---
tags:
- text-to-image
- stable-diffusion
- lora
- diffusers
- template:sd-lora
widget:
- text: >-
    CRCLG style, abstract graphic design, very sad, a young boy standing, an
    arcade cabinet engulfed in flames, vibrant colors on black background,
    detailed brushstrokes, expressive painting, micro detailed texture painting
  parameters:
    negative_prompt: >-
      poor composition, low quality, disfigured, chinese, asian, korean, photo,
      dull, fake, boring, cheap, sfw, wholesome, watermark
  output:
    url: images/cryptocollege-_00132_.png
- text: '-'
  output:
    url: images/cryptocollege-_00125_.png
- text: '-'
  output:
    url: images/cryptocollege-_00120_.png
- text: '-'
  output:
    url: images/cryptocollege-_00079_.png
- text: '-'
  output:
    url: images/cryptocollege-_00073_.png
- text: '-'
  output:
    url: images/cryptocollege-_00070_.png
- text: '-'
  output:
    url: images/cryptocollege-_00058_.png
- text: '-'
  output:
    url: images/cryptocollege-_00037_.png
- text: '-'
  output:
    url: images/cryptocollege-_00026_.png
- text: '-'
  output:
    url: images/cryptocollege-_00033_.png
- text: '-'
  output:
    url: images/cryptocollege-_00051_.png
base_model: stabilityai/stable-diffusion-xl-base-1.0
instance_prompt: CRCLG style
license: artistic-2.0
---
# CryptotoCollege Style

<Gallery />

## Model description 

Cryptotocollege Style
A very expressive painting style LoRA trained on 48 images scraped (with their blessing) from the X account @Cryptotocollege. While the style of their work varies, I chose their loose and expressive painting style that often features objects engulfed in flames and characters smoking cigarettes. I captioned the images via Claude3 API using my special art captioning prompt framework. I trained for 40 epochs with a batch size of 4 and a gradient accumulation of 2 using the prodigy optimizer. After reviewing test outputs and testing several epochs i chose epoch 26 to share. This LoRA doesn&#39;t quite capture the extreme texture that some of their work displays but it can be prompted toward that. This LoRA likes to create characters smoking cigarettes (even dogs and children) unprompted and will often incorporate flames into the scene.

Make sure to follow @Cryptotocollege on X

Since i&#39;ve been cancelled off X, you can find me on farcaster as @BlaireSilver13

## Trigger words

You should use `CRCLG style` to trigger the image generation.


## Download model

Weights for this model are available in Safetensors format.

[Download](/BlaireSilver13/Cryptotocollege_style/tree/main) them in the Files & versions tab.