A simple merge of several PonyXL models in an attempt to match the style of my base SDXL merge, a semi-realistic 3D Disney cartoon style with sharp details and vivid colors. It isn't quite as good as I'd like it to be, and base resolution faces are pretty terrible, but I'm getting good results with hiresfix + adetailer/face detailer, which are highly recommended. A simple ComfyUI workflow including both was used for all the examples.
I uploaded the workflow separately, you can find it on my profile and in the suggested resources.
Recommended settings:
Resolution: 812x1216, 1024x1024, 768x1024 (for slight speed boost)
Steps: 30-50
Sampler: Euler a (any should be fine)
Scheduler: Karras (AlignYourSteps scheduler used in examples)
CFG: 5-7
Hiresfix: 1.5x-2x using any upscale model (4x_foolhardy_Remacri used in examples)
10-20 steps at .3-.45 denoise
(examples use 1.5x, 12 steps, .4 denoise)
Adetailer/face detailer is HIGHLY recommended.
Please experiment further as I didn't change these settings much.
Recommended starting positive prompt:
"(score_9, score_8_up, score_7_up:1.1), score_6_up, score_5_up, score_4_up, (semi-realistic disney cartoon, volumetric lighting:1.1)"
Recommended starting negative prompt:
"score_4_up, score_5_up" (I used this for the examples, but I'm not sure it helps at all. Please experiment more, I haven't tested prompting very thoroughly.)
The models used for the merge are listed in the suggested resources section, please check them out as well.
描述:
训练词语:
名称: 3dponyvision_v10.safetensors
大小 (KB): 6775458
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
