|
also the SDXL models do seem to be a ton better than SD 1.5. i tried playground v2.5 (i think it's its own separate thing, the controls are completely different from SD*, anyway) but it just spit this out: which, don't get me wrong, is certainly visually kinda interesting. but it's also clearly not even close to what i wanted
|
# ? Apr 14, 2024 19:30 |
|
|
# ? May 2, 2024 21:19 |
|
my company just bought me a license for chode pilot
|
# ? Apr 14, 2024 19:37 |
|
|
# ? Apr 14, 2024 19:59 |
|
still using dreamshaper xl, but played around with clip skip, cfg scale and shift getting a little closer i guess, lol. prompt was "a color photograph of richard nixon, despondent, having a drinking contest, versus a horse. horse visible, horse drinking liquor, horse winning", same negative as before
|
# ? Apr 14, 2024 20:04 |
|
Beeftweeter posted:still using dreamshaper xl, but played around with clip skip, cfg scale and shift
|
# ? Apr 14, 2024 20:10 |
|
what are you using for those? i just tried "a color photograph of richard nixon, wearing a batman costume, in the oval office, drunk" and he's wearing a goddamn suit that's dreamshaper XL again, and is kinda funny because for the SD 1.5 models i usually had to specify him wearing a suit
|
# ? Apr 14, 2024 20:21 |
|
just loving around with midjourney. Beeftweeter posted:"a color photograph of richard nixon, wearing a batman costume, in the oval office, drunk" not very nixonian but it nailed the era and i love the background portraits it generated lol
|
# ? Apr 14, 2024 20:31 |
|
mediaphage posted:just loving around with midjourney. hmm, i suppose it is better than the open-source models, or at least probably includes way more stuff. the quality of the images seems to be about the same, though that does depend heavily on the model and what settings you use when generating with oss stuff locally anyway yeah those are great and are more in line with what i was picturing (the portraits are indeed the best part, lol). dreamshaper xl does seem to do a very good job of generating nixon, just not outside of very real scenarios lol. "dreamshaper" my rear end
|
# ? Apr 14, 2024 20:46 |
|
https://www.youtube.com/watch?v=KhQnnISdDIU
|
# ? Apr 15, 2024 04:56 |
|
https://twitter.com/skyezera/status/1778903035432870247 I'm Jorch
|
# ? Apr 15, 2024 09:56 |
|
i'm Big Chunus
|
# ? Apr 15, 2024 10:22 |
|
are you telling me that Shimp fried this rice?
|
# ? Apr 15, 2024 11:02 |
|
I suspect that the text is photoshop because the results never have clear font displayed like that.
|
# ? Apr 15, 2024 15:22 |
|
yeah, ai text isn't anywhere near that coherent usually. i guess they could've gotten lucky once or twice but that many different names? nah also lol @ all the cigarettes in the yearbook
|
# ? Apr 15, 2024 17:52 |
|
hell yeah a yearbook heater. we used to be a society.
|
# ? Apr 15, 2024 18:07 |
|
Matt Zerella posted:hell yeah a yearbook heater. we used to be a society.
|
# ? Apr 15, 2024 18:15 |
|
lol
|
# ? Apr 15, 2024 23:18 |
|
hell yes
|
# ? Apr 16, 2024 00:14 |
|
bahahaha this one is less funny, but still
|
# ? Apr 16, 2024 00:33 |
|
dreamshaper XL doesn't seem to know what "a mcdonalds" is, but hmm, maybe it does? lol i didn't specify "obese"
|
# ? Apr 16, 2024 00:46 |
|
idk what exactly went wrong here, it's the same settings and prompt as above (and the only difference between that and the previous two is substituting "the oval office" for "a mcdonald's" and "drinking liquor" with "smoking a cigarette"). the only change here was switching models from dreamshaper XL to juggernaut XL v9 kinda spooky i guess
|
# ? Apr 16, 2024 01:15 |
|
Han: punch it Chewie! Chewie: Arrwoooo
|
# ? Apr 16, 2024 01:16 |
|
these look like an actor playing nixon in a tv movie, none of them really look like actual richard nixon at all. you'd probably need to train a lora or whatever on him specifically for that.
|
# ? Apr 16, 2024 01:20 |
|
perhaps we can combine the two concepts dickdonalds
|
# ? Apr 16, 2024 01:38 |
|
Elder Postsman posted:these look like an actor playing nixon in a tv movie, none of them really look like actual richard nixon at all. you'd probably need to train a lora or whatever on him specifically for that. yeah, they're pretty close but you can still tell that it's not him. i guess a LoRA might work, but most things i've seen to force the model make the output resemble a certain person on civit.ai have been "embeddings", which draw things does support but i have no idea how they work, lol. you're supposed to be able to use them with just a keyword but it doesn't seem to work (well, i guess) ime then there's the problem of 99% of the poo poo out there being for SD 1.x e: for all of mine i've been leaving "RestoreFormer" (ostensibly some kind of face correction) enabled, maybe that's actually making it look less like him, idk. i'll try a couple without it Beeftweeter fucked around with this message at 01:46 on Apr 16, 2024 |
# ? Apr 16, 2024 01:40 |
|
yeah i'm just using 1.5 anyways. i've tried training embeddings locally but i've never gotten good results from it.
|
# ? Apr 16, 2024 01:43 |
|
Elder Postsman posted:yeah i'm just using 1.5 anyways. i've tried training embeddings locally but i've never gotten good results from it. i was wondering if i can use some 1.x model as a "refiner" with maybe dreamshaper XL as the primary model (it seems pretty good at doing the oval office at least, lol), and it seems like you can. since those seem to be better at more outlandish prompts (and XL is better at realism), maybe that might net better results, idk. you might have read my previous post before the edit too, i'll see if maybe turning off restoreformer might help
|
# ? Apr 16, 2024 01:49 |
|
Beeftweeter posted:bahahaha lol the nips TerminalRaptor posted:Han: punch it Chewie! also lol
|
# ? Apr 16, 2024 15:05 |
|
|
# ? Apr 16, 2024 16:36 |
|
That little "ethnically ambiguous" bit and the reason it keeps showing up is a fantastic example for the design ethics and bias course I am slowly putting together.
|
# ? Apr 16, 2024 16:49 |
|
Agile Vector posted:lol the nips yeah lmao. i, for one, blame joel schumacher
|
# ? Apr 16, 2024 16:56 |
|
was just looking for new models and this is kinda neat. it loops on the site, which works better https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bc0cc3a4-9981-4fa8-b309-d798960cbf2f/transcode=true,original=true/AD_00003%20%2812%29.mp4 (from https://civitai.com/images/10006126)
|
# ? Apr 17, 2024 05:19 |
|
that first shift makes my temples contract in a weird way
|
# ? Apr 17, 2024 15:34 |
|
it is genuinely upsetting me how much i appreciate meta/zuck for their continued wrench-throwing, with llama 3 being released randomly and wrecking a bunch of much larger (closed, investor storytime) models. like, i don't quite get why they do this, but it is fun.
|
# ? Apr 20, 2024 01:49 |
|
i was looking at downloading (stable diffusion, not LLM) models yesterday and found out that the same developer that made dreamshaper has another model pack called "AbsoluteReality", which is intended for photorealism. dreamshaper, apparently, is not lol, although it does a decent job of it considering i just remembered that now lol so i haven't tried it (or even downloaded it), but i will now and see what it can do. dreamshaper is pretty good ime, so hopefully this is too
|
# ? Apr 20, 2024 02:13 |
|
also idk what kind of format it's expecting but i can't get draw things to import a lot of SDXL stuff, which really sucks because e.g. there's a deep space 9 LoRA that looks cool lol
|
# ? Apr 20, 2024 02:19 |
|
i request an image of "tweetin beef, beeef tweet tweet" im joshin cause i think your username is funny but also im deadly serious about my request
|
# ? Apr 20, 2024 02:19 |
|
i wish i could use the ds9 LoRAs because quark's bar would be a great place to put yospos characters e: there's another SD v1.x LoRA for the enterprise-d bridge: https://civitai.com/models/197887/star-trek-bridge-galaxy-class but it's a little bare Beeftweeter fucked around with this message at 02:25 on Apr 20, 2024 |
# ? Apr 20, 2024 02:22 |
|
Beeftweeter posted:also idk what kind of format it's expecting but i can't get draw things to import a lot of SDXL stuff, which really sucks because e.g. there's a deep space 9 LoRA that looks cool lol the only real format for lora is safetensors, so i'd be surprised if it's expecting anything else, is it failing to import, not loading, or just not producing results?
|
# ? Apr 20, 2024 02:39 |
|
|
# ? May 2, 2024 21:19 |
|
Beeftweeter posted:also idk what kind of format it's expecting but i can't get draw things to import a lot of SDXL stuff, which really sucks because e.g. there's a deep space 9 LoRA that looks cool lol dang, gonna shelve my dreams of a fully caked up quark
|
# ? Apr 20, 2024 02:44 |