|
VectorSigma posted:computer, give me an excuse for being late to work ... gonna need a few thousand samples of doctor's signatures for a project
|
# ? Sep 14, 2022 04:59 |
|
|
# ? May 30, 2024 13:29 |
|
the magrathean planetyards
|
# ? Sep 14, 2022 17:33 |
|
Awesome. Sometimes I get caught up in the details, and forget these are honest to God dream printers.
|
# ? Sep 14, 2022 17:36 |
|
More progress, I'm now 27 epochs deep in the PC-98 hole and have some initial results. So far it seems to have mostly just learned how to make vaguely 90s era anime. Some highlights so far: Also I discovered that if your subject is extremely represented in the training set (like the Mona Lisa) that turning down the CFG scale actually helps reproduce the right output. Otherwise you get what I have dubbed "color frying" where it overshoots and the whole thing goes red/yellow/magenta. For Mona Lisa specifically I had to put the CFG scale all the way down to 2-3 (from 7!) to get the approximate painting back out. Full folder here: https://imgur.com/a/MObDoOa
|
# ? Sep 14, 2022 23:09 |
|
that's badass i wish someone would do a tune job on pixel art, it's maybe the one thing SD can't do well vs Dall-E 2
|
# ? Sep 15, 2022 02:48 |
|
BARONS CYBER SKULL posted:that's badass Try this: Positive prompt = pixelated BARONS CYBER SKULL, pixel perfect, art by paul kelpe, intricate, elegant, highly detailed, artstation, sharp focus Negative prompt = ((rounded corners)), messy, (childish), bad attempt, bland, plain, pencil, warped, smooth, blending, deformed, ugly, blurry, noisy, grunge, 3d, amateur, dirt, fabric, checkered, sloppy, scratchy, monotone, duotone, muted, vintage, lacking, washed out, muddy, ((((compressed)))), paper texture, mutilated, mutation, mutated The biggest change seems to always be by prefixing "pixelated" as the first word, no amount of weighting it later in the prompt seems to help as much. Also Paul Kelpe is a cubist artist I had never heard of, I found him by using the img2prompt interrogator on a bunch of pixel art and his name kept popping up. No idea why but it really does help. Also, mentioning things like "8bit" "16bit", "NES", "SNES", etc. universally turns out garbage prompts so avoid those for the 1.4 model at least. Maybe it will be better later. Objective Action fucked around with this message at 04:10 on Sep 15, 2022 |
# ? Sep 15, 2022 04:08 |
|
So when do we get high quality AI audio from text prompts? I want a nice easy source for samples like “avalanche of resonant metal tubes” “psychedelic sitars, discordant, seamless drone” “amen break played by explosions and small arms fire”
|
# ? Sep 15, 2022 04:12 |
|
fun with img2img
|
# ? Sep 15, 2022 04:26 |
|
Snowy posted:So when do we get high quality AI audio from text prompts? I feel like we're not all that many years away from AI-generated music that is indistinguishable from human-composed and performed music, including lyrics/vocals. That's weirdly unsettling to me in a way that the exact same thing with visual art isn't.
|
# ? Sep 15, 2022 05:17 |
|
WhiteHowler posted:I feel like we're not all that many years away from AI-generated music that is indistinguishable from human-composed and performed music, including lyrics/vocals. That's weirdly unsettling to me in a way that the exact same thing with visual art isn't. OpenAI published a song thing in 2020. It's kinda rough but close enough to good that I'm surprised we aren't there now 2 years later. You can read about it and listen to some samples here: https://openai.com/blog/jukebox/
|
# ? Sep 15, 2022 05:28 |
|
WhiteHowler posted:I feel like we're not all that many years away from AI-generated music that is indistinguishable from human-composed and performed music, including lyrics/vocals. That's weirdly unsettling to me in a way that the exact same thing with visual art isn't. AI music is basically already here. Its just not easily accessible to the public at the moment.
|
# ? Sep 15, 2022 05:31 |
|
I spent quite a while playing with AIVA (AI midi generation) a few months ago and it's just ok. Like our eyes are much better at smoothing over and filling in AI glitches in images, than our ears are at doing the same for music.
|
# ? Sep 15, 2022 05:40 |
|
https://www.youtube.com/watch?v=bkZrSAZEIeY
|
# ? Sep 15, 2022 05:48 |
|
Mola Yam posted:I spent quite a while playing with AIVA (AI midi generation) a few months ago and it's just ok. At the same time, music is much easier to reduce down to simple numbers and patterns. Garage band has already been doing a lot of this for like a decade. The only reason ai music isn't light years ahead of visuals is just a matter of attention and focus to develop it.
|
# ? Sep 15, 2022 05:49 |
|
I was pretty drat pleased with that old infinite jukebox thing where it analysed a song and looked for parts that could be flawlessly looped, so your favourite song could be played continuously in a new way for as long as it took for you to hate it. Anything even vaguely superior to that is just cream on top. e: oh hey, it still lives! https://eternalbox.dev/jukebox_index.html Gromit fucked around with this message at 05:52 on Sep 15, 2022 |
# ? Sep 15, 2022 05:49 |
|
Objective Action posted:Try this: not exactly pixel art but still rad
|
# ? Sep 15, 2022 06:06 |
|
IShallRiseAgain posted:AI music is basically already here. Its just not easily accessible to the public at the moment. Microsoft tried to launch Songsmith back in the 00s. It was sold as a tool to quickly put together musical backgrounds from just singning. It was a sonic dumpster fire. MS own commercial. Note the MacBook with a sticker. https://www.youtube.com/watch?v=3oGFogwcx-E Of course people had a lot of fun with it before it was quietly thrown in the memory hole. https://www.youtube.com/watch?v=mg0l7f25bhU axolotl farmer fucked around with this message at 07:15 on Sep 15, 2022 |
# ? Sep 15, 2022 06:34 |
|
axolotl farmer posted:Microsoft tried to launch Songsmith back in the 00s. It was sold as a tool to quickly put together musical backgrounds from just singning. It was a sonic dumpster fire. The fact that it uses the same exact song as the commercial has me rolling
|
# ? Sep 15, 2022 06:51 |
|
New AI on Huggingface: ERNIE-ViLG, a chinese AI that won't let you make pictures of controversial things like Tianmen Square https://huggingface.co/spaces/PaddlePaddle/ERNIE-ViLG
|
# ? Sep 15, 2022 07:14 |
|
Owns
|
# ? Sep 15, 2022 07:19 |
|
axolotl farmer posted:Microsoft tried to launch Songsmith back in the 00s. It was sold as a tool to quickly put together musical backgrounds from just singning. It was a sonic dumpster fire. I get it that it's more advanced cuz it's automatically matching the singing but I just hear Play It! In Microsoft Plus for Kids https://www.youtube.com/watch?v=xr8tBxAvW_c&t=1868s
|
# ? Sep 15, 2022 07:47 |
|
IShallRiseAgain posted:AI music is basically already here. Its just not easily accessible to the public at the moment. Marvel at the wonders of AI music! https://www.youtube.com/watch?v=xBaA-iTYwi4
|
# ? Sep 15, 2022 09:39 |
|
The world is not ready for the wonder that is AI music. High effort attempt at putting Eurovision out of business: https://www.youtube.com/watch?v=4MKAf6YX_7M Songsmith: https://www.youtube.com/watch?v=22AWPW5s4EA https://www.youtube.com/watch?v=WmC28cXWqLc
|
# ? Sep 15, 2022 09:56 |
|
The only thing holding AI music back is the same thing that was holding back AI image gen: the corps who have all the money SD was 800,000 USD worth of training given out for free, you cannot imagine how insanely mad that’s making the tech assholes who run “Open” AI Now, imagine you are the already astonishingly shady recording industry and how badly you absolutely do not loving want something like SD for music being out there
|
# ? Sep 15, 2022 10:20 |
|
yeah take a look the price on music gear and software and it become obvious why free music AI isn't widely available. there are lots of very fancy synth plugins to buy though!
|
# ? Sep 15, 2022 10:43 |
|
BARONS CYBER SKULL posted:The only thing holding AI music back is the same thing that was holding back AI image gen: the corps who have all the money This is an extremely good point Analytic Engine fucked around with this message at 01:40 on Aug 20, 2023 |
# ? Sep 15, 2022 10:59 |
|
BARONS CYBER SKULL posted:The only thing holding AI music back is the same thing that was holding back AI image gen: the corps who have all the money im fuzzy on SD's history, but isn't it something like a wealthy libertarian techbro going "woooo free internet! free movement of ideas! Here I spent 1 million on this pet project and now am releasing it for free for the world to enjoy!" while his peers looks at him dubiously thinking "dude, you can say it but you should never do it"? So basically we lucked out thanks to a good idiot?
|
# ? Sep 15, 2022 11:23 |
|
Popoto posted:im fuzzy on SD's history, but isn't it something like a wealthy libertarian techbro going "woooo free internet! free movement of ideas! Here I spent 1 million on this pet project and now am releasing it for free for the world to enjoy!" while his peers looks at him dubiously thinking "dude, you can say it but you should never do it"? he's very, very stupid, yes
|
# ? Sep 15, 2022 11:39 |
|
Objective Action posted:Try this:
|
# ? Sep 15, 2022 12:17 |
|
Made a dumb sizzle reel of all my June - August midjourney prompts https://www.youtube.com/watch?v=H_rPg2dvTIY I was gonna say "drat I need a hobby" but I guess this is my hobby now Elotana fucked around with this message at 15:22 on Sep 15, 2022 |
# ? Sep 15, 2022 13:14 |
|
ornate intricate filigree framed, vampire elf wearing ornate intricate detailed carved stained glass rose (((armor))), determined face, heavy makeup, covered in blood, blood stained clothing, led runes, inky swirling mist, gemstones, ((red pool background)), ((eyeshadow)), (angry), detailed, intricate,(Alphonse Mucha), (Charlie Bowater), (Daniel Ridgway Knight), (Albert Lynch), (Richard S. Johnson) Negative prompt: ugly, fat, obese, chubby, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), messy drawing, large_breasts, penis, nose, eyes, lips, eyelashes, text, Steps: 30, Sampler: Euler a using the waifu diffusion model
|
# ? Sep 15, 2022 14:10 |
|
Alpaca labs sending out invites
|
# ? Sep 15, 2022 17:19 |
|
Do the brackets in prompts add emphasis?
|
# ? Sep 15, 2022 18:18 |
|
Jonny Nox posted:Do the brackets in prompts add emphasis? on the AUTOMATIC1111 model, each set of parentheses are supposed to make it pay 10% more attention, and brackets make it pay 10% less* attention *It's actually multiplying/dividing by 1.1
|
# ? Sep 15, 2022 18:23 |
|
has that been verified by locking a seed and experimenting? SD-derived AIs seem to have a ton of trouble paying attention to more than a few words and i don't know that the ((())) weighting ever turned out to be worth a drat on the stock model (it always seemed to be strict left-to-right diminishing priority)
|
# ? Sep 15, 2022 18:48 |
|
Elotana posted:has that been verified by locking a seed and experimenting? SD-derived AIs seem to have a ton of trouble paying attention to more than a few words and i don't know that the ((())) weighting ever turned out to be worth a drat on the stock model (it always seemed to be strict left-to-right diminishing priority) It's a feature specifically added to the parser in this repo
|
# ? Sep 15, 2022 19:00 |
Elotana posted:has that been verified by locking a seed and experimenting? SD-derived AIs seem to have a ton of trouble paying attention to more than a few words and i don't know that the ((())) weighting ever turned out to be worth a drat on the stock model (it always seemed to be strict left-to-right diminishing priority) By default the power of the parentheses are just a myth, but the AUTOMATIC repo parses the prompt for the parentheses and manually manipulates the tokens.
|
|
# ? Sep 15, 2022 19:00 |
|
Specifically it goes from left ot right and multiplies on (] and divides on [) so you can also boost visibility by doing ]this[ like a psycho.
|
# ? Sep 15, 2022 19:03 |
|
Using brackets (especially in multiples) for that functionality strikes me as inelegant. I liked the other fork that just used numbers to denote weight (it didn’t have negative prompts, though).
|
# ? Sep 15, 2022 19:06 |
|
|
# ? May 30, 2024 13:29 |
|
Honestly 1.1 is way too small anyway
|
# ? Sep 15, 2022 19:10 |