Pixelle-Video AI Video Creation Platform

Pixelle-Video is an AI-powered engine designed to generate short videos (portrait or landscape) in approximately 3 minutes from a single topic input.

Core Capabilities

• Fully Automated Pipeline: Handles scriptwriting, image generation, voice synthesis, and BGM addition.
• AI Model Support: Compatible with various LLMs including GPT, Qianwen, DeepSeek, and Ollama.
• Voice Synthesis: Supports Edge-TTS, Index-TTS, and other mainstream solutions.
• Flexible Composition: Built on a ComfyUI-based architecture, allowing users to use preset workflows or customize capabilities.
• Visual Styles: Offers multiple templates and dimensions (portrait, landscape, etc.).

Pixelle-Video is an AI-powered fully automated short video engine. It allows users to generate complete videos by simply providing a single topic. The engine automates the entire creative pipeline, including:

• Scriptwriting: Writing video copy/scripts based on the topic.
• Visual Generation: Generating AI images or videos to match the script.
• Voiceover Synthesis: Creating AI-generated narration (TTS).
• Background Music: Adding BGM to enhance atmosphere.
• Video Assembly: One-click synthesis of all elements into a final video.

It is designed for users with zero video editing experience and supports various video formats (vertical/portrait and horizontal/landscape) and visual styles.

The Pixelle-Video Web interface is organized into a three-column layout to streamline the video creation workflow:

• Left Panel: Used for content input (scripts/topics) and background music settings.
• Middle Panel: Used for configuring voice (TTS) and visual (image/video generation) settings.
• Right Panel: Used for triggering video generation and previewing the final output.
• Sidebar: Contains system configuration access and a built-in FAQ.

Pixelle-Video utilizes the following technologies:

• Backend: Python 3.10+ and AsyncIO for asynchronous operations.
• Web Interface: Streamlit.
• AI Engines: OpenAI API (for LLM tasks) and ComfyUI (for image and audio generation).
• Configuration: YAML files.
• Package Management: uv.

Templates in Pixelle-Video are HTML-based layouts used to define the visual style of video frames. They are categorized by the type of media they use as a background:

• static_*.html: Static templates. These use pure text rendering and do not require AI-generated media. They are ideal for low-cost, quick generation.
• image_*.html: Image templates. These use AI-generated images as backgrounds by invoking ComfyUI image generation workflows.
• video_*.html: Video templates. These use AI-generated videos as backgrounds by invoking ComfyUI video generation workflows for dynamic content.

The system is composed of several specialized components that work together to automate video production:

• PixelleVideoCore: The central service class that coordinates all other sub-services.
• LLM Service: Manages interactions with Large Language Models to generate video scripts.
• Image Service: Interfaces with ComfyUI to generate visual assets.
• TTS Service: Interfaces with ComfyUI to generate speech/audio assets.
• Video Generator: The final stage component responsible for composing all assets into the final video file.

Pixelle-Video can be run using different combinations of services depending on your hardware and budget:

Recommendation: If you have a local GPU, use the local/Ollama solution for zero cost. If not, Qianwen provides the best value for high-quality generation.

Pixelle-Video uses a modular design to process a video from text input to final output. The workflow follows these four main stages:

1. 文案生成 (Script Generation): Creating the narrative text.
2. 配图规划 (Visual Planning): Planning which images or videos match which parts of the script.
3. 逐帧处理 (Frame-by-frame Processing): Generating the actual visual and audio assets.
4. 视频合成 (Video Synthesis): Combining all assets into the final video file.

Users can customize each stage by selecting different AI models (e.g., GPT, DeepSeek, Ollama), audio engines (e.g., Edge-TTS), and visual workflows (e.g., ComfyUI, RunningHub, or direct API connections to services like DashScope, OpenAI, or Kling).

Pixelle-Video uses a modular design to transform text into video through four main stages:

1. Script Generation: Creating the narration text based on the input topic.
2. Image Planning: Planning the visual elements for each part of the script.
3. Frame-by-Frame Processing: Generating the actual images or video frames.
4. Video Composition: Assembling the audio, visuals, and music into a final video.

Users can customize each step by selecting different AI models, audio engines, and visual styles.

Understanding the directory layout of the portable package:

• python/: Contains the embedded Python 3.11 runtime.
• tools/: Contains FFmpeg and other necessary utilities.
• Pixelle-Video/: The main application directory, including config.yaml.
• data/: Stores user data, including BGM (background music), templates, and workflows.
• output/: The destination for all generated videos.