lee101/appnz-depth-vfx

appnz-depth-vfx is a small tool that takes a single photo and turns it into different kinds of visual effects based on depth, meaning how far away each part of the image appears to be from the camera. From one input photo it can generate a depth map showing near and far areas in shades, a normal map describing surface angles, a heatmap style visualization, or a short looping video that gives the still photo a subtle sense of camera movement, known as a parallax effect. Under the hood it wraps a smaller, openly licensed AI model called Depth Anything V2 using a packaging format called Cog, which is designed to make AI models easy to run consistently in a container. The model itself is baked directly into the container image and loaded once, so it does not need to download anything extra while running, and it is tuned to run efficiently on GPU hardware. The README frames the purpose of this project as a follow up step after image generation: once you already have a picture, whether from an AI image generator or elsewhere, this tool adds the geometry and camera motion effects on top of it, rather than generating the original picture itself. It is meant to run self-hosted on a platform called app.nz. You can run it locally through the Cog command line tool by pointing it at an image and choosing an effect and strength setting, run it as a Docker container with GPU access, or deploy it directly to the app.nz platform. The output effects include parallax as a short video clip, and depth, heatmap, and normal maps as still images. Input photos are limited to 20 megapixels, and parallax video clips are capped at six seconds long. The underlying AI model has its own license terms described in a separate file, while the adapter code written for this project is released under the MIT license.