I generated this image with a 12gb 3060 and 16gb of system RAM
Prompt for the main image is:
gritty graphic novel art, a warrior holding up his hand which has turned into the head of a fire breathing dragon and shouting “forefathers one and all, bear witness!” speech bubble
Text is a lot easier on bing/dalle 3, for stable diffusion, its better to just do it manually, or use controlnet. There are loras like this one though: https://civitai.com/models/176555/harrlogos-xl-finally-custom-text-generation-in-sd
But its mostly for logos, and still takes a bit of prompting to get it to write what you want.
Self Destruct - kublai khan tx
yukikosan - midori (NSFW ALBUM COVER)
All songs that hit hard for different reasons… for me
They change the output. specifically wanted the loras to change the output design so it looks more like a mass of creatures fused together like the thing. For reference, here is the original output from bing:
And here is the bodhor lora on civitai:
edit: I linked the wrong lora before. here is the correct one, I think. I downloaded it some time ago. https://civitai.com/models/51288/body-horror-creatures
“Ghosts having a birthday party” was the thought I had in mind when generating this.
Kevin Bacon does kind of look like a skull
This image was made using Automatic 1111.
Controlnet model is qrcode monster: https://huggingface.co/monster-labs/control_v1p_sd15_qrcode_monster
Settings used is: Pixel Perfect, Balanced, control weight: 1.2
The controlnet image was a photo of skull i found on google that I applied a black and white filter to.
The effect was achieved using qrcode monster controlnet model
prompt: "retro_neon psychedelic neon art half body art portrait of a clean shaven man with short black hair looking up at a brilliant colorful tye-dye sky.front view.digital art, digital painting, "
using these two loras:
https://civitai.com/models/569937/retro-neon-style-fluxsdxlpony
https://civitai.com/models/637230/flux-lora-collection-xlabs (the art one)