abstract art of a beautiful portrait, solid shapes, geometric shapes, neotokyo colors, muted colors, pixar, artstation, greg rutkowski, samdoesart, ge

AI artwork generated with Stable Diffusion AI v1.4

 

Prompt: "abstract art of a beautiful portrait, solid shapes, geometric shapes, neotokyo colors, muted colors, pixar, artstation, greg rutkowski, samdoesart, ge "

 

Seed: 8356453

CFG usually either 7, or 14

 

Upscaled with Topaz Gigapixel AI:

- Upscale: 2x

- Mode: Standard

- De-noising: 0

- Blur Reduction: 20

- Face recovery: 10

 

My Workflow:

- Generate hundreds of 10-20 step "rough draft" images in Google Colab

- Grab the seeds of the good-ish ones

- Rerun in Google Colab with 50-100 steps using the seeds

- Send the good images to Gigapixel AI, export with Standard Mode settings

- Finally, send them through Sharpen AI, often with "Too soft - Normal" and 30-70 added grain. This step isn't that crucial, but it does help a tiny bit.

- Post to flikr with JS scripts to automate this description :)

- Probably about 1%-5% of AI generations get posted to Flickr. It depends a lot on the prompt, different prompts run into different issues. Bust portraits tend to be more successful since hands, arms, and legs almost always come out wrong.

 

Start creating right away with the official Stable Diffusion web UI: beta.dreamstudio.ai/

You get 200 free credits. Alternatively, find a Google Colab to make images at a cheaper rate. My personal one is too personalized to publicly share.

Join the Stable Diffusion community on Discord: discord.gg/stablediffusion

 

Almost all generations use v1.4 Stable Diffusion. However, a couple anime-related images may instead use the anime-trained Stable Diffusion model, "Waifu Diffusion".

 

Done