forge-film

Why we included this project

Anyone who has stitched together a multi-scene AI film knows the grind: logging into separate services, pulling frames, matching colors between models, then assembling the cut. Forge takes a plain story text, compiles it into a scene dependency graph, and uses Critical Path Method scheduling to run independent scenes in parallel across different video backends. When a scene switches models, it carries over the previous scene's last frame, applies histogram color matching, and feeds that as an image-to-video seed, so the final.mp4 doesn't have jarring cuts. The pipeline is configurable through forge.yaml, includes a mock backend for testing without API keys, and offers both a Gradio web UI and a Python API for embedding into larger workflows. That makes it a practical fit for teams producing narrative shorts, ads, or any story-driven video content.