Minor change in tensor data.
If your prompt too short, just copy-paste it twice.ππ
Illustrious variant of my DRm model derivative. Where as in DRm8.0, i already tweaked attn_ layers. Higher step (10+) tend to produce "overcooked" artifacts if inferring in low resolution, try lower the CFG.
This Papermae got:
minimized middle_block layers
reduced .block up/down
attn_ layers
conv_out and conv_in adjusted
added/extended output layers
These parameters are to increase the edge crispness a.k.a line art/art line/outlines. using negative prompt tend to overdone it, this model is absolutely possible generate decent-good image even without negative prompt, thanks to tweaked resnet blocks and conv_in/out layers.
