|
that wonka 
|
|
#
?
Aug 4, 2023 03:17
|
|
|
|
| # ? May 29, 2024 16:46 |
|
|
the concept of dad weed is fuckin hilarious for some reason same as succ to power bags
|
|
#
?
Aug 4, 2023 03:30
|
|
|
SniperWoreConverse posted:the concept of dad weed is fuckin hilarious for some reason "I smoke weed medicinally. I have CHRONIC BACK PAIN  "
|
|
#
?
Aug 4, 2023 03:39
|
|
|
black and white cartoon with bottom text "Death by cow tools." showing a human farmer surrounded by anthropomorphic cows standing on two legs holding disfigured but sharp tools made of calcified milk in the style of Gary Larson     I was definitely not expecting System Shock 2 or Dead Space enemies from this prompt lol Mercury_Storm fucked around with this message at 04:01 on Aug 4, 2023 |
|
#
?
Aug 4, 2023 03:54
|
|
|
            
|
|
#
?
Aug 4, 2023 07:06
|
|
|
|
|
#
?
Aug 4, 2023 09:00
|
|
|
hell yes, these are great.
|
|
#
?
Aug 4, 2023 13:28
|
|
|
The article text chatGPT slammed in post?
|
|
#
?
Aug 4, 2023 13:36
|
|
|
  Other than looking a bit melty I think this work flow is starting to come together. Still not where I want it but it's getting close and the inpainting workflow is calling to me but that seems like wasted effort in setting up until this is sitting out good images (it's my settings prompting, XL in comfy is a different beast but feels so much more powerful I must learn this)
|
|
#
?
Aug 4, 2023 17:29
|
|
|
That looks really nice. I hope we'll get some good negative embeddings for XL soon, those worked wonders with anatomy and composition on 1.5.  ^^^ My prompts are getting weirder. e: They look so happy! RIP Syndrome fucked around with this message at 17:52 on Aug 4, 2023 |
|
#
?
Aug 4, 2023 17:43
|
|
|
pixaal posted:The article text chatGPT slammed in post? https://claude.ai/ technically, but might as well be GPT yeah. I didn't even bother trying to ask for drug weed text which is why it's about the true dad weeds rather than drugs. Besides the edit I did at the end. pixaal posted:
While I have to stick with Auto1111 for editing I am curious about setting up comfy.ui for the initial images as it seems that for power users comfy.ui is taking over from auto1111. They look great, btw. RIP Syndrome posted:That looks really nice. I hope we'll get some good negative embeddings for XL soon, those worked wonders with anatomy and composition on 1.5. I had some saved negative text whatevers saved from months ago I reverted to using that seem ok, though it's photograph related and not artistic.
|
|
#
?
Aug 4, 2023 17:57
|
|
|
KakerMix posted:While I have to stick with Auto1111 for editing I am curious about setting up comfy.ui for the initial images as it seems that for power users comfy.ui is taking over from auto1111. They look great, btw. Comfy does inpainting and all that I'm just ironing out how many steps to do at different points and what schedules to use when. I'm using someone's waifu maker layout for the base and working from there. There's even a way to do the mask for inpainting in there under a rightclick menu according to a reddit post I was reading but again I haven't touched inpainting in comfy yet. I'm sure I can steal someone's base and combine it with whatever I end up liking for basic generation to start. e: current workflow (and prompt of related images) press download you should get the .json you probably need to install node manager and then open that up inside comfy and press "download missing nodes" to get the custom nodes. https://pastebin.com/178xbGhM e2: comfy also looks like it supports control net and all that fun stuff with some custom nodes in the manager again haven't had the chance to try that yet. I'm working the spire seed a bit (you can press load from history in comfy UI and get every single setting back from the historical generation including seeds from rands) 
pixaal fucked around with this message at 18:27 on Aug 4, 2023 |
|
#
?
Aug 4, 2023 18:12
|
|
|
KakerMix posted:I had some saved negative text whatevers saved from months ago I reverted to using that seem ok, though it's photograph related and not artistic. A1111 stopped showing my embeddings when I upgraded and switched to XL, so I assumed the old embeddings weren't compatible. I just tried it again, and it seems to be working now. Thanks! 
|
|
#
?
Aug 4, 2023 18:24
|
|
|
 
|
|
#
?
Aug 4, 2023 22:54
|
|
|
      
|
|
#
?
Aug 4, 2023 23:33
|
|
|
Killer robot posted:Gilbert Gottfried and Fran Drescher wearing lab coats in a secret underground laboratory, 35mm film still from a 1990s techno-thriller. I never realized how much Gilbert Gottfried and Fran Drescher looked like Fred Armisen and Carrie Brownstein.  
|
|
#
?
Aug 4, 2023 23:57
|
|
|
~~**fashionista**~~
|
|
#
?
Aug 5, 2023 00:20
|
|
|
KakerMix posted:~~**fashionista**~~ amazing, drat these are getting really cool
|
|
#
?
Aug 5, 2023 04:32
|
|
|
KakerMix posted:~~**fashionista**~~ i like it but i gotta admit i miss the dreamlike nonsense text
|
|
#
?
Aug 5, 2023 06:36
|
|
|
Yeah not sure why but this is still one of my favourite things produced by AI, even though it was only made by DALLE2, which really tried hard to be somehow grammatical despite still being complete gibberish. Too bad """Open"""AI will probably never release that model.
|
|
#
?
Aug 5, 2023 08:02
|
|
|
Children’s picture-book about a seal who is so happy about bread. I’d put that on the shelf by Goodnight Moon, ngl
|
|
#
?
Aug 6, 2023 00:34
|
|
|
Humbug Scoolbus posted:
   
|
|
#
?
Aug 6, 2023 01:08
|
|
|
So I tried plonking the new SD XL model into AUTOMATIC1111 and now my computer is running very, very slowly. I don't really know whats going on? According to the terminal it downloaded something called pytorch_model.bin and then nothing. Edit: Okay, seems to be running again after I let it go for a bit. Maybe it was just a big-rear end file it needed to download? MechanicalTomPetty fucked around with this message at 01:52 on Aug 6, 2023 |
|
#
?
Aug 6, 2023 01:48
|
|
|
MechanicalTomPetty posted:So I tried plonking the new SD XL model into AUTOMATIC1111 and now my computer is running very, very slowly. I don't really know whats going on? According to the terminal it downloaded something called pytorch_model.bin and then nothing. A1111 struggles with SDXL at the moment. Currently a lot of folks, myself included, are using ComfyUI to get around this, because it is enormously better optimized for SDXL, at the cost of caching the models in RAM. Comfy is… a trip, and frankly I don’t enjoy using it, but it’s undeniably very powerful.
|
|
#
?
Aug 6, 2023 01:51
|
|
|
For what it's worth, the new InvokeAI 3.01 has also partially implemented SDXL, txt2img and img2img work atm, (the latter a bit wonky due to Refiner shenanigans), support for in and outpainting (their "Unified Canvas") is still in development. Nodes are still experimental. Looks like they also got a bit better at updating/bugfixing it more frequently. That or the previous version was stuck in limbo for a while because everyone was working to get v3 out the door.
|
|
#
?
Aug 6, 2023 04:39
|
|
|
            
|
|
#
?
Aug 6, 2023 07:52
|
|
|
I saw this today and thought it might be interesting to someone in here. Baldur's Gate 3 came out recently and two of the mods listed on Nexus for it are actually LORAs for generating item icons and action icons 
|
|
#
?
Aug 6, 2023 08:04
|
|
|
Played with SDXL tonight. I added some enhancements for it in Automatic1111 like the refiner and ComfyUI extensions, but even the raw output is really impressive particularly with some of the early tuned models. Dom DeLuise is coming to murder you. He is holding a bloody knife. 
|
|
#
?
Aug 6, 2023 08:50
|
|
|
Crapple! posted:Neat! Good stuff!
|
|
#
?
Aug 6, 2023 14:35
|
|
|
Clipperton posted:that wonka That's Marty Feldman as Wonka, right? Hot drat that would have been a good one.
|
|
#
?
Aug 6, 2023 15:03
|
|
|
Interesting styles. Feel like sharing the prompts?
|
|
#
?
Aug 6, 2023 15:07
|
|
|
   
|
|
#
?
Aug 6, 2023 15:29
|
|
|
the famously well known boob bones
|
|
#
?
Aug 6, 2023 15:38
|
|
|
Boob armor protects your boob bones while boob windows let the chesticles breath. SD's understanding of anatomy may be sus.   
Sedgr fucked around with this message at 17:03 on Aug 6, 2023 |
|
#
?
Aug 6, 2023 16:52
|
|
|
I'm playing with Crystal Clear XL seems to be better than base XL (they want money if you use it commercial)  workflow still being optimized it's pretty taxing (about 20 minutes total with the upscale downscale load)
|
|
#
?
Aug 6, 2023 17:28
|
|
|
kuarduck posted:That's Marty Feldman as Wonka, right? Hot drat that would have been a good one. imgur strips the workflow from the image but I still have it this was from a workflow that wanted 2 positive prompts I went back to doing just one. The poster claimed SDXL worked better like this, It looked like they had the first prompt going to the first 10 steps then 20 steps on the 2nd prompt with the negative going to both. I'm having better results just using 1 prompt and increasing the CFG scale as steps increase. primary 69 year old man smoking a willy wonka rube Golberg bong he is exhaling through his nose the smoke curling around his head filling the dimly lit room the bong is in his hands being held to his lips secondary cinematic quality, old man, smoking bong, willy wonka, rube goldberg, vapor, clouds, bong negative error, watermark, cigarette, small
|
|
#
?
Aug 6, 2023 17:36
|
|
|
pixaal posted:I'm playing with Crystal Clear XL seems to be better than base XL (they want money if you use it commercial) I gotta check that out. Model is here: https://civitai.com/models/122822/crystal-clear-xl?modelVersionId=133832 though I have to wonder how they are supposed to enforce money for commercial use. If you don't tell them how are they supposed to know? I get in the corporates world you can have licenses and things but for a bunch of individuals every time someone talks about licenses attached to these models I just go  Regardless of where the technical legality lines up, how are you supposed to point to an image and go "Ah, yes, see, THIS one is x model because I know, somehow." Baffling! Especially if you do anything editing-wise to the thing and it isn't a 100% AI and nothing else image!
|
|
#
?
Aug 6, 2023 17:42
|
|
|
I think it's more a whistle blower thing where if your company internally is saying yeah just use Crystal or whatever. That said just rename it? Train like 1 more image on it to change the hash? That said if they offer support in that they will train future versions to help meet your needs that could possibly work out for them.
|
|
#
?
Aug 6, 2023 18:29
|
|
|
       
|
|
#
?
Aug 6, 2023 20:28
|
|
|
|
| # ? May 29, 2024 16:46 |
|
|
Junk posted:i like it but i gotta admit i miss the dreamlike nonsense text Yeah I get that, and honestly I do too because the surreal dream aesthetic is totally my jam. However, I'm more interested in actually making this tech work which means I can't leave it well enough alone. pixaal posted:I'm playing with Crystal Clear XL  First shot raw output using the model (Crystal Clear XL) and I gotta say, those are some MIGHT FINE dad hands. Blind luck, the rest seem better for the initial generation.
|
|
#
?
Aug 6, 2023 21:52
|
|