Heya @everyone we’re releasing a small update tonight:

New Features

1) V4 now supports --style and --stylize

a) The --stylize arg lets you change the strength of our ‘default styling’

– You can take it from a value of 0 to 1000 (default is 100)

– If you set --stylize 0 you may have worse images, but you may gain more fine-grained control

b) The --style arg lets you pick different ‘style biases’ for our model

– The current options are --style 4a and --style 4b

– 4a is closer to the previous v4 style (from the original v4 release)

– 4b is currently default (from the second v4 release)

Please note: These features may be very subtle unless you fix the random seed with something like --seed 1

– We will release new styles in the future that are more different/extreme from current ones

– We will also attempt to ‘keep the old styles’ from previous versions (by holding things like 4a and 4b and moving to 4c)

Model Upgrades

2) Our models (both regular and upscaler) have been updated

– Both should be slightly sharper with less undesired visual artifacts

– Both should do slightly better with small details like eyes and facial features

Bug Fixes:

3) We fixed some math with the way multiprompts and –no arguments work.

– Both should act ‘smarter’ and more literal now

– The arg --no now has a weight of -0.5 instead of -0.25 (same as –v3 and other models)

Leave a Reply

Your email address will not be published. Required fields are marked *