PLEASE NOTE!!! This is a DMD2-mode checkpoint. For optimal results, use LCM sampler, Beta scheduler, 8 steps and 1.2 CFG
Rapid generation of simple cartoon styles. e6 artist tags with simplistic styles will look the best. Artists with more detailed styles will be dark and noisy.
Description
Generation parameters:
LCM Sampler, Beta Scheduler, CFG 1.2, Steps: 8
For best results, use vpred LoRAs with this checkpoint.
FAQ
Comments (3)
v-pred made loras, locons and gloras work to an extent but mostly without issues, trust me on this.
This is an unusual, but interesting one. Thanks =)
this is very nice but also very, VERY, saturated.
Also.... why do i get what looks like AI text advertisement when i put the cfg around 0 to 0.4 (ComfyUI only can do this i believe), and using your settings has some strange noise added to the images for some odd reason but its even greater with every other sampler (only sampler I found to not have that noise at 15 steps and not as "baked" is a non-standard one called Kohaku_LoNyu_Yog)
how did you train this if its ok to ask cuz its got very good knowledge and understanding of characters but it seams a bit... broken.
Side note: I have never trained my own model (but i do merge models on occasion), only info i have to go off of is others ik personally that do a mix of training and merging and even some of them say this has good understanding but it's very "deep-fried"
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.