PIXEL·TRAINER

Conditional MLP: [class one-hot + noise] → pixels.
Train in Generate → MLP mode first.

Train MLP generator first.

Denoising Diffusion Probabilistic Model — same math as Stable Diffusion.
Learns to reverse Gaussian noise step-by-step. Generates from pure noise via iterative denoising.

Train DDPM first.

GAN — Generator vs Discriminator adversarial training.
Generator learns to produce images that fool the discriminator.

Train GAN first.