wai-A3-V140
Back

wai-A3-V140
529 0 7

Other

wai-A3-V140

WAI illustrious Rectified 4Steps
4 steps to generate high-quality illustrations, no need to worry about not having money to buy a 5090 anymore!

This time, a LORA was trained using WAI illustrious as the base model, so everyone can pair it with their favorite version.

Recommended parameters
Sampling steps: 4.

CFG scale: 1~1.5.

If CFG is set to 1, the speed of each step will be doubled, and if CFG is set to 1.5, the accuracy will be higher!

Sampling method: Euler, DPM (do not use Euler a or SDE versions).

Schedule type: Simple.

Schedule type is quite important! If you are using sd webui or ComfyUI, the default schedule type won't work, you must change it to Simple. If you're using diffusers, it corresponds to initializing the Scheduler with timestep_spacing="trailing".

About versions
A1 was trained on WAI NSFW illustrious SDXL's V10 and is suitable for WAI's V13 and earlier versions.
A2 and A3 were trained on V14 and are suitable for WAI's V14 version, as WAI's V14 has switched the base model to illustrious1.0.

Since it was trained on WAI, it works best when paired with WAI series models. Other illustrious and NoobAI series models can also be used, but the results may have slight imperfections.

Training code
The training code is here, I know no one wants to look at it anyway, so I just wrote it casually. If you don’t understand it, that's normal:

https://github.com/RimoChan/NoobAIXL Rectified

You can check out the principle in this Rectified Diffusion:

https://arxiv.org/html/2410.07303v2

This model is sourced from an external transfer (transfer address: https://civitai.com/models/1355945/wai-illustrious-rectified-4steps ）,if the original author has objections to this transfer, you can click，
Appeal
We will, within 24 hours, edit, delete, or transfer the model to the original author according to the original author's request

user_2wgcvmje

user_2wgcvmje

Other

Model Information

Active
Original author：
u9
Model Type：
LoRA
Basic Model：
IL-XL
Resource Name：
models/loras/wai-A3-V140.safetensors
MD5：
d1ec2d8d2df7a03d248d8ba281a1b188

WAI illustrious Rectified 4Steps
4 steps to generate high-quality illustrations, no need to worry about not having money to buy a 5090 anymore!

This time, a LORA was trained using WAI illustrious as the base model, so everyone can pair it with their favorite version.

Recommended parameters
Sampling steps: 4.

CFG scale: 1~1.5.

If CFG is set to 1, the speed of each step will be doubled, and if CFG is set to 1.5, the accuracy will be higher!

Sampling method: Euler, DPM (do not use Euler a or SDE versions).

Schedule type: Simple.

Schedule type is quite important! If you are using sd webui or ComfyUI, the default schedule type won't work, you must change it to Simple. If you're using diffusers, it corresponds to initializing the Scheduler with timestep_spacing="trailing".

About versions
A1 was trained on WAI NSFW illustrious SDXL's V10 and is suitable for WAI's V13 and earlier versions.
A2 and A3 were trained on V14 and are suitable for WAI's V14 version, as WAI's V14 has switched the base model to illustrious1.0.

Since it was trained on WAI, it works best when paired with WAI series models. Other illustrious and NoobAI series models can also be used, but the results may have slight imperfections.

Training code
The training code is here, I know no one wants to look at it anyway, so I just wrote it casually. If you don’t understand it, that's normal:

https://github.com/RimoChan/NoobAIXL Rectified

You can check out the principle in this Rectified Diffusion:

https://arxiv.org/html/2410.07303v2

This model is sourced from an external transfer (transfer address: https://civitai.com/models/1355945/wai-illustrious-rectified-4steps ）,if the original author has objections to this transfer, you can click，
Appeal
We will, within 24 hours, edit, delete, or transfer the model to the original author according to the original author's request