Abhorrent LoRA - Body Horror Monsters for Qwen Image
A new LoRA adapter lets users generate grotesque, malformed flesh monsters with specific anatomical prompts.
An independent developer has released 'Abhorrent LoRA,' a specialized adapter that transforms Alibaba's Qwen Image model into a generator for grotesque, customizable body horror monsters. The LoRA (Low-Rank Adaptation) fine-tunes the base model to produce malformed blobs of human flesh that respond to specific anatomical prompts. Users can generate creatures with multiple animal heads, tentacle masses, wings, or cloacas by including the trigger word 'abhorrent' in their prompts, treating it as a noun (e.g., 'The abhorrent is eating a birthday cake').
The creator, known as ThePoetPyronius, developed this as their second LoRA project after learning from issues with their first attempt—a punch pose adapter that struggled when transferred between models due to dataset and captioning problems. While the current version successfully creates detailed horror imagery, it suffers from doubling artifacts when generating images above 2000 pixels. The developer recommends generating at lower resolutions and upscaling while seeking community advice to fix the technical issue. This project demonstrates how specialized LoRAs can expand creative possibilities for open-source image models like Qwen.
- Uses trigger word 'abhorrent' to generate customizable body horror monsters with Qwen Image
- Creator's second LoRA project with improved dataset sampling and captioning techniques
- Currently has doubling artifacts above 2000-pixel generations, requiring upscaling workarounds
Why It Matters
Shows how specialized LoRAs can expand creative applications for open-source AI image models beyond standard generation.