🎨 NanoDiffusion-46M

A ~46M parameter latent diffusion model trained from scratch on BitTranslate/Bittensor_subnet_19_06_04_24. Uses a frozen SD VAE + CLIP encoder — only the UNet was trained.

⚠️ Early-stage model (500 steps of training). Expect abstract results!
10 100
1 15
1 8
Example prompts
Prompt Denoising steps CFG scale Number of images Seed (-1 = random)