AI artists bring back Greg Rutkowski with LoRA

This is a free LoRA fine-tune hosted on Civitai, built to bring digital artist Greg Rutkowski's surreal fantasy aesthetic right back into Stable Diffusion.

Rutkowski previously campaigned aggressively against the use of his work in AI training sets. In response, Stability AI removed his style when releasing Stable Diffusion 2.0. This community-made LoRA effectively undoes that removal by injecting the look back into open-source workflows.

What it does

During the heyday of Stable Diffusion 1.4 and 1.5, Rutkowski was invoked in text prompts over 400,000 times—surpassing legends like Picasso and Da Vinci. Adding his name became a cheat code for instant, high-quality digital oil paintings.

When Stability AI stripped out his style in SD 2.0, users widely complained that the model felt nerfed. By dropping this dedicated LoRA into your generation pipeline, you can recreate his dynamic lighting, dark fantasy compositions, and classical brushwork without complex prompt engineering.

How it works and where to get it

The model was trained by Lykon, a prominent creator known for early customized SDXL builds. Rather than retraining a full base model, LoRA freezes the underlying network and trains a lightweight adapter specifically focused on reproducing a single style, subject, or palette.

You can download the weights directly from Civitai at no cost. Because LoRAs are tiny files, they require negligible extra VRAM and run smoothly inside standard WebUI or ComfyUI setups on modest consumer GPUs.

Using it, however, comes with major ethical baggage. Rutkowski has fiercely opposed AI imagery, noting that he reached a point where he could barely distinguish AI outputs from his genuine catalog. Lykon stated that he would take the model down if Rutkowski asks him directly, though he argued that having an accurate immortal depiction of the style benefits everyone. The Civitai and Reddit communities remain split, with critics reminding peers that legal does not always mean ethical.

Why it matters

For digital creators and prompt engineers, this tool proves how easily small adapter models can resurrect removed aesthetics and capture complex brush dynamics on consumer hardware.

We think you should be extremely cautious about using it for client work or public projects, though. While it highlights how impossible it is to enforce artist opt-outs in decentralized open source, leaning on a model trained against a living artist's explicit wishes carries undeniable reputational risk.