This is a workflow for SDXL(NoobAI XL/Illustrious-XL) and Anima.

Last update: Anima XY Plot LoRA

Anima XY Plot LoRA

• XY Plot LoRA1

This is an example of comparing results without LoRA applied, with LoRA applied, or between multiple LoRAs.

• XY Plot LoRA2

This is an example of comparing different LoRA strength values.

Anima Turbo LoRA + NAG

By using Anima Turbo LoRA, you can use Negative Prompts with NAG while maintaining low steps and CFG 1.

Since applying NAG reduces generation speed, when generating with the 8–12 steps recommended for Anima Turbo LoRA, it is recommended to set it to approximately half of the total steps, such as start_percent=0 and end_percent=0.5.

Anima Couple

• v1

The mask is split to draw the characters separately. If a single LoRA is trained on multiple characters, it can distinguish between them, but it does not support combining multiple character LoRAs.

Since Anima generation takes time, this uses a speed-up node (Accelerate subgraph).
In exchange for faster generation, some degradation may occur. Adjust the parameters as needed.

• v2

Support for applying multiple character LoRAs has been added.

Anima

The processing order is: generation → Hires.fix → face-only Hires.fix → hands-only Hires.fix → upscale.

Anima XY Plot CheckPoint

This is a workflow for comparing Anima-based checkpoints.

Require

• SageAttention

here

Please download and install the appropriate version for your environment.

• triton-windows

ex)

pip install triton-windows

• sam3

v3.0 here

v3.1 here

• my custom nodes

https://github.com/hybskgks28275/ComfyUI-hybs-nodes

https://github.com/hybskgks28275/ComfyUI-Anima-NAG

• Build Tools for Visual Studio

If you enable torch.compile, please install triton-windows.

Also, install Build Tools for Visual Studio and make sure that cl.exe is added to your system PATH.