A 6GB VRAM-optimized, local AuraFlow Gradio web app with automatic installer — for the rest of us.
Welcome to AuraGen-AuraFlow-WebUI, the low-VRAM, no-nonsense interface for generating images with AuraFlow. It's for people who think ComfyUI is great but also mildly terrifying, and who’d rather click buttons than wrangle spaghetti.
This app runs locally, plays nice with ~6GB of VRAM, and doesn’t ask you to sign away your data to the cloud gods. You're welcome.
AuraGen-AuraFlow-WebUI is a lightweight, highly optimised Gradio app for running fal/AuraFlow-v0.3 locally. It's designed for:
- Modest hardware (think: 6GB VRAM cards that aren't dreaming of becoming A100s).
- Non-technical users (no Python kung-fu or YAML deciphering required).
- Quick setup (with
install.batandlaunch.bat, no ritual sacrifices needed). - Offline privacy (your cursed prompts stay on your machine).
- ✨ Generate images from text prompts using AuraFlow v0.3.
- 📏 Adjustable image dimensions (up to 1536x1536, if your VRAM permits).
- 🎛️ Control inference steps, guidance scale (CFG), and set seeds.
- 💾 Saves outputs to an
outputfolder. - 🎨 Clean, modern UI with sliders and buttons.
- 🧠 CPU offloading enabled via
diffusersto make sure your GPU doesn't spontaneously combust.
- NVIDIA GPU with ~6GB VRAM or more recommended.
- Will fall back to CPU mode if CUDA is unavailable (but expect glacial speeds). Do not try
- AMD GPUs are not officially supported (yet), but ROCm support is evolving.
- Windows (currently tested), though Linux/Mac users can adapt the scripts.
- Python 3.9+
- Git (optional, but convenient)
Skip the manual venv setup — I've scripted the whole ordeal:
-
Clone or download the repo:
-Extract to a desired location
or
git clone https://github.com/Raxephion/AuraGen-AuraFlow-WebUI.git cd AuraGen-AuraFlow-WebUI -
Run the installer script (Windows):
-
Double-click the install.bat file inside the extracted folder
or
install.bat
This:
- Creates a virtual environment
- Installs all dependencies
- Preps the app for launch
-
-
Launch the app:
-
Double-click on launch.bat
or
launch.bat
-
-
First-time run? Grab a drink — the model (10–15GB) will download via
diffusers. After that, it's cached.
Once launched, open the Gradio URL in your browser (e.g., http://127.0.0.1:7860) and:
- Type a prompt
- Adjust image size, steps, CFG, etc.
- Click Generate
- Admire or delete your creation
Images will be saved to the output folder, in case you accidentally make something good.
Powered by fal/AuraFlow-v0.3, a state-of-the-art, open-source text-to-image model.
"Fine-tuned on more aesthetic data. Now supports wider aspect ratios. SOTA results on GenEval. Fancy stuff." – fal.ai
Huge respect to @cloneofsimo, @isidentical, and the folks behind the scenes.
-
Out of Memory (OOM)?
- Lower the resolution
- Reduce steps
- Close Chrome tabs full of cat videos
-
Stuck downloading?
- Check your internet connection. Maybe even say something nice to it. Sacrifice a goat? Nah, just kidding, but check that connection
-
Gradio crashing or tooltip errors?
- You’re probably on an older/newer version. I stripped tooltips for max compatibility. I may bring them back once I've recovered from this
-
Protobuf errors?
- Make sure you ran
install.batin a clean environment. Reinstall dependencies if needed.
- Make sure you ran
Want to use a different AuraFlow-based model (e.g. a finetuned fork)?
- Open
app.py - Change the
MODEL_NAMEat the top to your preferred Hugging Face repo - Make sure it's compatible with
AuraFlowPipeline
- This tool is provided as-is. No warranties, no unicorns.
- AuraFlow is powerful. Use it ethically and responsibly.
- Results may vary depending on your prompt, GPU, or cosmic alignment.
This project is licensed under the MIT License.
It uses the AuraFlow model, which is licensed under the Apache 2.0 License. All rights and credit for AuraFlow go to its authors at fal.ai.
Happy generating, and may your VRAM survive the journey.
— The CipherCore-AuraFlow-webUI Dev Team (There is an I in team after all! - Raxephion)