This is my attempt at creating a Ballet Boys lora. It's not perfect though, based on limitations of SD1.5 and the complex nature of their dance moves and poses you will get some weird limbs, hands and feet or sometimes completely mangled bodies lol. The good news is it does not happen often if your prompt is fairly simple and the weird hands, fingers and feet are usually pretty minimal and can easily be fixed with some inpainting or after editing. Overall I am quite pleased with my first attempt, I may further improve this if it seems popular. My example prompts can be seen when clicking on the images. You will see that not all the images are perfect either.

Recommended Settings...

• Main Trigger: (B4lletB0ys)

• Checkpoint: Photonic Fusion 3 or 4 (I used V4 for the examples) I trained this lora on Photonic Fusion 4 so they pair perfectly.

• On A1111 for realism for this particular lora I recommend DDIM with Karras, (DPM++ SDE + Karras is also good),

• Steps 25-40 (30 is the sweet spot)

• CFG: 5-6

• Size: 512x768

• Hires: Denoising Strength: 0.30-0.35, Hires upscale: x2, Hires steps: 13, Hires Upscaler I highly recommend: 8x_NMKD-Superscale_150000_G also (1x-ITF-SkinDiffDetail-Lite-v1 or 8x_NMKD-Faces_160000_G or R-ESRGAN 4x are also good).

• Adetailer: YES: face_yolov8n as well as mediapipe_face_mesh_eyes_only and use the "<lora:lora_perfecteyes_v1_from_v1_160:1> photo of perfecteyes, perfect eyes, matching iris, subtle muted colour iris" in the positive prompt.

Also to the people who are well versed on ballet terminology this lora was not trained using actual ballet dance/pose terminology when captioning as I am not a dancer and I don't know the lingo. So using legit terms in your prompt may or may not produce the result you are looking for you will need to rely on the checkpoint you are using to hopefully know some terms.

Enjoy and please share your creations! If you like what you see please help support so I can keep creating. Thank you