Flux-Schnell

In terms of inference speed, the FLUX Schnell variants deliver up to 3× faster generation than the Base (compiled) model:

• Base (compiled): 0.9s

• Lightly Juiced: 0.6s

• Juiced: 0.4s

• Extra Juiced: 0.3s
  (For reference, the public Replicate endpoint reports 1.1s without go_fast, and 0.6s with it.)

In terms of quality, the Lightly Juiced and Juiced models closely match the Base model. Extra Juiced maintains similar performance but may occasionally degrade fine details like sharpness.