Speridlabs · Research

Iris-3B

Paper Research post Code

Pixel-space generation & general vision learner. No VAE: a 3B transformer generates every pixel directly. The same prior, fine-tuned, estimates depth and restores and upscales images.

Size (width × height)
10 150
1 10
Preset1024×1024 · seed 46001 · 100 steps · CFG 3.0
Presets — click to load prompt and settings