SOL-FIXED

SOL-ZIv1

A Stable Diffusion 3.5 Medium checkpoint merge by Brier Studios, combining SmoothMix Ultimate with zImage Turbo v1.0 via weighted sum. Focused on high-quality illustrative and artistic output with the power of the SD3.5 Medium DiT architecture.

Architecture: SD3.5 Medium (Stable Diffusion 3.5 Medium)

Parameters: 6.24B (BF16)

Total Tensors: 697

What's included:

- DiT backbone (30 transformer layers)

- SD3 VAE (16-channel)

- adaLN, context refiner, noise refiner, caption embedder

What's NOT included:

- Text encoders (CLIP-L, CLIP-G, T5-XXL must be loaded separately)

- Use CheckpointLoaderSimple (if merged as full checkpoint) or UNETLoader + VAELoader + CLIPLoader in ComfyUI

Important: This is an SD3.5 Medium model, NOT SDXL. You MUST use an SD3.5-compatible loader -- do NOT use SDXL or SD1.5 loaders with this checkpoint.

Merge Components:

- Base: Stable Diffusion 3.5 Medium

- Mix A: SmoothMix Ultimate

- Mix B: zImage Turbo v1.0

- Method: Weighted Sum

Recommended Settings (ComfyUI):

- Sampler: Euler / Euler a

- Steps: 28-40

- CFG: 3.5-5.0 (SD3.5 works at lower CFG than SDXL)

- Resolution: 1024x1024 or SD3.5 native aspect ratios

SOL-ILv1 🗡️

Checkpoint Merge • SDXL 1.0 Architecture

SOL-ILv1 is a full SDXL 1.0 checkpoint merge combining the illustrative power of Illustrious v2.0 with the versatile foundation of SmoothMix Ultimate. Designed for artists who need sharp, detailed illustrations with stable composition and strong color rendering.

✦ Details

| Property | Value |

|---------------|------------------------------------------------------------------------|

| Architecture | SDXL 1.0 (UNet 9+1+9) |

| Parameters | 3.47B total |

| Precision | FP16 (UNet + Text Encoders) / FP32 (VAE) |

| File Size | ~7.1 GB |

| Text Encoders | CLIP-ViT/L-14 (embedder.0) + OpenCLIP-ViT/bigG (embedder.1) — included |

| VAE | SDXL 4-channel (included, FP32) |

| Merge Method | Weighted Sum |

✦ Merge Recipe

- Base: stabilityai/stable-diffusion-xl-base-1.0

- Blend A: SmoothMix Ultimate

- Blend B: Illustrious v2.0

- Method: Weighted sum merge

✦ Usage

Load directly as a full checkpoint in:

- ComfyUI: CheckpointLoaderSimple → select SOL-ILv1.safetensors

- AUTOMATIC1111: Place in models/Stable-diffusion/, select from dropdown

- Forge: Same as A1111

No need for separate CLIP or VAE — everything is embedded in the checkpoint.

Recommended Settings

- Resolution: 1024×1024, 832×1216, 1216×832

- Sampler: DPM++ 2M Karras / Euler a

- Steps: 25–35

- CFG: 5–8

- VAE: Built-in (no external VAE needed)

⚠️ Known Issue

Layer 11 of CLIP-ViT/L-14 (conditioner.embedders.0.transformer.text_model.encoder.layers.11) contains NaN values — inherited from the original merge. This affects 0.2% of total parameters and does not impact generation quality in practice, as SDXL uses 12 CLIP layers and failure in one sublayer is compensated by the rest of the encoder.

✦ License

This model is provided for personal, non-commercial use. Redistribution, resale, or commercial deployment requires written permission from Brier Studios. Merged models must respect the licenses of all original base models.

Brought to you by Brier Studios — Dark Fantasy × AI

模型类型	检查点
基础模型	ZImageTurbo
发布时间	2026-05-19

SOL-FIXED

详情

下载文件 (1)

关于此版本

模型描述

此模型生成的图像