Updates for the SD 3.5 LoRA

I added the DaV1nc1 trigger word (for SD3.5 Large) after training without it. I think it works better. I edited the captions significantly for SD35 and tried different approaches. Most of my sample images have the prompt embedded if you use ComfyUI you can see it.

You can also see the range and variety of images it can produce in the samples.

The model should also respond to the following words:

• Biomech

• Anatomical Study

• Medical Study

• Vitruvian

• Annotation

• Sepia

This model creates biomechanical people and animals in the style of Leonardo Da Vinci's pen and ink studies. Seems to work well in the range of about 0.6 strength, but experiment as it gets stronger and has different effects all the way from 0.1 to 1.5

Try a simple prompt like "biomech _______ _______ , ink by Da_Vinci, sepia"

Keywords to experiment with:

anatomical or medical study - will show a partly cut-away drawing with more human elements and maybe with break-out drawings of internal bits

biomech, steampunk - will give you more gears and mechanical parts

cyborg - more robot than living

ink or ink drawing - can be important if you are not getting the ink drawing style

sepia - to make it look older and more like a Renaissance-style drawing

vitruvian - Should give you the famous arms and legs extended drawing style with double arms

The prompts don't come through since I used ComfyUI, but you can load them in Comfy and see the prompts. They are extremely simple, like "Biomech girl riding a horse, by Da_Vinci, ink, sepia, 8k, masterpiece, extremely detailed." If you read this far, fantastic congrats - hope you enjoy it.

Keep in mind it's my first Lora ever, I might post more or hopefully better ones with encouragement.

It wasn't trained on NSFW images but it seems to be able to do some - but YMMV and it may depend on which ckpt you use.

This is my first Lora, made from images I generated myself based on a Renaissance-style pen and ink sketch with annotations inspired by Da Vinci's anatomical/medical studies but with a biomechanical twist. I used 55 images I generated myself based on that concept and this was trained on Civitai - Epoch 8 seemed the best choice even though Civitai suggested Epoch 15 might be the best. I just found Epoch 15 to be too strong.