Pixel Dream Forge

Teach a tiny neural net to hallucinate. Real DCGAN + CPPN, 100% in your browser.

tf.js · …

01 / Dataset

0 imgs
Drop images here
10–500 recommended · click, drop or paste

No images yet. The generator is staring into the void.

Resolution
Channels

02 / Model

Batch size
Generator
Dense(64→4·4·128) → BN → ReLU → Reshape
ConvT(64, 4, s2) → BN → ReLU  [8²]
ConvT(32, 4, s2) → BN → ReLU  [16²]
ConvT(3, 4, s2) → tanh  [32²]
Discriminator
Conv(32, s2) → LReLU(.2) → Drop
Conv(64, s2) → LReLU(.2) → Drop
Flatten → Dense(1) logit

03 / Training Console

tensors 0
epoch
0
step
0
steps/s
0
D loss
0.000
G loss
0.000

> idle. awaiting your questionable decisions.

04 / Generation Gallery

Fixed noise · click = A, shift-click = B
Fresh samples
Timelapse (one per epoch)

Snapshots appear as epochs complete.

05 / Latent Walk

Morph between sample A#0 and B#15. Pick them in the gallery.

Hall of Hallucinations

login

Empty hall. Be the first to publish a beautiful mess.