💀🔊 StableAudioWebUI 💀🔊

A Lightweight Gradio Web interface for running Stable Audio Open 1.0

By @drbaph

Example

Ethereal_ambient_pads_with_pulsing_futuristic_bassline_and_hypnotic_shamanic_percussion.mp4

⚠ Disclaimer ⚠

I am not responsible for any content generated using this repository. By using this repository, you acknowledge that you are bound by the Stability AI license agreement and will only use this model for research or personal purposes. No commercial usage is allowed!

Recommended Settings

Prompt: Any
Sampler: dpmpp-3m-sde
CFG: 7
Sigma_Min: 0.3
Sigma_Max: 500
Duration: Max 47s
Seed: Any

> Saves Files in the following directory Output/YYYY-MM-DD/

> using the following schema 'your_prompt.mp3'

🚀Updates (0.4)

[10/06/2024]

✅ Added (Random_Seed) Checkbox

✅ Implemented Enhanced Filename Handling and Security Measures

Filename Length Control: Truncated long prompts to a maximum of 50 characters for filenames, preventing excessively long filenames.
Enhanced Sanitization: Applied strict rules to replace non-alphanumeric characters with underscores (_), ensuring valid and safe filenames.
Unique Filename Generation: Introduced a system to append numeric suffixes to filenames to avoid overwriting existing files, ensuring each file is uniquely named.
Safe Directory Handling: Utilized secure methods for path creation and directory handling to avoid risks from user input influencing file paths.

Click to expand for earlier updates

[08/06/2024]

✅ Added One-Click-Installer.bat for Windows NVIDIA / CPU Builds

✅ Optimised Code for efficiency

✅ Simplified UI

[06/06/2024]

✅ Updated UI elements to include Advanced Parametres dropdown

( CFG Scale, Sigma_min, Sigma_max )

✅ Added Use Half precision checkbox for Low VRAM inference

( Float 16 )

✅ Added choice for all Sampler types

( dpmpp-3m-sde, dpmpp-2m-sde, k-heun, k-lms, k-dpmpp-2s-ancestral, k-dpm-2, k-dpm-fast )

✅ Added link to the Repo

📝 Note: For Windows builds with Nvidia 30xx + or Float32 Capable CPU you can use the One-Click-Installer.bat to simplify the process, granted you have logged in to huggingface-cli and auth'd your token prior to running the batch script: Step 3 (the huggingface-cli is used for obtaining the model file)

Step 1: Start by cloning the repo:

git clone https://github.com/Saganaki22/StableAudioWebUI.git

Step 2: Use the below deployment (tested on 24GB Nvidia VRAM but should work with 12GB too as we have added the Load Half precision, Float16 option in the WebUI):

cd StableAudioWebUI
python -m venv myenv
myenv\Scripts\activate
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

(Note if you have an older Nvidia GPU you may need to use CUDA 11.8)

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Step 3: (Optional - read more): If you haven't got a hugging face account or have not used huggingface-cli before, create an account and then authenticate your Hugging face account with a token (create token at https://huggingface.co/settings/tokens)

huggingface-cli login

(paste your token and follow the instructions, token will not be displayed when pasted)

If you want to run it using CPU

omit 'pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121' and the process after it and just run

pip install -r requirements1.txt
pip install -r requirements.txt

Step 4: Run

python gradio_app.py

⭐ Bonus

If you are using Windows and followed my setup instructions you could create a batch script to activate the enviroment and run the script all in one, what you need to do is:

Create a new text file in the same folder as gradio_app.py & paste this in the text file

@echo off
title StableAudioWebUI
call myenv\Scripts\activate
python gradio_app.py
pause

then save the file as run.bat

Screenshots (older build)

(All with random seeds)

Prompt: a dog barking
CFG: 7
Sigma_Min: 0.3
Sigma_Max: 500

a_dog_barking.mp4

Prompt: people clapping
CFG: 7
Sigma_Min: 0.3
Sigma_Max: 500

people_clapping.mp4

Prompt: didgeridoo
CFG: 7
Sigma_Min: 0.3
Sigma_Max: 500

didgeridoo_.mp4

Model Details

Model type: Stable Audio Open 1.0 is a latent diffusion model based on a transformer architecture.
Language(s): English
License: See the LICENSE file.
Commercial License: to use this model commercially, please refer to https://stability.ai/membership

saganaki22 / stableaudiowebui Goto Github PK