---
license: other
library_name: diffusers
tags:
- text-to-image
- stable-diffusion-xl
- ties-merging
- fp32-merge
base_model:
- Bl4ckSpaces/SpacesFusion_V1
- IllumiYume-XL-Vpred
- NoobAI-XL-Vpred
---
# SpacesFusion_V3 (Final)

**The perfected iteration.** Built upon the solid anatomical foundation of SpacesFusion_V1, using High-Precision FP32 TIES-Merging to surgically inject missing capabilities.

## The Architecture
*   **Base:** `SpacesFusion_V1` (Ensures Anatomical Stability & Support for 1536px).
*   **Knowledge Delta:** Derived from `IllumiYume` (Weight: 1.1). Corrects the "Knowledge Gap" for 2025 datasets (e.g., Marvel Rivals, newer anime).
*   **Style Delta:** Derived from `NoobAI XL` (Weight: 0.8). Enhances line sharpness and shading depth.

## TIES-Merging Configuration
*   **Precision:** FP32 Calculation -> FP16 Storage.
*   **Density:** 0.5 (Noise Pruning).
*   **Conflict Resolution:** Enabled (Prevents Knowledge/Style clashes).

## Recommended Settings
*   **V-Prediction:** Enabled (Mandatory).
*   **CFG:** 5.0 - 7.0
*   **Sampler:** Euler a / DPM++ 2M SDE.