Animator published on GitHub: a drawn explainer video from a single phrase
XORit Editorial · Sources, AI assistance and corrections
A project called Animator has been published on GitHub: from a single text prompt it produces a whiteboard-style drawn explainer video. One command is enough for the tool to write the script, draw the frames, record the narration, and animate each icon at the moment its phrase is spoken.
How it works
Inside there is a five-step pipeline. The first four rely on Gemini models: storyboarding splits the idea into scenes consisting of one phrase and one icon, then the scene is drawn as a 16:9 frame, the text is voiced, and the icons are located in the image and linked to their phrases. The steps involving images, narration, and region detection run in parallel for all scenes. The fifth step is carried out by a local rendering engine: it draws the icons one after another and computes the animation on the CPU, so no graphics card is needed. A small built-in CRAFT model helps distinguish text from shapes.
Examples
The repository contains three examples, each assembled from a single prompt with no editing afterwards: Einstein's equation from 4 scenes (1:39), photosynthesis from 5 scenes (1:50), and wing lift from 5 scenes (1:36). According to the README, a video of 4–5 scenes takes about a minute.
Running it requires Python 3.10+ and FFmpeg. The Gemini key is needed only for generation from a prompt and for automatic region detection, and it can be obtained in Google AI Studio. If you use your own images, no key is required: clean marker drawings on a white background animate best, while photographs, gradients, and dark backgrounds work poorly.
The Animator project is a heavily extended fork of whiteboard-animator, and the repository is maintained by the user ashishjamarkattel. The base engine was created by Masih Sultani and contributors to the original project, while the fork added the pipeline from prompt to finished MP4. The prompts are kept as separate text files; they can be edited and rerun without changes to the code.
Translation editor: Салтанов А.
Source: github.com ↗