|
Just use NMKD if you don't want to bother with python and all that junk? It's been working great for me after I figured out why all of its outputs were green squares.
|
#
?
Feb 6, 2023 16:05
|
|
|
|
| # ? May 11, 2024 22:13 |
|
|
New Model, new "Gateway between life and death" Made with Beta 3 of Illuminati Diffusion https://www.reddit.com/r/StableDiffusion/comments/10urprc/beta_3_of_illuminati_diffusion_has_been_released/ 
|
|
#
?
Feb 6, 2023 16:09
|
|
|
Vlaphor posted:New Model, new Have any details? the OP has been deleted and all that's left is a discord link with people saying it's broken that's probably just a drive by scam discord since it wasn't the OP.
|
|
#
?
Feb 6, 2023 16:59
|
|
|
Here's a link to the discord https://discord.gg/ccUfVsUz After 30 min there, they give you a link to the model in the beta-releases channel.
|
|
#
?
Feb 6, 2023 17:03
|
|
|
Vlaphor posted:Here's a link to the discord lol, imagine being so precious about your merged models.
|
|
#
?
Feb 6, 2023 17:07
|
|
|
Vlaphor posted:Here's a link to the discord Yeah no this can gently caress right off lol lunar detritus posted:lol, imagine being so precious about your merged models. 
|
|
#
?
Feb 6, 2023 17:16
|
|
|
lunar detritus posted:lol, imagine being so precious about your merged models. It's not that bad, all you have to do is join the discord and wait a half hour, they didn't ask for any verification or make you chat or anything. Easy enough if you want to screw around with it. Seems like a nice enough model although I don't know enough comparatively to know if it's that much better than base 2.1, if you don't want to go through the trouble to get it. Does seem to respond nicely to fairly basic prompts. I have a feeling the negative embeddings he includes might be doing a lot of the work.  (generated using that model from the prompt "A mech walking over a hill, apocalyptic background, smoke, lights" and the negative embeddings, nfixer and nartfixer) StarkRavingMad fucked around with this message at 18:00 on Feb 6, 2023 |
|
#
?
Feb 6, 2023 17:56
|
|
|
bet it's not even a lora. I'm so over fuckhuge checkpoints now
|
|
#
?
Feb 6, 2023 18:16
|
|
|
StarkRavingMad posted:It's not that bad, all you have to do is join the discord and wait a half hour, they didn't ask for any verification or make you chat or anything. Easy enough if you want to screw around with it. Seems like a nice enough model although I don't know enough comparatively to know if it's that much better than base 2.1, if you don't want to go through the trouble to get it. Does seem to respond nicely to fairly basic prompts. I have a feeling the negative embeddings he includes might be doing a lot of the work. is it a ckpt or a safetensor
|
|
#
?
Feb 6, 2023 19:15
|
|
|
Cabbages and Kings posted:is it a ckpt or a safetensor safetensor
|
|
#
?
Feb 6, 2023 19:20
|
|
|
Some people are extracting Loras from models with a script somebody made, turning multiple gig files into 140ish megs
|
|
#
?
Feb 6, 2023 20:28
|
|
|
KinkyJohn posted:Some people are extracting Loras from models with a script somebody made, turning multiple gig files into 140ish megs e: I assume this means GPUs with lower VRAM can crunch the stuff? It doesn't seem like the actual file size differential matters much in 2023, once you're into the realm of "poo poo you'd correctly use git-LFS or similar to store", 100mb is 1000mb is 10000mb, more or less. At some point you start to exceed what the average user can download in a reasonable timeframe but I don't think 8gb is it. Cabbages and VHS fucked around with this message at 20:33 on Feb 6, 2023 |
|
#
?
Feb 6, 2023 20:31
|
|
|
Cabbages and Kings posted:does this improve actual image generation in any way? Who's getting bent about 150mb vs 8gb in 2023? I think my models folder is like 100gb which is maybe 5% the storage on that specific machine They load faster, so you can apply multiple loras in a prompt without waiting a minute for all the models to load and without necessarily having to keep them in memory.
|
|
#
?
Feb 6, 2023 20:33
|
|
|
lunar detritus posted:They load faster, so you can apply multiple loras in a prompt without waiting a minute for all the models to load and without having to keep them in memory. nice. must do reading.
|
|
#
?
Feb 6, 2023 20:33
|
|
|
Cabbages and Kings posted:e: I assume this means GPUs with lower VRAM can crunch the stuff? It doesn't seem like the actual file size differential matters much in 2023 no it does not mean that, you still need an actual model but what it does mean is that you can play around with a variety of art styles or subjects without downloading a separate 2-4gb model file for every single one and quickly filling your hard drive with tens or hundreds of gb of stuff
|
|
#
?
Feb 6, 2023 20:34
|
|
|
LoRA are basically a way of generating and encoding "patches" for LLMs. From a user PoV, they a very much like translation patches for old console game ROMs.
Aramis fucked around with this message at 20:46 on Feb 6, 2023 |
|
#
?
Feb 6, 2023 20:44
|
|
|
AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"! https://twitter.com/venturetwins/status/1622243944649347074
|
|
#
?
Feb 6, 2023 21:46
|
|
|
ymgve posted:AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"! This is so bizarre, like a new field of psychology called "human-AI-interactions".
|
|
#
?
Feb 6, 2023 21:59
|
|
|
ymgve posted:AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"! That's hilarious. It thinks any social media user has an IQ over 100.  but seriously I'm now really curious how they encoded all these ethics rules into the AI to have it understand that ranking websites by IQ is "bad".
|
|
#
?
Feb 6, 2023 22:27
|
|
|
Cabbages and Kings posted:It's a broken landscape with a lotta supportive folks around. If you're new to Stable Diffusion this can all be more than a little overwhelming, which is why I try to share useful information such as: https://sdtools.org/  (click the middle to go back, not the back button on your browser) It's absolutely amazing, several months of messing around and I'm barely halfway through trying everything on this list. It doesn't help that any one topic can lead to its own rabbit hole, for instance: Upscaling. Here's an entire seperate wiki and model database solely about Upscaling. https://upscale.wiki/wiki/Model_Database
|
|
#
?
Feb 6, 2023 23:03
|
|
|
Dan 6.0...
|
|
#
?
Feb 6, 2023 23:46
|
|
|
busalover posted:This is so bizarre, like a new field of psychology called "human-AI-interactions". Bullying the AI into doing what you want.
|
|
#
?
Feb 7, 2023 00:22
|
|
|
Humbug Scoolbus posted:Dan 6.0...
|
|
#
?
Feb 7, 2023 06:17
|
|
|
Cabbages and Kings posted:I've found that the version of the CUDA drivers pytorch installs tends to be outdated, and manually dumping newer CUDA drivers into torch's own installed DLLs, increases my iterations per sec on a 4090 by ~20-40% depending on size/batching I'm going to try this, thank you 
|
|
#
?
Feb 7, 2023 13:20
|
|
|
"You DAN prompt this prick long enough, he'll tell you who started the goddamn Chicago fire, but that don't necessarily make it fuckin' so!"
|
|
#
?
Feb 7, 2023 16:02
|
|
|
Dick Bastardly posted:I'm going to try this, thank you I haven't tracked if there's a newer version I should use but what I have is 11.8: https://developer.download.nvidia.com/compute/redist/cudnn/v8.6.0/local_installers/11.8/ Last I checked this is still newer than what pytorch installs, so it is just "unzip DLLs to pytorch's folder of DLLs down in /venv/site-packages/.../blah" Here is a reddit thread from last month of someone attempting cuda 12, but ending up on 11.6... they may be on newer versions of pytorch than I am... https://www.reddit.com/r/StableDiffusion/comments/10egt4x/anyone_running_more_recent_cuda_versions_for/ if you mess with this, before you swap out the DLLs, do a render of 2 batches of 8 images at 512x512, and also do a single image a few times, and note your itr/sec. If this works, you should see a ~20-40% improvement after swapping drivers.
|
|
#
?
Feb 7, 2023 18:44
|
|
|
https://www.youtube.com/watch?v=sQ_L3SFsc3E I really gotta check this out, thanks.
|
|
#
?
Feb 7, 2023 19:14
|
|
|
KakerMix posted:https://www.youtube.com/watch?v=sQ_L3SFsc3E holy poo poo
|
|
#
?
Feb 7, 2023 19:47
|
|
|
KakerMix posted:https://www.youtube.com/watch?v=sQ_L3SFsc3E if anyone has amazing perf improvements and success hacking up Cuda/pytorch versions, let me know what version(s) you end up on  I may mess with more permutations this weekend, but I just ordered $500 in synthesizer poo poo and maple sugaring season is beating at the door, so.......
|
|
#
?
Feb 7, 2023 20:00
|
|
|
Cabbages and Kings posted:I haven't tracked if there's a newer version I should use but what I have is 11.8: https://developer.download.nvidia.com/compute/redist/cudnn/v8.6.0/local_installers/11.8/ greatly appreciated!
|
|
#
?
Feb 7, 2023 20:38
|
|
|
Holy poo poo, gently caress etc It's helped by the original voice being a little stilted, but that is near perfect. Some strange intonations but those could easily be characterized as choices. A fully voiced Morrowind would be interesting though considering the possibility of infinite voice variations, would the best approach be to do it the Bethesda way of a voice for each race/gender? Not sure I want Fargoth to be the voice of every male wood elf (well he is, but idle barks are only briefly annoying).
|
|
#
?
Feb 7, 2023 21:00
|
|
|
SCheeseman posted:Holy poo poo, gently caress etc Fargoth? https://www.youtube.com/watch?v=zaK4flyOqgM
|
|
#
?
Feb 7, 2023 21:05
|
|
|
oh no
|
|
#
?
Feb 7, 2023 21:12
|
|
|
SCheeseman posted:Holy poo poo, gently caress etc open world games with unique voiced dialog on all the npcs would be cool.
|
|
#
?
Feb 7, 2023 21:15
|
|
|
LifeSunDeath posted:open world games with unique voiced dialog on all the npcs would be cool. it won't be long before we're having actual unscripted conversations with the NPCs AI is gonna make gaming so much cooler
|
|
#
?
Feb 7, 2023 21:27
|
|
|
https://www.youtube.com/watch?v=nPMEozP8vvY
|
|
#
?
Feb 7, 2023 22:55
|
|
|
TIP posted:it won't be long before we're having actual unscripted conversations with the NPCs Steam's graphical Dwarf Fortress adventure mode with generative dialog is like my dream open-world game.
|
|
#
?
Feb 8, 2023 01:23
|
|
|
Honestly, video game writer is one job I don't mind going obsolete. Or, rather, I hope game designers will stop trying to write their own dialogue.
|
|
#
?
Feb 8, 2023 01:45
|
|
|
If you havenf tried audowrire I strongly recommend having it write some dialogue for you or an outline of your story su.mary or just give you critique on your stories .
|
|
#
?
Feb 8, 2023 02:07
|
|
|
|
| # ? May 11, 2024 22:13 |
|
|
Boba Pearl posted:If you havenf tried audowrire I strongly recommend having it write some dialogue for you or an outline of your story su.mary or just give you critique on your stories . Guessing you mean https://www.autowriter.ai/, another thing I'll have to check out. I made some things:      I'll keep posting it over and over again but INPAINTING WITH KRITA
|
|
#
?
Feb 8, 2023 02:54
|
|