Illustrious Bored XL
This model aims to fuse the unique illustrative style of Obsession (Illustrious-XL) with the high-resolution capabilities and refined prompt understanding of the Illustrious XL series.
Key Features:
Hybrid Style: Combines the aesthetic sensibilities of Obsession (trained on AI-generated and Danbooru images) with the high-fidelity, detail-rich output characteristic of Illustrious XL.
High-Resolution Focus: Leverages the foundation of Illustrious XL v1.0 and v1.1, known for pushing the native resolution boundaries of Stable Diffusion XL (up to 1536x1536).
Enhanced Prompt Handling: Inherits the Natural Language Processing (NLP) + Tag-based hybrid prompting system from Illustrious XL, allowing for flexible and nuanced control using both natural language descriptions and specific tags.
Refined Foundation: Incorporates improvements from Illustrious XL v1.1 for potentially better character coherence and stability, utilizing its CLIP and VAE components.
Illustrious Bored XL is built by sequentially merging the UNets of Obsession v3.1, Illustrious XL v1.0, and Illustrious XL v1.1, creating a unique blend designed for versatile, high-quality image generation.
Consider this model a starting point for exploring the combined strengths of these powerful base models.
Description
This version was created using the ComfyUI Model Merge Blocks node.
The UNet for this model was constructed through a two-step merging process:
First Merge (output_unet1):
obsessionIllustrious_v31 was merged with Illustrious-XL-v1.0.
The merge ratios applied to obsessionIllustrious_v31's blocks were:
Input blocks: 0.6
Middle block: 0.5
Output blocks: 0.4
(Illustrious-XL-v1.0 contributed the remaining weight for each block accordingly, i.e., IN: 0.4, MID: 0.5, OUT: 0.6).
Second Merge (output_unet2 - Final UNet):
The output_unet1 (from the first merge) was merged with Illustrious-XL-v1.1.
The merge ratios applied to output_unet1's blocks were:
Input blocks: 0.3
Middle block: 0.5
Output blocks: 0.7
(Illustrious-XL-v1.1 contributed the remaining weight for each block accordingly, i.e., IN: 0.7, MID: 0.5, OUT: 0.3).
CLIP and VAE:
The CLIP (text encoder) from Illustrious-XL-v1.1 was used.
The VAE from obsessionIllustrious_v31 was used.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.