ATH-MaaS/Pixelle-Video

Modular AI engine that auto-generates video scripts, images, voice, and final compositions from topic input.

What it does

• Automatic script generation — input topic; LLM writes narration without manual effort
• Multi-model image/video synthesis — pluggable backends: ComfyUI workflows, DashScope, OpenAI, Kling APIs
• Voice + audio composition — Edge-TTS, Index-TTS, background music; supports voice cloning
• Template-driven layout — vertical/horizontal formats, multiple genre templates (education, narrative, comedy, etc.)
• Extended capabilities — digital avatars, image-to-video, motion transfer from reference video
• Configurable model stacking — swap LLMs (GPT, Qwen, DeepSeek, Ollama), image models, TTS, video backends on-the-fly

How it works

Python Streamlit web app implementing a modular pipeline: topic → LLM script generation → image planning → per-frame synthesis → FFmpeg composition. Supports three generation paths: local ComfyUI workflows, direct API calls (DashScope/OpenAI/Kling for images; Edge-TTS/Index-TTS for audio), or RunningHub remote nodes. Atomic capabilities compose independently; users configure providers + API keys via WebUI. FFmpeg handles final video encoding; templates control layout/styling.

Standing & signals

27k GitHub stars, last pushed June 2026, regular releases through early 2026. Ships Windows one-click installer, bilingual docs (Chinese/English), and production features (avatars, motion transfer added 2026-01-26). Active maintenance and ecosystem suggest production readiness and ongoing development.

Q2 — Apply where

Tool-level integration for Auto Social Posting: evaluate Pixelle-Video as an upgrade or replacement for current make-reel.py video generation layer, particularly for automatic script generation and broader model flexibility.

Q3 — Fit for current projects + sketch

Direct fit for Auto Social Posting's video generation. Currently uses make-reel.py (Edge-TTS + Whisper, $0.02/min VI cost). Pixelle-Video offers automatic script generation, composable model backends, and motion/avatar features. Integration sketch: (1) evaluate ComfyUI vs. direct-API cost/quality vs. current pipeline, (2) wire topic/article body → Pixelle-Video script → video output, (3) regression test new reel quality against make-reel.py baseline. Not urgent; make-reel.py is proven and low-cost.

Q4 — Future-watch verdict

WATCH — production-ready and actively maintained, but ROI depends on moving beyond current make-reel.py use cases (e.g., avatar content, multi-language expansion). Architecture is well-documented and modular; revisit when scaling or adding new video features.