elder-plinius/GL4SS

the looking glass — spatiotemporal image + video engine

What it solves

It addresses the difficulty of visualizing how a specific geographic location looked at different points in history. Instead of relying on fragmented memories or vague descriptions, it provides a tool to generate visual representations of any place on Earth across a vast timeline.

How it works

GL4SS is a client-side application with no backend that uses a combination of AI models via OpenRouter to generate spatiotemporal content. Users select a location on a map, a year (ranging from 25 million years ago to 3050 AD), and a time of day. The app then uses a multi-stage pipeline:

  1. Scene Planning: Uses a model (defaulting to Gemini 3 Flash) to plan the visual scene.
  2. Stills: Generates a static image using an image model (e.g., Grok Imagine or FLUX).
  3. Film + Sound: Optionally renders the frame as a short video with audio using video models (e.g., Seedance or Kling).

Frames are stored locally in the browser's IndexedDB for instant retrieval. The interface features a custom "dial" for time navigation and a "sundial" for lighting control.

Who it’s for

People interested in history, geography, and visual storytelling who want to explore the world's past and future through AI-generated imagery and video.

Highlights

  • No Backend: Operates entirely in the browser; users provide their own OpenRouter API key.
  • Spatiotemporal Range: Covers a timeline from the Great Dying (252 Myr) to the year 3050 AD.
  • Multi-modal Output: Generates both high-quality still images and short films with sound.
  • Temporal Navigation: Features a quantized time dial with 284 stations and a geometric sundial for lunar/solar progression.
  • Comparison Tool: Allows users to pin a frame and wipe between two different eras at the same location.

Related

  • Project
    zju3dv/Diffuman4D

    Diffuman4D is a spatio-temporal diffusion model for high-fidelity 4D consistent human view synthesis, enabling free-viewpoint rendering from sparse-view videos.

  • Project
    EricLBuehler/mistral.rs

    A fast, flexible LLM inference engine written in Rust that supports multimodal models, agentic tool execution, and OpenAI/Anthropic compatible APIs.

  • Project
    dekart-xyz/geosql

    A geospatial analysis skill for AI agents that enables natural language querying of PostGIS, BigQuery, and Snowflake, featuring a visual map-in-the-loop feedback system for self-correction.

  • Project
    haoyi-duan/WorldScore

    WorldScore is an open‑source benchmark for evaluating world‑generation models (3‑D, 4‑D, video). It provides a dataset, generation adapters, a suite of spatial‑temporal metrics, and a public leaderboard. Users register their model via a small YAML config and a Python class, generate videos, run the evaluation script, and submit the resulting JSON score.

  • Project
    Visionary-Laboratory/holi-spatial

    A data curation pipeline that transforms video streams into annotated 3D spatial intelligence datasets, including 3D Gaussian Splatting geometry and spatial QA pairs.