Illustrious

🚀 gguf Quantized and FP8 Scaled Versions of Illustrious (Test Pack)

This project offers gguf quantized and FP8 scaled versions of the Illustrious model, falling under the text - to - image pipeline.

📄 License

The project is licensed under the Apache-2.0 license.

📦 Installation

General Setup

• Drag the gguf file(s) to the diffusion models folder (./ComfyUI/models/diffusion_models).
• Drag the clip or encoder(s), such as illustrious_g_clip and illustrious_l_clip, to the text encoders folder (./ComfyUI/models/text_encoders).
• Drag the vae decoder(s), like vae, to the illustrious_vae folder (./ComfyUI/models/vae).

Run Without Installation

• Obtain the Comfy pack with the new gguf - node (pack).
• Run the .bat file in the main directory.

💻 Usage Examples

Workflow

• Drag any workflow JSON file to the activated browser.
• Alternatively, drag any generated output file (e.g., picture, video, etc., which contains the workflow metadata) to the activated browser.

Widget Examples

Here are some examples of using the widget:

Example 1

• Text: "masterpiece, best quality, vibrant, very aesthetic, high contrast, semrealistic, highly detailed, absurdres, masterful composition, cinematic lighting, score_9, score_8_up, score_7_up, score_6_up, score_5_up, rating_questionable, source_anime, 1girl, portrait, multicolored hair, fringe, bare shoulders, upper body, cosmic"
• Parameters:
  • Negative Prompt: "femboy, low quality, 2koma, 4koma, bad anatomy, jpeg artifacts, signature, watermark, lowres, bad hands"
• Output: samples\ComfyUI_00001_.png

Example 2

• Text: "drag it to browser same descriptor to the 1st one with gguf q4_0"
• Output: samples\ComfyUI_00002_.png

Example 3

• Text: "drag it to browser same descriptor to the 1st one with gguf q4_0"
• Output: samples\ComfyUI_00003_.png

Example 4

• Text: "drag it to browser same descriptor to the 1st one with gguf q4_0 (v9 model)"
• Output: samples\ComfyUI_00007_.png

Example 5

• Text: "drag it to browser same descriptor to the 1st one with gguf q8_0 (v11 model)"
• Output: samples\ComfyUI_00014_.png

Example 6

• Text: "drag it to browser same descriptor to the 1st one with full set gguf (new v13 model)"
• Output: samples\ComfyUI_00015_.png

📚 Documentation

Review

• For legacy models, using tags/words as input can yield more accurate results. However, it's not very convenient compared to recent models at the beginning.
• Credits should be given to the contributors from the Civitai platform.
• fast - illustrious gguf was quantized from FP8 scaled safetensors, while illustrious gguf was quantized from the original bf16. (This is just an attempt to test: Does the trimmed model with 50% fewer tensors really load faster? Please test it yourself. Note that some models might have unique structures/features that affect the loader performance, and there's no one - size - fits - all solution.)
• The FP8 scaled file works well in this model, including the vae and clips.
• It can run on old machines, such as those in the 9xx series or earlier (legacy mode [--disable - cuda - malloc --lowvram] is supported) and is compatible with the new gguf - node.
• Disclaimer: Some models (original files) are provided by others, and it might be difficult to identify the creator/contributor(s) unless specified in the source. It's better to leave it blank rather than using "anonymous/unnamed/unknown". If it's your work, please let us know, and we'll handle it properly.

🔗 References

• Wai: creator
• ComfyUI: comfyanonymous
• gguf - node: ([pypi](https://pypi.org/project/gguf - node)|repo|pack)

📋 Information Table