Description
The basic model of SDXL's LCM2 has been adjusted and optimized, so it is not suitable for LCM samplers and is very friendly to SDE samplers. (Contact me on WeChat: QQ1007872291)
If you don't read tedious documentation, then the recommended configuration here is as follows: Sampler: DPM++ SDE Karras CFG: 2 Steps: 5
Sampler Recommended for SDE The sampler is relatively friendly, the recommended sampler is as follows 1. DPM++ SDE Karras - slow speed, more perfect detail (5 steps +) 2. DPM++ 2M SDE Karras - fast (5 steps +) 3. DPM++ SDE - suitable for high steps (8 steps+), slow speed, more perfect details CFG recommended
CFG is usually set to 1, and can be adjusted to 1-3. The higher the value, the faster the folding speed, that is, the image can be generated using fewer steps.
Test case Prompt: (Well-Bred's works: 1.13), a beautiful girl, 24yo, skinny, pretty, [k-pop: detailed eyes, detailed beautiful face: 0.5], Negative Prompt: ng_deepnegative_v1_75t, negative_hand, disfigured, ugly, deformed, worst quality, low quality, extra legs, extra arms, deformed hands, deformed anatomy, umbrella, sampling: DPM++ 2M SDE Karras or DPM++ SDE Karras Step: 5+ Size: 768 x 1024 or above Sampler tes* ***ng 5 steps as an example, the effects of the different samplers are as follows
in 10 steps For example,
the effects of different samplers are as follows
in 20 steps For example,
the effects of different samplers are as follows
DPM++ 2M SDE Karras step count test DPM++ 2M SDE Karras is more suitable for fast card drawing. The results of steps 1 to 8 are as follows
Step 4 is A usable image was formed, and the details gradually improved as the number of steps increased.
More test reports:
https://www.yuque.com/mengxiangyuan-7fhri/eo7oa2/gtifyimxamueeixg?singleDoc # SDXL-LCM Sampler Test Report My Github:
https://github.com/SoftMeng/ComfyUI_Mexx_Styler
https://github.com/SoftMeng/ComfyUI_Mexx_Poster
https://github.com/SoftMeng/ComfyUI_Mexx_Image_Mask
https://github.com/SoftMeng/stable-diffusion-prompt-pai
https://www.liblib.art/userpage/0f37258e861044879b8728309c0ac3b0/publish
If you don't read tedious documentation, then the recommended configuration here is as follows: Sampler: DPM++ SDE Karras CFG: 2 Steps: 5
Sampler Recommended for SDE The sampler is relatively friendly, the recommended sampler is as follows 1. DPM++ SDE Karras - slow speed, more perfect detail (5 steps +) 2. DPM++ 2M SDE Karras - fast (5 steps +) 3. DPM++ SDE - suitable for high steps (8 steps+), slow speed, more perfect details CFG recommended
CFG is usually set to 1, and can be adjusted to 1-3. The higher the value, the faster the folding speed, that is, the image can be generated using fewer steps.
Test case Prompt: (Well-Bred's works: 1.13), a beautiful girl, 24yo, skinny, pretty, [k-pop: detailed eyes, detailed beautiful face: 0.5], Negative Prompt: ng_deepnegative_v1_75t, negative_hand, disfigured, ugly, deformed, worst quality, low quality, extra legs, extra arms, deformed hands, deformed anatomy, umbrella, sampling: DPM++ 2M SDE Karras or DPM++ SDE Karras Step: 5+ Size: 768 x 1024 or above Sampler tes* ***ng 5 steps as an example, the effects of the different samplers are as follows
in 10 steps For example,
the effects of different samplers are as follows
in 20 steps For example,
the effects of different samplers are as follows
DPM++ 2M SDE Karras step count test DPM++ 2M SDE Karras is more suitable for fast card drawing. The results of steps 1 to 8 are as follows
Step 4 is A usable image was formed, and the details gradually improved as the number of steps increased.
More test reports:
https://www.yuque.com/mengxiangyuan-7fhri/eo7oa2/gtifyimxamueeixg?singleDoc # SDXL-LCM Sampler Test Report My Github:
https://github.com/SoftMeng/ComfyUI_Mexx_Styler
https://github.com/SoftMeng/ComfyUI_Mexx_Poster
https://github.com/SoftMeng/ComfyUI_Mexx_Image_Mask
https://github.com/SoftMeng/stable-diffusion-prompt-pai
https://www.liblib.art/userpage/0f37258e861044879b8728309c0ac3b0/publish