---
license: creativeml-openrail-m
language:
- en
library_name: diffusers
pipeline_tag: text-to-image
base_model: stabilityai/stable-diffusion-xl-base-1.0
tags:
- stable-diffusion
- stable-diffusion-xl
- sdxl
- text-to-image
- photorealistic
- fp16
- vae-fix
- juggernaut
- rundiffusion
- kandooai
---
> A drop-in replacement for [**Juggernaut XL v7**](https://huggingface.co/RunDiffusion/Juggernaut-XL-v7) with the **SDXL fp16 VAE fix** baked in. Use this if your fp16 inference produced black images, NaN errors, or noise artifacts on the original v7. Output is otherwise identical.
## When to use this variant
- ๐ ๏ธ You're running **fp16 inference** and the standard v7 produces black or NaN-corrupted outputs.
- ๐ป You're memory-constrained and need fp16 for **lower VRAM**.
- ๐ฌ You have **fp16 LoRAs or pipelines** validated against this fix.
If you're running fp32/bf16 or have no fp16 issues, the standard [**v7**](https://huggingface.co/RunDiffusion/Juggernaut-XL-v7) works fine.
---
๐ Skip the VAE drama. Try Juggernaut on RunDiffusion.
Every model pre-loaded with the right VAE. No installs. No precision gotchas.
Pre-loaded into ComfyUI ยท Forge ยท Automatic1111 ยท Fooocus ยท InvokeAI ยท SwarmUI
---
## The Newer Juggernauts
| Model | Why upgrade |
| --- | --- |
| [**Juggernaut XL v9**](https://huggingface.co/RunDiffusion/Juggernaut-XL-v9) | Direct successor on SDXL โ RunDiffusion Photo v2. **6M+ downloads.** |
| [**Juggernaut X / XI**](https://huggingface.co/RunDiffusion/Juggernaut-XI-v11) | Ground-up retrains with GPT-4V captions |
| [**Juggernaut Z**](https://huggingface.co/RunDiffusion/Juggernaut-Z-Image) | Lumina-Image-2 architecture |
| [**Juggernaut Pro Flux**](https://www.rundiffusion.com/juggernaut-pro-flux?utm_source=huggingface&utm_medium=model_card&utm_campaign=juggernaut_xl_v7_fp16&utm_content=cross_promo_pro_flux) | FLUX.1 backbone |
## Recommended Settings
Same as standard v7:
| Parameter | Value |
| --- | --- |
| Resolution | `832 ร 1216` (portrait) ยท `1216 ร 832` (landscape) |
| Sampler | `DPM++ 2M Karras` |
| Steps | `30 โ 40` |
| CFG scale | `3 โ 7` |
| VAE | **Fp16-fixed VAE already baked in** |
๐ [**Prompting library**](https://www.rundiffusion.com/prompting?utm_source=huggingface&utm_medium=model_card&utm_campaign=juggernaut_xl_v7_fp16&utm_content=prompting_inline) โ free to read.
## Commercial Use
This model **may not be deployed behind paid API services** without explicit licensing. Contact **[juggernaut@rundiffusion.com](mailto:juggernaut@rundiffusion.com)**.
Personal/creative use under [CreativeML Open RAIL-M](https://huggingface.co/spaces/CompVis/stable-diffusion-license).
## Credits
Juggernaut XL v7 was created by **[KandooAI](https://twitter.com/Juggernaut_AI)** with **[RunDiffusion](https://www.rundiffusion.com/?utm_source=huggingface&utm_medium=model_card&utm_campaign=juggernaut_xl_v7_fp16&utm_content=credits)**. FP16 VAE fix applied for stable half-precision inference.
---
