This is the result of a fun experiment: I fused the DMD2 LoRA into a model I really like (the original checkpoint can be found here: Dollamor) and then quantized it to Q5_1.

I actually managed to run this model on a decade-old laptop with a GTX 960M, so it is extremely lightweight and fast. Just keep in mind: because of the distillation process, do not expect the exact same quality as the original model.

Recommended Settings

• Steps: 8

• CFG: 1.3 - 1.8

• Sampler: LCM

• Scheduler: Simple or Normal

Additional Instructions

• Please use the CLIP L and CLIP G models, as well as the provided VAE.

• A ComfyUI workflow is also included in the files. Note: You will need the appropriate custom nodes to load GGUF models to use it.