GlitchHunter is Flux dev Lora trained on my outputs from MJ and SDXL. Focus is on unusual erratic characters. It is trained on few types of styles and merges them really well into new forms. It carries style into other subjects pretty well too. Biased on neon cyberpunk comic style.Training data includes part of my <a target="_blank" rel="ugc" href="https://magiceden.io/launchpad/supernatural_creatures">Supernatural Creatures</a> animation collection. You can find more info about the collection and the process behind it in this <a target="_blank" rel="ugc" href="https://medium.com/@glitchcandies/supernatural-creatures-fcb993254d04">medium article</a>. Trained on 52 images. Dataset styles includes: - abstract comic (2D) - toon shader like kind of 2.5D - designer toy inspired mecha sculptures for 3D To get those styles type in "3D", "comic", into prompts. Dataset was captioned using gpt-4o-mini and my manual input. It likes simple prompts and transfers styles pretty well. Prompt example: A vivid, surreal mecha creature with colorful flora and mushrooms with whimsical shapes emerges, blending nature, metallic like monster, 3D imagination in harmonious chaos. Settings: CFG: 1, 28 or 38 steps. Trigger word: [GlitchHunter] Trained using Glif Loradex while beta testing the platform. Licensed CC0.

GlitchHunter

Initial version trained on 2300 steps.

Files with SHA256: 85ae5861703c5d1fafc9d863bca15a71067cb066a2bee6d49184744017b00499