Happy cakeday to you!
This model creates custom cakes covered in marzipan, shaped like the prompt.
Usage
Use the trigger word: "CakeStyle" along with the LoRA to use it.
Recommended weights: ~0.55 (0.1-0.7), it's quite overbaked.
Recommended sampler: Euler-a
Use simple prompts
Don't use too many beautifying keywords
Don't use too many negative prompts
Most base models seem to handle shaped cakes pretty well already, which is why this works well with a very low weight. The model gives it more of a creative boost, and produces more varied cakes, and is often more successful at making the custom shape you want.
It works surprisingly well with v1-5 base model. It's clean and simple in it's style, which suits the rough shapes you would expect from a custom shaped cake.
Tips & Tricks
Useful keywords for the prompt:
Marzipan cake shaped like ...
Description
FAQ
Comments (13)
OUTSTANDING SIR! BEAUTIFUL!!!
BEST LORA EVER IN THE HISTORY!
Thank you! Very nice pictures you shared! Gotta try some of those prompts myself, and that model seems sweet!
Yeah I gotta say, this LoRA is awesome, and I'm a connoisseur. You can make a cake out of anything with this. I kinda feel sorry for the poor people who make actual cakes now. LOL They won't be able to compete... but they will have a lot of awesome inspiration.
@parallelepipedon Yeah, visually you may be right. However, I think I'll take the flavor of the real ones over the generated images. At least until we can connect images to our taste stimuli with a neural interface :)
But yeah, as you said, use it as cake inspiration.
Someone please, make that shark cake! It must exist!
Um, well this is special.
@Murdo Aren't we all? :)
Im new to this, im trying to understand what lora is and how to install and use this cake model
Hi! Welcome to AI models :D
There are several different kinds. You will want to start downloading some CHECKPOINTS and LoRAs.
Checkpoints are considered your BASE MODEL. Consider it the engine that drives each image generation. The backbone of it. It often has a style (like realistic, anime, 3D or something like this).
Then you have LoRAs (and many other types). Think of these as add-on models that you put ON TOP of a Checkpoint.
If you are using Automatic 1111 as your Stable Diffusion software, your Checkpoint models will go here:
\stable-diffusion-webui\models\Stable-diffusion
and your LoRA's go here:
\stable-diffusion-webui\models\Lora
The file path is based on where you installed the Automatic 1111 to.
Place the downloaded files in there.
After restarting Automatic 1111 or reloading your models, you can now choose a "Checkpoint" in the drop-down at the top of the screen.
For LoRAs, you need to trigger the keywords needed for them, as well as link to the LoRA. This is done by writing the name of the LoRA-file like this:
<lora:CakeStyle:1>
And with the trigger keyword, it looks like this:
<lora:CakeStyle:1>CakeStyle
In the case for my LoRA. I suggest you go into an image here, and copy the generation data/prompt from there, and start off by using that to get a similar result (it's not gonna be exactly the same unless you get everything perfect).
Here's a video guide: https://www.youtube.com/watch?v=ZHVdNeHZPdc
Good luck!
Would love to see this for XL
I need to figure out how to do XL training on colab. If I do I may give these an XL spin.
But I think that XL does these things pretty well out of the box already. Try using some prompts found on these cakes and maybe you'll get something nice already?
Man, you won the Heaven, this is the BEST LoRA I seen in my life, I can make my waifus transform into cake with this LoRA.
This is also the only model where SD1.5 base model actually turns out good results!
Shitty unprofessional looking cakes actually fit this very well!