search

Stability-AI / stable-audio-tools

Generative models for conditional audio generation
MIT License
2.55k stars 237 forks source link

Mask_args for Outpaintinig #134

Open JLenzy opened 2 months ago

JLenzy commented 2 months ago

I'd like to create my own inference/gradio script specifically for Outpainting tasks, but finding it a bit tough to determine the 'optimal' mask settings for this. Here is what I have landed on so far:

mask_args = {
    "cropfrom": 0,  
    "pastefrom": 0,  
    "pasteto": 100,  
    "maskstart": 10,  
    "maskend": 100,   
    "softnessL": 30,   
    "softnessR": 0,  
    "marination": 0.2 
}

However I'm finding the transition from original -> model generation to be quite jarring. Any guidance on how to improve the outpainting experience would be greatly appreciated!

JLenzy commented 2 months ago

@piwell tagging you in here as I can see you've been working on similar issues

Taikakim commented 1 month ago

I saw Dadabots Tweeting about an upcoming in/outpainting model being the next SA release, so it's probable that this release is not simply built for these tasks.