v1.0 — AI Video Generator with Web UI + Keyframes Chaining
Release v1.0 — AI Video Generator with Web UI + Keyframes Chaining
Release date: 2026-06-13
Overview
v1.0 is the initial release of Agnes Video Generator: a completely free AI video generation tool powered by the Agnes AI API. It turns a text idea into multi-scene AI videos automatically — with a Web UI, character reference images, end-frame control, three video chaining modes, and checkpoint resume.
Usage
Prerequisites
- Python 3.10+
- ffmpeg (for AI video concatenation)
One-Click Launch
git clone https://github.com/lcy362/agnes-video-generator.git
cd agnes-video-generator
./start.shYour browser will automatically open http://localhost:8765.
Manual Launch
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python server.pyThen configure your Agnes API key in the Web UI (also settable via the /api/config endpoint).
What's New
Features & Improvements
- Web UI — one-click launch, operate entirely in the browser, no command line needed.
- AI-powered full pipeline — idea → story → character reference → script → per-scene AI video → final video.
- Custom reference image — upload a character reference image for consistent character appearance across all scenes.
- Custom end frames — specify end frame images for each scene to precisely control AI video output.
- Image-to-image end frames — auto-generate scene end frames via img2img based on the reference image.
- Three video chaining modes —
keyframes(recommended) /ti2vid(transition frames) /none(independent scenes). - Checkpoint resume — auto-resume interrupted AI video tasks from the last checkpoint; no duplicate uploads or generation.
- Real-time progress — WebSocket-powered live progress updates for every step of AI video generation.
- Multi-language support — 中文 / English / Русский / 日本語 / 한국어 / Bahasa Melayu / Bahasa Indonesia.
- In-page API Key configuration — configure your Agnes API key directly in the Web UI.
Bug Fixes
- Standardized all API timeouts (image 120s, video 300s, LLM 120/300s) with separate connect/read timeouts.
- i2i end-frame read timeout fix — async base64 encoding, timeout raised to 120s, exponential backoff.
- Video duration parsed from the user requirement text; frontend duration selector wired through server and pipeline.
- Task list crash and API key status overwrite caused by i18n variable shadowing.
- Resume-flow bugs, log noise, and missing network logs.