Primary settings are settings that provide a large impact, and are the major settings that typically needed to consider when generating an image
prompts are the most direct and most meaningful of settings - it is the text that is used to guide how the drawer generates images.
Multiple simultaneous prompts can be input, with the pipe/vertical bar character "
|" as separation between prompts. For each prompt, it is possible to specify weight and when to stop optimizing for a prompt, with syntax like
weightis a constant multiple that enhances the importance and impact of a prompt to the generation, and between multiple prompts. The
stopis for telling the drawer/optimizer when to stop optimizing for this particular prompt, often used in conjunction with a negative weight: this is because the possibility for making an image look like "not something" is quite easy, it just has to look like absolutely nothing.
stopto some extent prevents that.
prompts="suicide at the brooklyn bridge | film noir"
prompts="Squid game| art by Hwang Dong-hyuk"
--prompts "seaside sunset| blue seas:3"
aspectcan be one of three:
widescreen. By specifying aspect, it automatically defines
sizefor the user, where the one-to-one relationship is defined as follows:
'square' -> size=[144, 144] 'portrait'-> size=[128, 160] 'widescreen'-> size=[192, 108]
Recommended to use along with
scaleto increase size, note that quality automatically defines
The type of drawer defines what technology generates the images:
vqganuses VQGAN from taming-transformers,
line_sketchall use diffvg, and
pixeluses dribnet's original pixel drawer. the new
diffusiondrawer makes higher quality images (subjective), but needs more iterations that vqgan.
difference in style:
pixelgenerates pixel art,
vqgangenerates GAN-images, which are often trippy or realistic, and
line_sketchgenerates stroke-based images as if it was a drawing and strokes were drawn down.
qualityparameter is able to give defaults to many different other arguments. In general, the better the quality, the larger the image, more CLIP models would be incorporated, more iterations it would run, and more batches would be run. This gives us an easier time to specify image quality.
Often, when testing a prompt or init_image or technique, start with "draft", and then go larger as testing progresses.
The name says it all, specify the number of iterations that a prompt would run with.
Typically for a prompt, 30 iterations would show the basic layout and focus, while from 100 on, the iterations may not change as much. Over 300 would have minimal effects.
Use scale to increase size of output. The pixel size of the output is just aspect defaults multiplied by
scale. Scale need not be integer. Scale nor size does not effect pixel number of pixel drawer, or strokes of clipdraw drawer.