This is a workflow for NoobAI XL and Illustrious-XL.

(v1.0)It includes upscaling and FaceDetailer x3 (face, eyes, hands)

(v2.0)In v2.0, the Detailer uses an LLM for element detection.
To detect more elements, please add them to the prompt in the Florence2Run node.

The workflow is structured to perform upscaling first, followed by local correction using FaceDetailer.

(v3.0)Reverted FaceDetailer to use the Ultralytics YOLO.(face only)

(v4.0)In addition to the Detailer using Ultralytics YOLO, a Detailer utilizing MediaPipe FaceMesh has also been added.

In the default settings, the upscale nodes are disabled.
Please enable them as needed.

Includes WaveSpeed inference cache.

The Euler Ancestral sampler is not recommended because it has a low effect on cached inference.

Installation of Triton and SageAttention is required.

ex) pip install triton-windows

SageAttention is here

Please download and install the appropriate version for your environment.

The default settings are configured for models using V-Prediction + zsnr.
If you're using an Epsilon-Prediction model, please adjust the settings in the Checkpoint Loader accordingly.

Supports Dynamic Prompts and A1111 syntax (e.g., BREAK).
When using A1111 syntax, set weight_interpretation to A1111 in the CLIP Text Encode SDXL node.

If generating an illustration larger than the resolution supported by the checkpoint without upscaling, enable the Apply Kohya's HiresFix node.

Apply Kohya's HiresFix node is here.

Please install custom nodes via "Install Missing Custom Nodes" in the ComfyUI Manager.

The generated image files have prompts and workflows embedded, but the Civitai system cannot properly read the prompts.

The default values are the recommended ones provided by ChatGPT.

• Basic v1.0(2025/04/22 update)

  This is a basic generation workflow.

• Basic v2.0(2025/05/02 update)

  Changed the Detailer from YOLO to Florence-2.

• Basic v3.0(2025/05/05 update)

  The workflow has been reorganized to group frequently modified nodes together for easier access.

  Reverted FaceDetailer to use the YOLO model.

  Added support for random resolution selection.

• Basic v4.0(2025/05/14 update)

  In addition to the Detailer using Ultralytics YOLO, a Detailer utilizing MediaPipe FaceMesh has also been added.

  Set it up so that Set/Get nodes automatically switch when toggling the switch On/Off.

• Couple(2mask) v1.0(2025/04/22 update)

  This is a workflow for distinguishing multiple regions.

  Please use the Load Image node to load an image that is divided into red and blue regions.

• Couple(2mask) v2.0(2025/05/03 update)

  This is a workflow for distinguishing multiple regions.

• Couple Region(2025/05/03 update)

  This is a workflow for distinguishing multiple regions.

  While generation speed may decrease, it allows you to limit the scope of application when using multiple LoRAs.

• XY Plot(2025/04/07 update)

  This is a workflow for XY Plot.

• ModelMerge(2025/04/22 update)

  This is a workflow for model merge and preview.