice though but where are the pencils and finally donald trump serves mcdonalds at the white house
|
# ? Sep 1, 2022 11:06 |
|
tote up a bags posted:couple of fun ones (all 512x512, k_euler_a, 50 steps), grabbed prompts from https://lexica.art/'s new grid mode and messed around a bit:
|
# ? Sep 1, 2022 11:16 |
|
the training data has an idea of what a "Maluch" looks like but unfortunately doesn't quite get the details. actual car: SD:
|
# ? Sep 1, 2022 12:43 |
|
Slightly modified from someone else's prompt, but here's Portrait of futuristic samurai holding a corgi, sci - fi, tech wear, intricate, elegant, highly detailed, digital painting, artstation, cinematic lighting, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha, 8 k (I added the part about holding the corgi) The outfits for the doggies are the cutest thing ever.
|
# ? Sep 1, 2022 12:55 |
|
Bula Vinaka posted:DALL-E outpainting: Now that's a body pillow
|
# ? Sep 1, 2022 13:15 |
|
AARD VARKMAN posted:can someone with actual SD please outpaint the Nathan Fielder "friends laughing just out of frame" picture
|
# ? Sep 1, 2022 13:40 |
|
I saw in this thread a couple weeks back that Dalle was going to be dropping prices starting today but that doesn't seem to be the case when I go to buy credits. Does anybody know if they changed their minds? I like Dalle for a few reasons but its definitely the most expensive.
|
# ? Sep 1, 2022 15:18 |
|
hydroceramics posted:I saw in this thread a couple weeks back that Dalle was going to be dropping prices starting today but that doesn't seem to be the case when I go to buy credits. Does anybody know if they changed their minds? I like Dalle for a few reasons but its definitely the most expensive. You are confused, OpenAI is reducing the cost for GPT models, these are the ones that write text not the diffusion AI models, which is what Dall is. It was announced the same day Stable Diffusion released and people speculated it was related. I'm pretty sure it was misreported some places too so don't feel like you missed something, your source may have been bad.
|
# ? Sep 1, 2022 15:21 |
|
pixaal posted:You are confused, OpenAI is reducing the cost for GPT models, these are the ones that write text not the diffusion AI models, which is what Dall is. It was announced the same day Stable Diffusion released and people speculated it was related. I'm pretty sure it was misreported some places too so don't feel like you missed something, your source may have been bad. What a bummer but thanks for the information!
|
# ? Sep 1, 2022 15:37 |
|
hydroceramics posted:What a bummer but thanks for the information! Hopefully they'll reduce it. You can always use hugging face SD for free if you don't have the hardware to run SD https://huggingface.co/spaces/stabilityai/stable-diffusion I know Dalle has a style and some people want that, the prompt style is also different and doesn't translate well I kinda feel like I'm suggesting you play PS3 with free online because Xbox360 requires Xbox Live. (PS4 was the first one to require PSN for online play). Maybe you missed this option though. I find it useful when I'm away from my computer and just have to know what some idea would look like.
|
# ? Sep 1, 2022 15:43 |
|
|
# ? Sep 1, 2022 15:52 |
|
pixaal posted:Hopefully they'll reduce it. You can always use hugging face SD for free if you don't have the hardware to run SD I was in the SD beta and can use the website they've recently moved to, I just find it really frustrating to use. There's a ton of slider bars but I frequently get errors that one of them is wrong but they won't tell me which one or why. If I set everything to the minimum everything looks like crap. I just got fed up. Dalle makes more sense to me but like I said its definitely the most expensive option.
|
# ? Sep 1, 2022 15:53 |
|
hydroceramics posted:I was in the SD beta and can use the website they've recently moved to, I just find it really frustrating to use. There's a ton of slider bars but I frequently get errors that one of them is wrong but they won't tell me which one or why. If I set everything to the minimum everything looks like crap. I just got fed up. Dalle makes more sense to me but like I said its definitely the most expensive option. Honestly, of the big three AI image generators, I think DALL-E is the worst. It just feels so restricted in what it can do. I feel like it doesn't allow you to generate any image that could even be vaguely inappropriate. It can produce great results, but everything feels so sanitized.
|
# ? Sep 1, 2022 16:13 |
|
When you think about it, AI's limitation is only time. It's driving us to create it out of our desire to make things easier, but we're never going to be able to restrict it once it really gets going, so the only brakes on this is the time it takes to take over. We're all doomed and these fun pictures are just the bait of us walking right into it's trap.
|
# ? Sep 1, 2022 16:17 |
|
IShallRiseAgain posted:Honestly, of the big three AI image generators, I think DALL-E is the worst. It just feels so restricted in what it can do. I feel like it doesn't allow you to generate any image that could even be vaguely inappropriate. It can produce great results, but everything feels so sanitized. I fed your post into DALL-E and got back 4 screen of static. DALL-E is pissed! Usually when I feed it nonsense words I still get something.
|
# ? Sep 1, 2022 16:21 |
|
LifeSunDeath posted:When you think about it, AI's limitation is only time. It's driving us to create it out of our desire to make things easier, but we're never going to be able to restrict it once it really gets going, so the only brakes on this is the time it takes to take over. We're all doomed and these fun pictures are just the bait of us walking right into it's trap. It started with the wheel and was a slippery slope of laziness
|
# ? Sep 1, 2022 16:21 |
|
I actually think the limitation, at least of the current models, is the training data. It can only make derivatives of what people have already made before and it really shows. Current models are definitely not doing a good job of making cheesy bread stick rats with bread stick tails attacking Sean Hannity on his own show, that's for sure. Also the appropriation of people's art/photos in that data is more than a bit suspect (lol at shutter stock logos being spontaneously reproduced in some images, for example.)
|
# ? Sep 1, 2022 16:25 |
|
Stable diffusion. I gave it a non-breaking space. First an actual one, then and it gave me the same four images. Don't know what the actual implications of this are. E: lol "& n b s p ;" without spaces the second time. Apparently the awful apk translates that into an actual non-breaking space. Flipperwaldt fucked around with this message at 16:27 on Sep 1, 2022 |
# ? Sep 1, 2022 16:25 |
|
"The thing that should not be", colorful photograph by Michal Karcz and Gregory Crewdson
|
# ? Sep 1, 2022 16:28 |
|
Mercury_Storm posted:I actually think the limitation, at least of the current models, is the training data. It can only make derivatives of what people have already made before and it really shows. Current models are definitely not doing a good job of making cheesy bread stick rats with bread stick tails attacking Sean Hannity on his own show, that's for sure. Also the appropriation of people's art/photos in that data is more than a bit suspect (lol at shutter stock logos being spontaneously reproduced in some images, for example.) Mostly they are good at making a single subject right now and not an entire scene. You can sometimes trick them into making simple scenes but actually setting up something that complicated it still on the to do list for many and after video. Which ugh video seems weeks away I'd be very surprised if I'm not able to make little AI gif(t)s for Christmas. Everyone reading groans at gif gift, one person says it's Jiff this joke doesn't make sense.
|
# ? Sep 1, 2022 16:32 |
|
|
# ? May 29, 2024 20:53 |
|
mobby_6kl posted:
poo poo I really like these E: is there any way to add data to SD's learning set? Of course I'm asking this as someone who can't figure out how to make it use my gpu so maybe I'm not the best person, but I'm curious anyway. TheWorldsaStage fucked around with this message at 16:55 on Sep 1, 2022 |
# ? Sep 1, 2022 16:34 |