Pixelle-Video: One-Click AI Video Generator With ComfyUI Integration

By Prahlad Menon 2 min read

Need to pump out short-form videos without touching a timeline editor? Pixelle-Video is an open-source engine that automates the entire workflow: script generation โ†’ image/video creation โ†’ voiceover โ†’ final render.

The killer feature: a Windows one-click installer that bundles everything. No Python setup, no ffmpeg hunting, no dependency hell.

Repo: github.com/AIDC-AI/Pixelle-Video

What It Does

Give it a topic like โ€œWhy you should build a reading habitโ€ and it:

  1. Writes the script using your choice of LLM (Qwen, GPT-4, DeepSeek, Ollama)
  2. Generates images for each scene via ComfyUI workflows or API calls
  3. Creates voiceover with Edge-TTS, Index-TTS, or custom voice cloning
  4. Assembles the video with your chosen template and background music
  5. Outputs a ready-to-post video

Real-time progress shows each step: generating scenes โ†’ creating images โ†’ synthesizing speech โ†’ rendering video.

Installation

  1. Download the latest release
  2. Extract and run start.bat
  3. Browser opens to http://localhost:8501
  4. Configure your API keys in Settings
  5. Start generating

Thatโ€™s it. All dependencies are bundled.

Manual Install (Mac/Linux)

# Install uv (Python package manager)
# See: https://docs.astral.sh/uv/getting-started/installation/

# Install ffmpeg
brew install ffmpeg  # macOS
sudo apt install ffmpeg  # Ubuntu/Debian

# Clone and run
git clone https://github.com/AIDC-AI/Pixelle-Video.git
cd Pixelle-Video
uv run streamlit run web/app.py

Configuration

LLM Setup

Pick from presets or configure manually:

ProviderNotes
Qwen (้€šไน‰ๅƒ้—ฎ)Recommended, very cost-effective
GPT-4oHigher quality, higher cost
DeepSeekGood balance
OllamaFree, runs locally

Image Generation

Three options:

  1. Local ComfyUI โ€” Point to your ComfyUI server (default: http://127.0.0.1:8188)
  2. RunningHub โ€” Cloud-based ComfyUI hosting
  3. Direct API โ€” DashScope, OpenAI DALL-E, Volcengine Seedream

Video Generation (Optional)

For templates that support video backgrounds:

  • DashScope Wan / HappyHorse
  • Kling AI
  • Volcengine Seedance

TTS Options

  • Edge-TTS โ€” Free Microsoft voices
  • Index-TTS โ€” With voice cloning support
  • Upload reference audio for custom voice cloning

The Interface

Three-column layout:

Left Panel โ€” Content

  • AI-generated or fixed script input
  • BGM selection (built-in or custom)
  • TTS workflow and voice preview
  • Reference audio upload for cloning

Middle Panel โ€” Visuals

  • ComfyUI workflow selection
  • Image dimensions
  • Prompt prefix for style control (e.g., โ€œMinimalist black-and-white matchstick figure styleโ€)
  • Style preview

Right Panel โ€” Output

  • Video template selection (static/image/video backgrounds)
  • Aspect ratio (vertical/horizontal/square)
  • Generate button with real-time progress
  • Video preview and download

Template System

Templates are HTML-based and organized by type:

PrefixDescription
static_*.htmlText-only, no AI media
image_*.htmlAI-generated images as backgrounds
video_*.htmlAI-generated video clips as backgrounds

Create custom templates by dropping HTML files in the templates/ folder.

Cost Options

Completely Free:

  • LLM: Ollama (local)
  • Images: ComfyUI (local)
  • TTS: Edge-TTS
  • Total: $0

Budget Option (Recommended):

  • LLM: Qwen (very cheap)
  • Images: ComfyUI (local)
  • TTS: Edge-TTS
  • Total: Pennies per video

Cloud Option:

  • LLM: OpenAI
  • Images: RunningHub or API
  • TTS: Premium voices
  • Total: Higher but no local GPU needed

Workflows Included

The workflows/ folder contains ready-to-use ComfyUI workflows:

  • image_flux.json โ€” FLUX image generation
  • Various TTS workflows
  • Video generation workflows

Add your own by dropping JSON files in the folder.

Why This Over Other Tools

Pixelle-Video stands out because:

  1. True one-click setup โ€” Windows bundle includes everything
  2. ComfyUI native โ€” Not a wrapper, integrates with your existing workflows
  3. Flexible pipeline โ€” Swap any component (LLM, image gen, TTS)
  4. Template system โ€” HTML-based, easy to customize
  5. Voice cloning โ€” Upload reference audio, get that voice
  6. Local-first โ€” Can run 100% offline with Ollama + ComfyUI

Academic Foundation

The project builds on research from SIGGRAPH Asia and ACL:

  • FilmAgent (SIGGRAPH Asia 2024) โ€” Multi-agent virtual film production
  • Anim-Director (SIGGRAPH Asia 2024) โ€” Controllable animation generation
  • ComfyUI-Copilot (ACL 2025) โ€” Intelligent workflow assistant
  • AniMaker (SIGGRAPH Asia 2025) โ€” MCTS-driven clip generation

Use Cases

Content Creators:

  • Explainer videos from blog posts
  • Educational content at scale
  • Social media shorts

Marketing:

  • Product feature videos
  • Localized content (swap TTS voice/language)
  • A/B test different visual styles

Personal:

  • Turn notes into video summaries
  • Create presentations automatically
  • Language learning content

Limitations

  • Chinese-focused UI and documentation (though the tool works with English content)
  • Best results require local GPU for ComfyUI
  • Video quality depends on your workflow/model choices

Links: