/SDG/ - Stable Diffusion General

Prompt: 'portrait of a redhead female busty tankoban cowgirl, muscular, upper body, D&D, huge tits, cinematic lighting, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by Artgerm and Greg Rutkowski and Alphonse Mucha' * 0.6 + 0.5 * 'Emma Watson'
seed: 2408829516
Prompt mixing can really fuck things up but it's working to tame the Emma Watson vectors.
I modded it into the other user's script()
filebin.net/2etz10f7n2zz3mnm
I just noticed that I fucked up the prompt and went over 1.0 total but it didn't blow up so whatever.

Attached: emmatest.png (1027x514, 1.02M)

nevermind, it always saves to the img2img folder instead of the txt2img one

So tattoos are easy, anyone found the prompts for body jewelry/piercings?

forgot to say but you need to install lark' with pip.

Attached: download (7).png (512x512, 372.33K)

Oh shit sorry, I fucked up the default, around line 153 should be this:
default=txt2img_grid_output_dir or "outputs/txt2img-samples"

author removed it(finally)

is there a way to bootstrap this to make moving pictures yet?

more emma
using asuka in a yellow dress as a source image

Attached: grid-0068.png (516x2572, 1.46M)

That first one is just beauty and the beast

Attached: emma_watson_as_hermione_granger_clean_cel_shaded_vector_art._shutterstock._behance_hd_by_lois_van_baarle_artgerm_helen_huang_by_makoto_shinkai_and_ilya_kuvshi_-S_4145385614_ts-1660617415_idx-0.png (512x512, 374.16K)

So I should just ignore step altogether?

I wish I could generate at 1024x1024. The future is looking bright.

Attached: index.png (512x512, 501.19K)

I find this very useful base blocks of my prompts that I take inspiration from art that's already out there and just build upon it.
colab.research.google.com/gist/aredden/5028802c3824170150af9eaee91db358/clip_interrogator-w-checkpointing-adjectives.ipynb#scrollTo=G_Lo91Kktd2j

Wait, can you weight parts of the prompt as standard or do you need the modified script?
What is the syntax for doing it 3-4-n ways?

Some quick tips for using SD

Attached: stable diffusion tips.png (1871x767, 163.99K)

is the hidamari diffusion colab updated with the new model? if not, is there a colab that is updated?

how hard would it be to use text to AI image generation then run that through to generate a next "frame" and then use another gpt-3 or even 4 when it comes out AI audio and run then through eachother for say a short 30 second video and audio totally AI generated?

i think disco diffusion has a video option already, never tried it before because it takes so long

What dimensions are allowed? It keeps failing if I put in something besides 512x512 or 384x384. Does it have to be in multiples of 128?

64

multiples of 64

By 2025 you're going to be watching feature length films made with this tech

Attached: film_still._screenshot._young_woman._sitting_at_a_table._holding_a_newspaper._extreme_close_up._focused_on_newspaper._in_public_cafeteria._-W_704_-C_10.0_-n_5_-i_-S_2339853981_ts-1660275332_idx-4.png (704x512, 438.77K)

Need the modified script.
'my perfect prompt' * 0.9 + 0.1 * ('anime' * 1.0 - 'painting' * 1.0) is the only example that the original author of the code gave.
It's easily modded into the normal scripts though.
Original here:
github.com/dogewanwan/stable-diffusion

sweet, what about video to audio?

There was an user that listed like 6 dimensions that worked flawlessly on the last thread

8, see colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_diffusion.ipynb#scrollTo=uf9pbS3kCsUf

Let's try different actresses:
Chloe Moretz
Jennifer Connelly
Ellen Page
Sigourney Weaver
Denise Richards
Milla Jovovich
Angelina Jolie
Brooke Shields
Jessica Alba

ok but what about right now is there a video to audio AI?

What do you mean by video to audio?
Like you put in a silent video and it generates what it thinks the audio tracks should be?

do I need to rebuild the entire thing into a different environment like going from Stable > Waifu/k/gui or are there specific scripts that just need swapping out/editing

nice
semi related, stability ai are working on harmonai sound/music generation, not sure of the current state of it
harmonai.org/

Attached: 01267-1430560134_film_still._young_woman._sitting_in_a_train_carriage._holding_a_newspaper._wide_shot._side_angle..png (704x512, 512.94K)

yes exactly that

bless you

but then why tf do I keep getting this, even with multiples of 8? E.g. even with the 768x512 like on the site you linked.

>RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 24 but got size 23 for tensor number 1 in the list.

did you change the C or f vars?

Yeah okay that's not going to be for a while yet user.
Something like that is probably closer to 2030 than 2025.
On the more immediate side of things just posted the most promising AI music model currently in production.
Supposedly the beta is next month.

If you're already using the one with gradio, just put that zip's contents into the scripts folder, 'pip install lark' and 'pip install pynvml'.
I didn't include the other user's restart script so it'll error if it runs out if ram but that should be fine.

Nah it runs fine with 512x512

thanks I'll have a play with this later. Maybe I can finally force it to give the responses wanted.