ATH-MaaS/Pixelle-Video

Pixelle-Video is an AI-powered automated short video engine that generates videos by orchestrating a modular pipeline of LLM, TTS, and image/video generation services. It supports multiple AI providers and ComfyUI workflows, allowing users to create content through a Streamlit web interface or a FastAPI REST API. The system handles the entire lifecycle from topic-based narration and script generation to visual asset production and final video assembly.

The Windows packaging process now includes a \start.bat\ launcher and a \README.txt\ template. The launcher configures the environment (Python path, FFmpeg, and \PIXELLE\_VIDEO\_ROOT\) and starts the Streamlit Web UI, while the README provides users with quick-start instructions, configuration guidance, and troubleshooting steps for the portable Windows distribution.

packaging/windows/templates · high confidence

This change introduces the foundational project structure for Pixelle-Video, an AI-powered automated short video engine. It adds a Dockerfile based on Python 3.11-slim that supports configurable China mirrors for faster dependency installation, along with a docker-compose.yml and docker-start.sh script to orchestrate the API (FastAPI) and Web UI (Streamlit) services. The release includes .gitignore and .dockerignore files to manage build artifacts and sensitive data, a NOTICE file detailing open-source license compliance (MIT, Apache-2.0, BSD-3-Clause), and a config.example.yaml that defines settings for LLM providers, direct API media models (OpenAI, DashScope, Kling, etc.), ComfyUI/RunningHub workflows, and video templates. Documentation is provided via README.md, README\_EN.md, and a comprehensive mkdocs.yml for multi-language site generation.

(repo-wide) · high confidence

Added a new task management subsystem under api/tasks that provides in-memory storage and lifecycle management for asynchronous video generation jobs. This includes data models for task status and progress, a TaskManager class to handle task creation, execution, and cleanup, and a public API exposing these components for integration with the FastAPI service.

api/tasks · high confidence

This change establishes the foundational service layer for video generation by introducing new Pydantic and dataclass models for media results, progress events, and storyboards, alongside a suite of core services. The \services\ package now includes \MediaService\ and \VideoService\ for handling generation and processing, \FrameProcessor\ to orchestrate the TTS-to-video pipeline, and \APIProviderMediaService\ to support direct API calls for image and video generation (e.g., via DashScope, Kling, Seedance). Additionally, \APIAssetAnalysisService\ enables VLM-based asset analysis, while \ComfyBaseService\ and \HTMLFrameGenerator\ provide the infrastructure for ComfyUI workflow execution and HTML-based frame rendering.

_pixelle\video/services · high confidence

The application now uses a centralized configuration module (\pixelle\_video/config\) built on Pydantic models to manage settings for LLMs, ComfyUI, and direct API providers (OpenAI, DashScope, DeepSeek, Gemini, Ark, Kling). This change introduces support for configuring local HTTP proxies per provider, setting RunningHub concurrent execution limits and instance types (e.g., 48GB VRAM), and defining default prompt prefixes for image and video generation. Users can now manage these settings through a unified interface that validates required fields and persists changes to YAML.

_pixelle\video/config · high confidence

The project now includes a new FastAPI-based REST API layer (Pixelle-Video API) that exposes endpoints for LLM, TTS, image, content, and video generation. This API provides both synchronous and asynchronous video generation modes, task tracking, and file management capabilities. The implementation includes CORS middleware support, configurable API prefixes, and proper lifecycle management for the core video service instance.

api · high confidence

The web interface has been restructured into a modular Streamlit application named Pixelle-Video, replacing the previous entry point. This change introduces a multi-page layout with dedicated Home and History pages, alongside a new internationalization (i18n) system that supports dynamic language switching and automatic system language detection. The update also includes the addition of Apache 2.0 license headers across the web package files.

web · high confidence

The \api/schemas\ directory now defines the Pydantic models that govern the API contract for the Pixelle-Video platform. This change introduces strict request and response structures for video generation (including scene counts, TTS workflows, and template parameters), content creation (narration, image prompts, titles), media processing (image generation, frame rendering), and resource discovery (workflows, templates, BGM). For users, this establishes the definitive input/output formats for all API endpoints, ensuring consistent validation and documentation for services like LLM chat, TTS synthesis, and video rendering.

api/schemas · high confidence

Five new HTML templates have been added to the 1920x1080 directory, providing users with additional visual styles for video generation. The new options include an 'image\_book' design with a starry background and handwritten fonts, an 'image\_film' layout featuring a top title, central image, and bottom text, a full-screen 'image\_full' template with background image coverage, a minimalist 'image\_ultrawide\_minimal' layout with a three-column grid, and a 'image\_wide\_darktech' template with a dark tech aesthetic, grid overlays, and glowing orbs.

templates/1920x1080 · high confidence

The API now exposes a comprehensive set of endpoints organized into dedicated routers for health checks, LLM chat, text-to-speech synthesis, image generation, content creation (narrations, prompts, titles), video generation (sync and async), file serving, task management, and resource discovery. This structure introduces specific capabilities such as template parameter introspection, cross-platform URL path handling for generated videos, and a secure file access service with strict directory whitelisting.

api/routers · high confidence

Developers can now launch the Pixelle-Video project directly in GitHub Codespaces. The new configuration sets up a Python 3.11 environment, installs system dependencies (including FFmpeg and CJK fonts), and uses the uv package manager to install Python dependencies and Playwright browsers. Upon starting the container, the Streamlit web UI automatically launches on port 8501, providing immediate access to the application without manual setup.

.devcontainer · high confidence

The web interface now includes a Home page that presents available video generation pipelines as selectable tabs, allowing users to choose and run specific pipelines, along with an integrated FAQ sidebar for support. A new History page has been added to display past generation tasks in a grid view, providing detailed statistics, status filters, sorting options, and the ability to view video previews and task details for completed or failed jobs.

web/pages · high confidence

Added a new automated build system for creating Windows portable packages of Pixelle-Video. This includes a Python build script (\packaging/windows/build.py\) that downloads and bundles Python and FFmpeg, installs project dependencies, and generates launcher scripts. A corresponding README provides documentation on prerequisites, configuration options, and usage instructions for building the package.

packaging/windows · high confidence

The \pixelle\_video/prompts\ package has been introduced to centralize and standardize all LLM prompt templates used in the video creation pipeline. This change adds dedicated modules for generating topic-based narrations, refining content narrations, creating video titles, and producing prompts for both image and video generation assets. It also includes a style conversion module to translate user style descriptions into technical prompts for models like Stable Diffusion or FLUX. A key behavioral improvement across these templates is the strict enforcement of language consistency, ensuring that narration and title outputs match the input language, while image and video prompts are consistently generated in English for compatibility with AI models. The package also introduces predefined image style presets (e.g., stick figure, minimalist) and enforces structural constraints like word counts and JSON output formats to improve reliability.

_pixelle\video/prompts · high confidence

A new \api\_services\ module has been introduced to centralize and route calls to external AI providers. This layer includes a unified \ImageClient\ and \VideoClient\ that automatically dispatch requests to specific backend clients based on the model name: DashScope (Wan, Seedream, GPT-image) for images, and DashScope (Wan), Kling, and Seedance for video. The implementation supports lazy loading of provider clients, configurable proxy settings per provider, and a compatibility config layer that maps old-style configuration keys to the new provider-specific settings (e.g., \DASHSCOPE\_API\_KEY\, \KLING\_ACCESS\_KEY\).

_pixelle\_video/services/api\services · high confidence

The \pixelle\_video/utils\ package has been introduced, providing core helper modules for the application. This includes \llm\_util.py\ for discovering and testing LLM API connections, \content\_generators.py\ for creating titles and narrations via LLM, and \template\_util.py\ for parsing video dimensions and managing template resources. Additionally, \os\_util.py\ standardizes path resolution using the \PIXELLE\_VIDEO\_ROOT\ environment variable, while \workflow\_util.py\ handles ComfyUI workflow path resolution and \tts\_util.py\ contains Edge TTS logic.

_pixelle\video/utils · high confidence

This update introduces a suite of new visual templates for the video generation engine, specifically targeting 1080x1920 (vertical) and 1080x1080 (square) aspect ratios. The diff adds multiple HTML-based template files including \asset\_default\, \image\_blur\_card\, \image\_book\, \image\_cartoon\, \image\_default\, \image\_elegant\, \image\_excerpt\, \image\_fashion\_vintage\, \image\_full\, \image\_healing\, \image\_health\_preservation\, \image\_life\_insights\, and \image\_minimal\_framed\. These templates provide diverse styling options such as blurred backgrounds, book excerpts, cartoon styles, and elegant layouts, allowing users to generate videos with varied aesthetic themes.

templates/1080x1920 · high confidence

Added several new utility modules to the web UI backend: \batch\_manager.py\ introduces a lightweight batch generation system that processes multiple topics sequentially with shared configuration and progress callbacks; \history\_persistence.py\ enables automatic saving of generation history records (including video metadata and task details) for web workflows via the persistence service; \streamlit\_helpers.py\ adds a safe rerun mechanism compatible with different Streamlit versions and implements a JavaScript-based alert to warn users when switching to SelfHost workflows; \async\_helpers.py\ provides a Windows-specific async runner using \ProactorEventLoop\ to fix Playwright HTML rendering issues under Streamlit on Windows, along with a helper to read the project version from \pyproject.toml\.

web/utils · high confidence

The video generation logic has been restructured into a modular pipeline system located in \pixelle\video/pipelines\. This change introduces a \BasePipeline\ abstract class and a \LinearVideoPipeline\ base class that enforces a standardized lifecycle (setup, content generation, visual planning, asset production, post-production, and finalization) via the Template Method pattern. New concrete implementations include \StandardPipeline\ for general topic-based video creation, \CustomPipeline\ as a template for user-defined workflows, \AssetBasedPipeline\ for generating videos from user-provided media assets, and \LinearVideoPipeline\ for linear workflows. The \\\init\\_.py\ module now exports these components, centralizing the video generation strategy.

_pixelle\video/pipelines · high confidence

The web/pipelines module has been refactored to support a dynamic, extensible UI architecture. A new base registry (base.py) allows individual pipeline modules (standard, asset\_based, digital\_human, i2v, action\_transfer) to self-register, replacing the previous static structure. This change introduces unified support for three workflow sources—RunningHub, local ComfyUI (selfhost), and direct API models—via new helper utilities (api\_workflows.py) that filter and render options based on the selected source. Users can now switch between these sources in pipelines like Digital Human and Image-to-Video, with the UI dynamically adjusting available workflows and model controls accordingly.

web/pipelines · high confidence

The web UI now maintains user session state, including language preferences and a cached instance of the PixelleVideoCore service. This change ensures that the core service is lazily initialized and automatically recreated when configuration changes are detected, improving stability and resource management during user interactions.

web/state · high confidence

The Pixelle-Video package (version 0.2.0) introduces a unified service layer that exposes LLM, TTS, and video generation capabilities. The LLM integration now supports predefined presets for Qwen (via Alibaba Cloud Bailian), OpenAI, Claude, DeepSeek, Ollama, and Moonshot, with Qwen specifically configured to use the Bailian API key endpoint. Additionally, the TTS system has been expanded to include a comprehensive list of Edge TTS voices for Chinese, English, Korean, French, Portuguese, German, Russian, Turkish, and Spanish, enabling local synthesis with a wider variety of language and gender options.

_pixelle\video · high confidence

The web interface components have been reorganized into a modular structure, introducing a dedicated sidebar for displaying the FAQ (loading localized content from docs/FAQ.md) and a new batch generation mode in the content input section that allows users to process multiple topics at once. The settings panel now supports LLM preset selection with automatic model loading and connection testing, while the style configuration has been updated to manage TTS inference modes (local vs. ComfyUI) and API-based video workflows.

web/components · high confidence

The project introduces its core dependency configuration files, establishing the package as 'pixelle-video' (version 0.2.0) with a Python 3.11+ requirement. The main manifest (pyproject.toml) defines the runtime stack, including FastAPI and Uvicorn for the API service, Streamlit for the UI, and specific libraries for AI and media processing such as edge-tts (locked to 7.2.7), ComfyKit, Playwright, and MoviePy. Additionally, Windows packaging requirements and documentation build dependencies (MkDocs) are added.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• First survey — no prior run to compare against. CAI 42.

• Code Health 75
• Architecture 82
• Maturity 54
• Readiness 25
• Security 63
• Accessibility 52

• 300 commits — 279 feature/other, 21 fixes

• (root) — 92 commits
• pixelle_video/services — 35 commits
• (repo) — 22 commits
• web/i18n — 18 commits
• templates/1080x1920 — 16 commits
• web/pipelines — 14 commits
• pixelle_video/pipelines — 11 commits
• docs/images — 10 commits
• docs/en — 9 commits
• workflows/runninghub — 9 commits
• packaging/windows — 7 commits
• web/components — 7 commits
• api/routers — 5 commits
• pixelle_video/config — 5 commits
• .devcontainer/postCreate.sh — 4 commits
• docs/FAQ.md — 4 commits
• docs/FAQ_CN.md — 4 commits
• workflows/selfhost — 4 commits
• api/app.py — 3 commits
• web/app.py — 3 commits

• fix: Fix: the title in standard pipeline returned by llm may be empty
• fix: (fix) Merge pull request #192 from hit-cxf/feature/api-media-support
• fix: (fix) import error & max_token
• fix: Fix: Accidental deletion of OS imports
• fix: Revert "指定默认RUNNINGHUB_BASE_URL"
• fix: Revert "添加反向代理协议头中间件，支持 Nginx/Traefik 后正确识别 HTTPS 请求"
• fix: fix: The issue of null content field returned by LLM
• fix: fix: Windows Playwright HTML rendering under Streamlit
• fix: fix: correct TTS voice parameter name in asset_based pipeline
• fix: fix: guard cancel_task against overwriting terminal task status
• fix: fix: handle videos without audio in _trim_video_to_duration
• fix: fix: https://github.com/AIDC-AI/Pixelle-Video/issues/35
• fix: fix: lazy ffmpeg check, workflow scan caching, index-based task listing
• fix: fix: lazy load API clients and handle null LLM content
• fix: fix: playwright not found in docker build
• fix: fix: remove broken CLI entry points referencing non-existent cli module
• fix: fix: remove duplicate delete_task that doesn't update index
• fix: fix: restore local workflow selection after API integration
• fix: fix: wrong paper title
• fix: fix: 添加 docs/images 到 Docker 镜像以支持模板预览画廊
• …and 280 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

ATH-MaaS/Pixelle-Video was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.