EVO 3: v0.2 (SD 2.1 768): evo-car-lora-sd21-768_v02.safetensors

Trained on a dataset of 118 hi-res images (wiith BLIP captions manually edited) with Repeats = 20 and 2360 regularisation images, source model: stabilityai/stable-diffusion-2-1 with v2 and v_parameterization enabled.

Relevant Kohya settings:

• Learning rate: 0.0001

• LR Scheduler: cosine

• LR warmup (% of steps): 10

• Optimizer: AdamW8bit

• Text Encoder learning rate: 0.00005

• Unet learning rate: 0.0001

• Network Rank (Dimension): 128

• Network Alpha: 1

• Max resolution: 768,768

• Clip skip: 1

• Min SNR gamma: 0
  

First of a series of EVO brand cars LoRA models.

All training suggestions are welcome.

v0.741 first SDXL version, includes EVO 3 car, Cattedrale di Palermo square, Teatro Massimo & Palermo streets