My LoRAs

My LoRAs

LoRAs (Low-Rank Adaptations) are a technique used to fine-tune models with specific styles or concepts. Here are some of mine.

Flux
LoRAs trained for Flux.
NoobAI LoRAs
NoobAI is a finetune of Illustrious using the full Danbooru and e621 datasets. There are two different versions of the model, one with eps and one with v-prediction. I’m currently training for the v-pred 0.65S version unless otherwise specified.
Pony Diffusion V6 LoRAs
Pony Diffusion V6 XL is a specialized Stable Diffusion XL model optimized for generating anthropomorphic and feral creature artwork, with particular expertise in the ‘My Little Pony’ aesthetic. Built on the SDXL architecture, it combines high-resolution output capabilities with precise character generation, supporting both SFW and NSFW content creation. The model excels in maintaining consistent character designs while offering flexibility across various artistic styles, making it a valuable tool for both fan artists and general anthropomorphic content creators. This list contains all the LoRAs I have made for this model.
Stable Diffusion 3.5 Large
Stable Diffusion 3.5 Large represents a significant advancement in the Stable Diffusion model family, building upon the success of its predecessors. This model introduces improved image generation capabilities with enhanced understanding of text prompts, better composition, and more accurate rendering of human features including hands and faces. It utilizes a larger parameter count and more sophisticated architecture compared to earlier versions, resulting in higher quality outputs and better coherence in complex scenes. The model demonstrates particular strengths in artistic compositions, and handling intricate details in both natural and synthetic scenes. It maintains SDXL’s base resolution capabilities while offering improved prompt adherence and more consistent style reproduction across different types of generations.
CompassMix XL Lightning LoRAs
CompassMix Lightning is a refined SDXL-based model developed by Lodestone for frosting.ai, designed to deliver high-quality image generation in significantly fewer steps than traditional models. Distinguished by its ability to produce detailed outputs in just 8-16 steps while maintaining SDXL’s high resolution standards, it strikes an optimal balance between generation speed and image quality. The model features enhanced prompt adherence and supports ControlNet compatibility, making it particularly valuable for users who need quick yet high-quality image generation capabilities. This list contains all the LoRAs I have released for this model.

Related Posts

docs

My Documents

LoRA Training Guide

The LoRA Training Guide explains Low-Rank Adaptation (LoRA), a technique for fine-tuning large language and diffusion models efficiently by introducing small, trainable low-rank matrices instead of modifying all model parameters. This approach keeps the original model weights frozen and injects two additional matrices into each layer to learn necessary adjustments. LoRA is lightweight, making it feasible to train multiple adaptations without hefty storage requirements. The guide also compares LoRA with LyCORIS, an advanced extension that offers more control and flexibility, and introduces LoKr, which uses Kronecker products for matrix decomposition, enhancing memory efficiency and control over the adaptation process.

Custom Nodes