LeafReel turns any document into a narrated explainer video in eight completely different visual styles — storybook watercolor, cut paper, pixel game, claymation, chalkboard, kinetic type, flat motion graphics and cinematic LTX clips. Script fact-checked against your text, every frame inspected by a vision model, everything on open weights through ComfyUI and the Comfy SDK.
The same 400-word text (examples/volcanoes.md), rendered in each style. All made on one RTX 3090.
Two machines, each doing what it is best at.
leafreel doctor checks both nodes, every model file and tells you which styles are ready; scripts/check_open_weights.py proves no workflow contains a partner/API node.
| Style family | End-to-end (measured) |
|---|---|
| Code-rendered (flat, chalkboard, kinetic) | 2.2–2.7 min |
| ComfyUI images (watercolor, papercut, pixel, clay) | 4.6–5.7 min |
| Cinematic (FLUX frames → 10 LTX-2.5 clips with sound) | 35 min |
An MCP server lets Hermes Agent, OpenClaw, pi, Claude Code… send a PDF from a chat and get the reel back. Verified with Hermes Agent driven by a local Qwen.
mcp_servers:
leafreel:
command: /path/to/leafreel/.venv/bin/leafreel-mcp
env: { COMFY_BASE_URL: "http://127.0.0.1:8189", LLM_BASE_URL: "http://127.0.0.1:8090/v1" }git clone https://github.com/Editorenbici/leafreel && cd leafreel
python3 -m venv .venv && . .venv/bin/activate && pip install -e '.[agents]' && scripts/setup_motion.sh
leafreel doctor # checks LLM + vision, Comfy SDK proxy, models, narration
leafreel serve # web app on http://127.0.0.1:8100
leafreel make examples/volcanoes.md --style papercut --language Spanish