The parallax-photo effect, minus the watermark and the upload

What it does Feed it a still image and it produces a parallax video: the picture appears to shift in 3D as if a camera were dollying through the scene. The pipeline is monocular depth estimation (or your own depth maps) feeding a heavily optimized GLSL shader that displaces pixels according to their depth. Output is video at whatever resolution, codec, and supersampling you like, with post effects such as depth of field, lens distortion, and vignette layered on top. There’s a Gradio-based WebUI for interactive work and Python scripting for batch production.

The interesting bit The animation itself is pure rasterization — a depth map used as a displacement field while a virtual camera moves — not frame-by-frame generation, which is why the project reports up to 8k50fps on an RTX 3060. It’s also the open, self-hosted answer to watermark-and-subscription web tools doing the same trick; the repo’s own topic list name-checks Immersity.

Key highlights

• Built-in monocular depth estimation, or bring your own depth maps
• GPU-side GLSL rendering; the project claims up to 8k50fps on an RTX 3060
• Post effects including upscalers, lens distortion, depth of field, and vignette, with loops meant to be artifact-free at the edges
• Gradio WebUI for hands-on tweaking, Python scripts for mass production, plus portable ready-to-run executables
• Self-hosted with no watermarks or usage caps; commercial use explicitly encouraged, and a community ComfyUI node pack exists

Caveats

• “The latest AI models available” is as specific as the README gets — which depth estimators you actually get isn’t named there
• Rendering runs on the GPU via GLSL, so plan on having one
• Performance figures are the project’s own claims, not independent benchmarks

Verdict Worth a look if you make stock footage, social motion content, or just want the Immersity-style effect without uploading your images to someone else’s server. Pass if you expected actual 3D reconstruction — this is a convincing illusion built from one photo and one depth map.