ad astra per asperga 
(added colab instructions)
Colab:
-
Open automatic1111's colab: https://colab.research.google.com/drive/1Iy-xW9t1-OQWhb0hNxueGij8phCyluOh
-
Replace this cell:
#@title Normal 1.4 model
# get a token from https://huggingface.co/settings/tokens
user_token = "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" #@param {type:"string"}
user_header = f"\"Authorization: Bearer {user_token}\""
!wget --header={user_header} https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt -O model.ckpt
With:
!apt install megatools
!megadl "https://mega.co.nz/#!FxclSKoL!EFSM4nLlXMuOvBLkoZNmtOH4Y8oycjrU7h2Hn6mKl1k"
-
Run (it'll take a few minutes to download)
-
Refer to Marsey as
a sks catin your prompt.a photo of a sks cat as a gardeneretc
Download the weights: https://models.rdra.ma/
Development on Stable Diffusion has been happening at a wild pace, and there's already written a working implementation of DreamBooth for it. Whereas Textual Inversion refines your prompt, DreamBooth is able to finetune the model itself to teach it new concepts. I did one run with 87 Marseys and another with just 3 upscaled ones. So far the upscaled run (u1000.ckpt) is looking best.
See the Textual Inversion posts for a comparison with that. DreamBooth is far better at keeping her colors consistent and producing recognizable images. Decent results were something like 1 in 50 before, now they're nearly every image.
a photo of Marsey
a photo of a spaceman Marsey in outer space
a photo of Marsey as a gardener
I love that it creates real Marseys too 
a photo of Marsey running away from a dinosaur
Missing the 
. Might be overfitting and improve with better regularization in DreamBooth.
a photo of Marsey cooking in the kitchen
These models also work with Deforum to create animations:
There's still plenty of low-hanging fruit to suggest these will get better quickly - the DreamBooth repo doesn't do regularization correctly, and we'll get a free bump in quality when SDv1.5 is released. The plague of AIposting continues
Jump in the discussion.
No email address required.
my mind is literally blown right now, this is insane. it has the style, the structure, the shape perfect.
currently downloading the model, can you just swap out the original SD model with the marsey model? if so I could use AUTOMATIC1111's webui
also, why is the model so big? its like 3x larger than the SD model lmao
also also, models.rdra.ma, when did that happen, lol
edit: column 2 row 3 is actually really cute
Jump in the discussion.
No email address required.
Yeah it's a drop in replacement so automatic's UI should work. The extra size is from the "ema" version of the SD model which has extra metadata for training
I was high one night and got rdra.ma off njalla, I'm glad I'm getting some use out of it lmao
Jump in the discussion.
No email address required.
Can you use the usual pruning script on that model?
I'm talking about shit like https://github.com/harubaru/waifu-diffusion/blob/main/scripts/prune.py
I use a "custom" .bat (https://pastebin.com/R15jTB8U) .py (https://pastebin.com/W4mrB0nf) where you just have to edit the last line of the .py for your model. I think it needs to be in the root webui folder but I'm not certain. It works on Waifu-Diffusion so it might work on Marsey-Diffusion
Jump in the discussion.
No email address required.
Oh perfect, thanks, I'll give that a try in the future
Jump in the discussion.
No email address required.
Also, if you want to do crossovers between models (like waifu-diffusion and marsey-diffusion for anime marseys), it's dead easy to do, you just need to add the weights lmao
There's an automated script .bat (https://pastebin.com/0J9TPEcs) along with the .py (https://pastebin.com/056xeuVm). This one doesn't require changing anything, just type what's needed in cmd.
It will ask you for a ratio between the two models, there's no general rule on merges yet so just guess.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
based af
Jump in the discussion.
No email address required.
Darn that's a LOT of unused shit, most models "only" get halved in size
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context
More options
Context