Generates images
Synas Image 1: an SDXL LoRA for wallpaper-style images
A small adapter for Stable Diffusion XL 1.0, tuned to produce high-resolution, aesthetic digital wallpapers.
- Type
- Text-to-image LoRA adapter (Diffusers)
- Base model
- stabilityai/stable-diffusion-xl-base-1.0 (3.5 billion parameters)
- You give it
- A text prompt
- You get back
- An image
- Training data
- A curated 500-image subset of Navanjana/image-wallpapers-dataset
- Training hardware
- Two NVIDIA T4 GPUs
- License
- CreativeML OpenRAIL-M
What it does
Synas Image 1 is a Low-Rank Adaptation (LoRA) of Stable Diffusion XL 1.0. A LoRA doesn't replace the base model. It adds a small set of trained weights on top, which nudges the base model toward a particular look. Here, that look is high-resolution digital wallpaper: landscapes, vehicles and abstract art.
Because it is an adapter, you need the SDXL base model too, and anything you can do with SDXL in Diffusers you can do with the adapter applied.
Run it yourself
Install diffusers, transformers and accelerate, then load the base model and apply the LoRA. This is the example from the model card.
import torch
from diffusers import DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
dtype=torch.bfloat16,
device_map="cuda", # "mps" for Apple devices
)
pipe.load_lora_weights("Atugil-Intelligence/synas-image-1")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]The card also lists Google Colab and Kaggle notebooks, and the Draw Things and DiffusionBee local apps. Running it yourself means your prompts and images stay on your own machine.
License and honest limits
The model card lists the CreativeML OpenRAIL-M license, which allows use but includes use-based restrictions. Read it before building on the model, and see why we say open-weight rather than open-source.
The card doesn't publish benchmarks or a limitations section, so we don't make performance claims here. Generated images can mislead: label them, and consider content credentials where your tools support them. For images whose origin you don't know, Synas Detect 1 is the companion model.
Questions
What is Synas Image 1?
It is a LoRA, a small adapter, trained on top of Stable Diffusion XL 1.0 to generate high-resolution wallpaper-style images: landscapes, vehicles and abstract digital art. It is published as a Diffusers model on Hugging Face.
Do I need Stable Diffusion XL as well?
Yes. A LoRA is loaded on top of its base model, so you load stabilityai/stable-diffusion-xl-base-1.0 and then apply the Synas Image 1 weights to it. The code on this page shows how.
What license does it use?
The model card lists CreativeML OpenRAIL-M. That license allows use but carries use-based restrictions, so read it before you build on the model. It is one reason we call the release open-weight rather than open-source.
How was it trained?
According to the model card, it was trained on a curated 500-image subset of the Navanjana/image-wallpapers-dataset, using two NVIDIA T4 GPUs.
Are there benchmarks?
The model card doesn't publish any, and neither do we. The best test is your own prompts: the card includes four example generations to start from.
Try it
Open the model card for the weights, example generations and notebooks. No sign-up needed.