infernalzeus/chronicle-forge

Chronicle Forge is a Python pipeline that takes a topic as input and produces a fully narrated MP4 video ready for YouTube, with no cloud services required. You give it a subject, such as the history of a city or the life of a historical figure, and it automatically writes a script, generates images, records a voiceover, and stitches everything into a 1080p video complete with YouTube title, description, and tags. The pipeline runs six steps in order. It first looks up Wikipedia background on the topic, then writes a narration script in three rounds: one pass for the story itself, a second to break it into scenes with image descriptions, and a third to expand each scene into more detailed visual moments. After that, it generates YouTube metadata, records the narration using a text-to-speech system, creates one image per scene using a local AI image model, and finally composes everything into a Ken Burns-style slideshow at 1920 by 1080 pixels. Every step is saved to disk as it completes, so if you stop the process you can pick up exactly where you left off using a resume flag. Running Chronicle Forge requires a reasonably powerful local machine. The image generation step uses a type of AI model called SDXL, which needs at least 8 gigabytes of graphics card memory. On NVIDIA graphics cards it uses CUDA, on AMD cards it uses a Windows-only path called DirectML. A CPU-only mode is available but skips image generation. You also need at least 16 gigabytes of system RAM, 20 gigabytes of free disk space, and Python version 3.11 specifically. Three external programs must be installed before the pipeline can run: Ollama (which runs the language model that writes the script and metadata), an SDXL-format image model (available via InvokeAI or manual conversion), and FFmpeg for video composition. Any Ollama-compatible language model can be substituted for the default. An optional web interface built with Flask lets you enter topics and watch progress in a browser. Setup involves cloning the repository, installing Python packages, creating a configuration file with paths to your model and voice settings, and starting the Ollama server before running the pipeline.