|
BrainDance posted:What's the deal with DPM? When did this become a thing? DPM++ SDE requires half the steps but each step takes twice as long as DPM++ 2M Karras. I don't know about the others just those two but that's the likely culprit. Ninja Edit Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 2, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Denoising strength: 0.48, Version: v1.2.1, Hires upscale: 1.5, Hires steps: 32, Hires upscaler: 4x-UltraSharp KwegiboHB fucked around with this message at 06:51 on Jul 27, 2023 |
# ? Jul 27, 2023 06:23 |
|
|
# ? May 30, 2024 06:03 |
|
normal human hand with five ordinary fingers Steps: 32, Sampler: DPM++ 2M SDE Karras, CFG scale: 7.5, Seed: 1332882019, Size: 1024x1024, Model hash: 31e35c80fc, Model: sd_xl_base_1.0, Version: v1.5.0 This was my first attempt at creating an image with the prompt above using the sd xl 1.0 base model, followed by the sd xl 1.0 refiner model. The results were fantastic. Inertia fucked around with this message at 06:48 on Jul 27, 2023 |
# ? Jul 27, 2023 06:38 |
|
A way to bypass the 4 second limit of RunwayML video. Although I figure you can't go on too long as the image will start to degrade. https://twitter.com/maxescu/status/1684121140019490817
|
# ? Jul 27, 2023 08:10 |
|
Inertia posted:
Tried that same prompt on SDXL as well, also got a hand with five fingers. To be fair, I did get some proper hands as well, I just like how this looks Edit: Yo, gimme a high 7 bro! Vlaphor fucked around with this message at 14:42 on Jul 27, 2023 |
# ? Jul 27, 2023 14:33 |
|
Tried the hand thing on MJ. "A normal human hand with 4 fingers and 1 thumb"; Hey, not bad MJ, you've made a lot of prog.... Huh. Okay. Let's try "a normal orc ish hand with 4 fingers":
|
# ? Jul 27, 2023 17:48 |
|
|
# ? Jul 27, 2023 18:01 |
|
hydroceramics posted:Huh. Okay. Let's try "a normal orc ish hand with 4 fingers": Bing does not understand this
|
# ? Jul 27, 2023 18:04 |
|
Lmao
|
# ? Jul 27, 2023 18:29 |
|
|
# ? Jul 27, 2023 19:29 |
|
it's weird to me that AI images are getting so much better but doing poo poo like reverse image search online is still kinda dogshit.
|
# ? Jul 27, 2023 20:52 |
|
"perfectly normal human hand with the correct number of fingers" SDXL, comfyUI, using this workflow
|
# ? Jul 27, 2023 21:01 |
|
|
# ? Jul 27, 2023 22:51 |
|
for fucks sake posted:"perfectly normal human hand with the correct number of fingers" why are there warts on that hand. what the gently caress
|
# ? Jul 27, 2023 22:54 |
|
https://th.bing.com/th/id/OIG..hQ41GCZS5pub1MIPaFM?pid=ImgGn https://th.bing.com/th/id/OIG.rVW6R84vyVzu9oaRBmfF?pid=ImgGn two men having panic attacks due to burgers
|
# ? Jul 27, 2023 23:01 |
|
|
# ? Jul 27, 2023 23:15 |
|
Dodgson. Dodgson! We've got Dodgson here!
|
# ? Jul 28, 2023 00:00 |
|
Don't sleep on Stable Diffusion XL 1.0s text capabilities: Stable Diffusion XL 1.0 with Auto1111 and lots of inpainting and full resolution. There are surface-level capabilities in this new Stable Diffusion that MidJourney can't keep up with, it's wild.
|
# ? Jul 28, 2023 08:57 |
drat a bottle of chug would go down smooth right about now
|
|
# ? Jul 28, 2023 12:54 |
|
So I take it you can't just grab the model files and run them in auto1111 (in that I just tried that, it took ages to load them and then threw an error), or is that just a problem on my end?
|
# ? Jul 28, 2023 14:55 |
|
I think there is an update for auto1111 for sdxl 1.0 to make it useable, there are also some caveats with using it like in order to use the refiner model you have to send the original image to img2img and run it again. Olivio sarakas (I think I got his name right?) Has a good explainer video on it on youtube that he posted the other day, he talks about how to avoid errors that he ran into and how to get it updated. I've tried comfy UI with sdxl and it worked great but was honestly too complex for me to customize and set up the way I would like it. I also tried invoke ai which was fantastic except it takes way way longer than it should to generate anything and I haven't been able to successfully troubleshoot it. I haven't tried auto1111 with sdxl but it seems the easiest to use still, although it seems you currently have to fiddle with and switch models around to run the refiner and stuff.
|
# ? Jul 28, 2023 15:17 |
|
someone will make it easy in a week and let you automatically download the model like you can do with 2.1 and 1.5
|
# ? Jul 28, 2023 15:27 |
I just got SDXL working in auto1111 myself. Seems pretty neat! Not too difficult. Roughly... Command prompt and navigate to your install. Git pull. Once thats updated Civitai or huggingface have the models SDXL 1.0 and the refiner model. Download and drop in the models\stable diffusion folder and the VAE in the vae folder. Launch auto1111 like normal, it'll do some updating. Open up the webui, go to the settings and under user interface go to the quick settings drop down and add the sd_vae, that will give you an additional drop down in the interface that lets you select the VAE. Generate at 1024. I got confused at first because I was generating nonsense images but it turned out it was just set to 512 by default and doesnt work well with the small size image. Getting some fun stuff just rerunning old prompts.
|
|
# ? Jul 28, 2023 15:35 |
|
So basically like installing any other model, I remember something had to be done for 2.1 to get support in A1111 maybe I will take a look at it this weekend.
|
# ? Jul 28, 2023 15:38 |
|
|
# ? Jul 28, 2023 16:10 |
|
E: sorry this is not the chat GPT thread. IGN Headline: Someone Used ChatGPT to Finish the Game of Thrones Book Series Link: https://www.ign.com/articles/someone-used-chatgpt-to-finish-the-game-of-thrones-book-series Spoiler: That 'independent developer' is a moron who knows nothing about how Chat-GPT works, and didn't bother to find out. I hope it's not a Goon. IGN posted:If you're thinking that's a lot to read and just want a summary of how well the AI actually did at this task, thankfully, Swayne shared some highlights with us. First off, credit where it's due, ChatGPT is apparently very good at tracking character continuity even with a web as complex as that in the Game of Thrones books. One example Swayne posted on GitHub shows ChatGPT including the character Illyrio in the first few paragraphs, then disappearing for over a hundred thousand words before returning for a single scene with Varys, and then disappearing again for another massive chunk of time. ChatGPT was able to remember Illyrio and introduce him again where it made sense to, highlighting ChatGPT's impressive ability to "remember" information and call it back up where needed. That's the exact opposite of Chat-GPT. Let's investigate further... Liam Swayne posted:ChatGPT is capable of keeping track of a massive web of characters over the course of the novel. For example, you can see in this screenshot that Illyrio phases in and out of the story, doesn't appear for hundreds of pages, but is still kept in the back of the mind of the AI for when the character is needed: Really? ENHANCE ENHANCE ENHANCE HE LITERALLY ASKED IT TO WRITE ABOUT ILLYRIO. And then there's this: Or maybe... JUST MAYBE.... you could read the documentation about how ChatGPT uses tokens? Do you think that would work? Or would that spoil the mystique? https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them Oy vey. Doctor Zero fucked around with this message at 00:49 on Jul 29, 2023 |
# ? Jul 28, 2023 18:53 |
|
|
# ? Jul 28, 2023 20:04 |
|
|
# ? Jul 28, 2023 20:35 |
|
|
# ? Jul 28, 2023 21:07 |
|
The man in black fled across the desert, and the gunslinger followed.
|
# ? Jul 28, 2023 21:11 |
Mirrors are still a little goofy but interesting.
|
|
# ? Jul 28, 2023 21:25 |
|
KakerMix posted:Don't sleep on Stable Diffusion XL 1.0s text capabilities: You can't fool me, that text was clearly inserted by a photoshop human. Only a question of time before we'd start getting fake deepfakes, but still, to see it in this thread...
|
# ? Jul 28, 2023 23:42 |
|
Image to Video is definitely the cool fake trailer/pretty twitter video king, but I decided to try my hand at something resembling a real movie scene. I think I did this one thing well, but definitely lol at anyone who thinks this could replace Hollywood right now. https://twitter.com/AnomalyZSeries/status/1685003751260618752
|
# ? Jul 28, 2023 23:43 |
I like this set of gunslingers a lot.
|
|
# ? Jul 28, 2023 23:49 |
|
action scene of winnie-the-pooh breaking out of copyright jail and running into the forest, pursued by police, police cars, sirens, officer in pursuit Juggernaut model I was pretty impressed by the dust.
|
# ? Jul 29, 2023 00:11 |
|
KakerMix posted:Don't sleep on Stable Diffusion XL 1.0s text capabilities: prompts were all simple for SDXL, default settings with a restaurant sign with the words" eat rear end" and a cake with the word "gently caress" on it Soulhunter fucked around with this message at 03:30 on Jul 29, 2023 |
# ? Jul 29, 2023 01:04 |
|
quote:Sausage Envoy provides in-depth coverage of the high-stakes world of global sausage diplomacy and trade. Countries across the globe vie for supremacy in the competitive arena of sausage production and export. Each monthly issue unravels the Machiavellian schemes and manipulations undertaken in the pursuit of sausage dominance: quote:Japanese pop icon Beef Diplomat's sultry new album "On Hasole" is taking the music world by storm. The first single "Sear My Surface" uses steakhouse imagery to convey desire, as Beef croons "let your flame kiss my flesh till I'm well-done."
|
# ? Jul 29, 2023 01:56 |
|
I keep wanting to tap the thread title in silent disapproval for most of the really cool images on here. Can we at least agree on posting the style description if not the whole prompt? Thank youuuuu ...
|
# ? Jul 29, 2023 02:01 |
|
mcbexx posted:I keep wanting to tap the thread title in silent disapproval for most of the really cool images on here. Can we at least agree on posting the style description if not the whole prompt? Thank youuuuu ... I mean I'll dump the whole prompts, just usually there is specifics to it because I'll post stuff that runs the gamut. BUT: Stable Diffusion via Auto 1111 using the DreamShaper XL1.0 model bold text reading "BEEF DIPLOMAT" circa 1988 album cover, studio photo of a stylish old asian man with rainbow coat, Kodak Ektachrome 100 neg: cow, drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly With lots of 'only masked' inpainting
|
# ? Jul 29, 2023 02:05 |
|
how close are we to stable diffusion that will work with radeon GPUs?
|
# ? Jul 29, 2023 02:47 |
|
|
# ? May 30, 2024 06:03 |
Those gunslingers were all some variation of: solo gunslinger skeleton riding his horse crossing a river, (cowboy costume), in the style of prismatic magenta, dripping paint, gothcore, (in the style of Yoann Lossel, moebius, greg tocchini and frank quitely), ((occult symbols, ethereal, otherworldly, mystical, enigmatic, cinematic, ultrafine details, 1970s, vintage)) minimalist, faded Negative prompt: moon, desert, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), messy drawing, penis, nose, eyes, lips, eyelashes Steps: 25, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 123456789, Size: 800x1000, Model hash: 31e35c80fc, Model: sd_xl_base_1.0 variations in the setting and color. Mountain pass, desert, obsidian and gold, prismatic teal, etc. If anyone wants a prompt from anything I post just quote it with a ? mark and I'll dig up the prompt when I get a chance. Can use clip in Auto1111 or I think the describe command in Midjourney if needed as well.
|
|
# ? Jul 29, 2023 03:14 |