|
Goons, help. I keep bouncing off of ComfyUI, even though it seems like it has the tools (RePoser Plus) to create the consistent characters that I need but I load it up, stare at it for a while, generate images that are inferior in every way to Midjourney V6, then close it. Then I struggle for an hour trying to get MJ to create the same character in a new scene, even with reference images. I don't know how to pull stuff from github, I don't know how to do anything with ComfyUI other than adding prompts and getting garbage, and I have no idea how to get all of the RePoser Plus stuff loaded in order to even start messing around with it. Where do I begin? Is there a series of tutorials that will take me from zero to basement-dwelling pants-making GBS threads SDXL supergoon? Keep in mind, the only reason I even want to go down this route is that I can't get consistent characters and art styles from MJ, so I'm hitting a wall in my create process. Any help would be enormously appreciated.
|
# ? Jan 18, 2024 02:32 |
|
|
# ? May 30, 2024 14:49 |
|
|
# ? Jan 18, 2024 02:42 |
|
I really want to know what the ai draws from whenever I put in a prompt that contains the words "manic pixie dream". sometimes it's a mix of hipster and lisa frank but not always Bing Image Generator, prompt "manic pixie dream sailor saturn" this set is the closest it got to a sailor scout that didn't look like a recolored sailor moon zetamind2000 fucked around with this message at 04:31 on Jan 18, 2024 |
# ? Jan 18, 2024 03:51 |
|
null_pointer posted:
This guy is German but every video has English subtitles, and he deals with ComfyUI and loving around in it exclusively. I can't say he's amazing but he's pretty thorough and his discord is ok at explaining some things. https://www.youtube.com/@ferniclestix This guy deals with ComfyUI exclusively too, and he has stuff for beginners and is more approachable for beginners. Others that I found either deal with surface level stuff or don't dive deep enough for my liking. But if you want those you can check out Scott Detweiler, as one. If all you want is consistent characters maybe start with A1111 instead as ComfyUI can be intimidating and complicated. https://www.youtube.com/@BitesizedGenius That's the best tldr a1111 channel I found so far explaining some tricky concepts in detail Finally here's my go-to for every single AI generated thing in detail at once https://m.youtube.com/@SECourses/videos You might think he only makes videos to shill his Patreon with scripts and you'd be partially right, but the man genuinely spends a big amount of time playing with AI, training his LORAs etc and it shows in his videos Strotski fucked around with this message at 09:40 on Jan 18, 2024 |
# ? Jan 18, 2024 09:15 |
|
Toxic Mental posted:This isn't AI art but I just wanted to share it here better watch out..
|
# ? Jan 18, 2024 19:19 |
|
|
# ? Jan 19, 2024 00:01 |
|
|
# ? Jan 19, 2024 05:33 |
|
zetamind2000 posted:I really want to know what the ai draws from whenever I put in a prompt that contains the words "manic pixie dream". sometimes it's a mix of hipster and lisa frank but not always If you are using SD, there is this extension https://github.com/castorini/daam It shows you an attention heat map for each word in the prompt. It's really useful to understand what a prompt does and why it works/doesn't work(and also for finding out that 80% of the words in your prompt are doing absolutely nothing and are just cargo culting )
|
# ? Jan 19, 2024 13:13 |
|
Hard to pick out anything anymore, it's like the tech (and people's technical abilities) are so good now that everything's "good" in an AI art sense. I guess that means the distinguishing criteria are being pushed back into the realm of individual taste/creative talent. Anyway, I love these.
|
# ? Jan 19, 2024 14:50 |
|
I SAW HOW YOU'LL DIE, ENSIGN! WELCOME ABOARD!
|
# ? Jan 19, 2024 18:38 |
|
Is there a way to make an AI horror along the lines of the Pizza Nuggets commercial, but without paying RunwayML or whatever else now? For reference: https://www.youtube.com/watch?v=Zrg4t3_PdLM
|
# ? Jan 19, 2024 19:54 |
|
Mercury_Storm posted:Is there a way to make an AI horror along the lines of the Pizza Nuggets commercial, but without paying RunwayML or whatever else now? For reference: Pike gives you 30 credits a day free (1 generation seems to be 10 credits), You can pay Pika for more https://pika.art/ they were free unlimited for a bit. My stuff was taking about 4-8 generations per 3seconds of footage to find one I liked. https://i.imgur.com/zD2lEgb.mp4 About 40 generations were used to make this. Seed image was made with Bing - so about a week to make for free now, when it was unlimited free this took about 2 hours waiting on generations. pixaal fucked around with this message at 20:01 on Jan 19, 2024 |
# ? Jan 19, 2024 19:58 |
|
|
# ? Jan 19, 2024 20:01 |
|
|
# ? Jan 20, 2024 20:12 |
|
Not mine, but "Organ Vending Machine"
|
# ? Jan 21, 2024 00:06 |
|
You can tell it's AI generated because the anime doctor is throwing the entire Blood gang hand sign with one hand.
|
# ? Jan 21, 2024 01:08 |
|
|
# ? Jan 21, 2024 02:38 |
|
|
# ? Jan 21, 2024 02:38 |
|
I love making trippy hallways. Vlaphor fucked around with this message at 04:42 on Jan 21, 2024 |
# ? Jan 21, 2024 04:36 |
|
Bing Image Generator, prompts "backrooms creepypasta" and "the poolrooms"
|
# ? Jan 21, 2024 07:21 |
|
My friend tried to use bing ai art and his prompt of "Dr Doom and Superman fighting" did not go as planned...
|
# ? Jan 21, 2024 13:25 |
|
Guess the song: (Hint: it''s one song represented by all 4 images, not one each)
|
# ? Jan 21, 2024 14:37 |
Bing and dall-e 3 both refuse to depict violence. I've tried pretty hard to get a good fight scene out of them but the closest they'll do is like the described characters posing menacingly at each other like a comic book cover
|
|
# ? Jan 21, 2024 15:00 |
|
Javid posted:Bing and dall-e 3 both refuse to depict violence. I've tried pretty hard to get a good fight scene out of them but the closest they'll do is like the described characters posing menacingly at each other like a comic book cover Yeah, you either gotta go into Stable diffusion (Note: I have yet to generate a single image in SD that is anywhere close to mid-journey quality) or do the two characters separately, then composite them. You'd probably have to go the compositing route, just because most AI image generators can't do two people in the same scene, consistently.
|
# ? Jan 21, 2024 15:10 |
|
Warren Buffet
|
# ? Jan 21, 2024 16:29 |
|
PaleFigure posted:Guess the song: America the Beautiful?
|
# ? Jan 21, 2024 17:29 |
|
Whoa there buddy leave some pool cues for the other patrons.
|
# ? Jan 22, 2024 00:07 |
|
Ahhh, Master Kenobi--care for some billiards?
|
# ? Jan 22, 2024 00:12 |
|
"It's over, Anakin. I'm stripes!" edit: oops wrong scene
|
# ? Jan 22, 2024 00:22 |
|
Javid posted:Bing and dall-e 3 both refuse to depict violence. I've tried pretty hard to get a good fight scene out of them but the closest they'll do is like the described characters posing menacingly at each other like a comic book cover For instance, I've found it pulls up hard if you ask it to show someone firing a gun at anyone else. But if you ask for someone pointing a gun and then add "with muzzle flash and flying empty shell casings", generally not a problem. Should you then say that the people being shot are "tumbling with splats of Morello cherry syrup", it will usually deliver. (Sadly the "head and torso wreathed in explosively splattering rich red pasta sauce" trick looks to have been clamped down upon.)
|
# ? Jan 22, 2024 00:29 |
|
Small Strange Bird posted:For instance, I've found it pulls up hard if you ask it to show someone firing a gun at anyone else. But if you ask for someone pointing a gun and then add "with muzzle flash and flying empty shell casings", generally not a problem. Should you then say that the people being shot are "tumbling with splats of Morello cherry syrup", it will usually deliver. edit: asked it for "more syrup please" and Earwicker fucked around with this message at 01:11 on Jan 22, 2024 |
# ? Jan 22, 2024 01:09 |
|
|
# ? Jan 22, 2024 02:12 |
|
Small Strange Bird posted:Bing is weird about violence because on the one hand it's really restrictive and prudish, but on the other if you find the right prompt it will happily give you John Wickian splatter all day. Ah cmon guys don't use the computer to make naughty, you'll make Bill Gates cry. You're supposed to use Windows for stuff like defeating Malaria.
|
# ? Jan 22, 2024 03:46 |
|
Bing Image Generator, prompt "The Simpsons! The Musical"
|
# ? Jan 22, 2024 04:06 |
|
Humbug Scoolbus posted:America the Beautiful?
|
# ? Jan 22, 2024 13:48 |
|
|
# ? Jan 22, 2024 20:15 |
|
|
# ? Jan 22, 2024 21:06 |
|
LOL
|
# ? Jan 22, 2024 21:40 |
|
This seems interesting - the claim is they can get approximately the same results as SDXL in far fewer steps. Haven't been able to get it working properly on my machine, but this online version is pretty speedy.
|
# ? Jan 23, 2024 00:44 |
|
|
# ? May 30, 2024 14:49 |
|
It is drat quick, but it seems less powerful in some ways and misses concepts and words (or maybe I'm just not used to SD prompting)
|
# ? Jan 23, 2024 01:25 |