All the content I make is and will be completely free on CivitAI.
But if you want to support me with a coin, then you can do it here.
Trained on AbyssOrangeMix2 with clip skip 2.
Recommended embeddings:
1. easynegative
2. bad-hands-5
Recommended extension:
1. Pixelization
Strength 1 - 0.6 is optimal.
PLEASE CHECK YOUR LORA NAME AFTER DOWNLOAD.
There is no trigger word, but i highly recommend to use tag pixel art in the end of your prompt.
If you want a full body sprite image, then you should go with this structure: full body of CHARACTER, very detailed, realistic, simple background, transparent background, pixel art
This lora is resolution and hires.fix friendly, so feel free to experiment and share your feedback. Enjoy!
Description
FAQ
Comments (28)
What is the master model you are using?
I just tried to generate images using your settings and abyssorangemix2_hard, it doesn't look anything like your images. It's still pretty standard looking, like an illustration.
Did you used clip skip 2? Upload some pics, i will try to check your generation info and help.
Ok I see the problem, civitai has renamed the LORA, it didn't match your prompt.
After changing the lora name, it works now.
Did you use Capcom sprites? Like the Marvel and Street Fighter games?
@sdf34gdssdg4 Yeah, i wrote about lora name in the description. I really don't remember which sprites i used, cause there was about 100+ of them, so capcom sprites maybe was included.
@sdf34gdssdg4 what name should lora be?
what name should lora be?
@SteelmanStar you may rename lora file whatever you like, but you need to recall it your prompt with the same exact name if you wnat to use it. So basically, if on my PC file of this lora is named as "sxz-pxl-v1", then in prompt i should use it as <lora:sxz-pxl-v1:1>. It means, that lora file name and lora name in prompt tab should match. Nothing complicated here.
So after I Install the Github thing
https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization
And place the 3 checkpoints into it, what next?
How do I use it? or does it just use it on everything automatically and I should just activate it for your model, and disable it if I dont want to use the Pixel Extension, you didn't really explain this and how to use it/ not use it.
First of all, i am not a creator of this extension, i just mentioned it in description for post-processing stuff.
Second, if you did everything right with this extension it should appear in Extra tab with new slider and checkboxes.
And finally, lora works fine itself without this extension. Extension is quite optional.
@sadxzero Sorry, I really don't know where to put these three files, can you help me?
@yang9361 stable-diffusion-webui/extensions/stable-diffusion-webui-pixelization/checlpoints
Its said just above download links in github
This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument
Did you use a particular VAE ? I can't quite reproduce your images, especially for the faces that are often poorly detailed/blurred, even if I use the same prompts and parameters...
I'm using this one. Did you use hires.fix? I highly recommend to use it with x1.5 or x2 multiplier.
@sadxzero Thank you for your answer, I'll try to use that and hires fix. Also, did you have to change the '.ckpt' to '.pt' so that it's considered to be vae and not a model ?
@cibaxih nah, i just put it as it is in "models/VAE" folder
Had the same issue~ might worth mention this in the description~
I really like this style you have, and I want to make the same. I'm really new to this, and since I have a slow PC I have to run it on Google Colab. Could you help me set it up so I can get the same results? It's totally fine if not :)
Hi @lotfisunni197 you may contact me in discord (sadxzero#2715) or telegram (username_v2) where i can share some screenshots ant try to help you.
After I learned to use XYZ plot, I finally knew that besides prompt, the secret of satisfying my heart lies in the Sampling method and weight~~
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.