selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Agnes Video Generator logo

Self-host Agnes Video Generator

v7.2.2

Free AI video generator with narration and subtitles

ReleaseOpen SourceMIT
Updated5 days ago
LanguagePython
Setup difficultyEasy
Stars
596
CPU
Any (no GPU required)
RAM
No minimum stated
Storage
No minimum stated
Port
8765 (Web UI and API)
Docker
ghcr.io/lcy362/free-short-video
Install with Build from Source
Agnes Video Generator web UI showing the video creation interface

What is Agnes Video Generator?

Agnes Video Generator is a free, open-source tool for creating AI-generated videos on your own machine. All heavy AI processing runs in the cloud through the free Agnes AI API, so no local GPU is needed. It produces multi-scene videos complete with AI voice narration (via Microsoft Edge TTS), word-level subtitles, and optional virtual presenter avatars. Five creation modes handle everything from single clips to full stories, long-form scripts, poetry animations, and anchor-style presentations. The web interface supports 22 languages and runs on port 8765. You can deploy it via Docker, a simple start.sh script, an npm command (npx free-short-video), or an AI-agent skill. Manual installs need Python 3.10+ and ffmpeg. The code is MIT-licensed with no paid tiers or hidden fees.

Should you self-host Agnes Video Generator?

Agnes Video Generator scores 53/100 on selfhost.directory (how scoring works) and ranks #5 of 18 in Video Processing. It has 596 GitHub stars.

It shipped 10 stable releases in the past year. The latest, v7.2.2, came out on 23 October 2026. New versions land roughly every 1 days.

Setup is easy: most people get it running in one sitting. An official Docker image is available. It is MIT-licensed and written mostly in Python.

Its closest alternative on selfhost.directory is VideoHighlighter (51/100, 345 stars). Agnes Video Generator scores higher overall. See Agnes Video Generator vs VideoHighlighter side by side.

Features

  • AI narration (TTS)
  • Auto subtitles
  • Checkpoint resume
  • Custom reference images
  • Digital anchor
  • Image-to-video generation
  • Keyframes animation
  • Multi-API-key rotation
  • Multi-language UI
  • Multiple resolution and aspect ratio options
  • Multi-scene video pipelines
  • Poetry video mode
  • REST API
  • Text-to-video generation

Agnes Video Generator README

How to install Agnes Video Generator

Install Agnes Video Generator with Build from Source

Environment variables
AGNES_API_KEY (via .env)

git clone https://github.com/lcy362/agnes-video-generator.git cd agnes-video-generator ./start.sh

Auto-creates venv, installs deps, opens browser. Requires Python 3.10+ and ffmpeg.

How to configure Agnes Video Generator

  • AGNES_API_KEY── 1. API Key(必填)────────────────────────────────────────────── 从 https://platform.agnes-ai.com 获取免费 Key ⚠️ 主 Key 的变量名不带序号(就是 AGNES_API_KEY,没有 AGNES_API_KEY_1 这个槽位)

    ── 1. API Key(必填)────────────────────────────────────────────── 从 https://platform.agnes-ai.com 获取免费 Key ⚠️ 主 Key 的变量名不带序号(就是 AGNES_API_KEY,没有 AGNES_API_KEY_1 这个槽位)

  • AGNES_API_KEY_2── 2. 多 Key 轮询(可选,推荐)──────────────────────────────────── 每个 Key 的配额独立,总量 ≈ 20 × Key 数 / 分钟。 命名规则:主 Key 用 AGNES_API_KEY,额外 Key 从 AGNES_API_KEY_2 起序号依次递增 (_2、_3、_4 ...),中间不可断号(断号之后的 Key 不会被加载)。 遇到 429 会自动轮换到下一个 Key;限速配额与并发上限随 Key 数线性放大。

    ── 2. 多 Key 轮询(可选,推荐)──────────────────────────────────── 每个 Key 的配额独立,总量 ≈ 20 × Key 数 / 分钟。 命名规则:主 Key 用 AGNES_API_KEY,额外 Key 从 AGNES_API_KEY_2 起序号依次递增 (_2、_3、_4 ...),中间不可断号(断号之后的 Key 不会被加载)。 遇到 429 会自动轮换到下一个 Key;限速配额与并发上限随 Key 数线性放大。

  • AGNES_API_KEY_3
  • AGNES_API_KEY_4
  • AGNES_API_KEY_5
  • AGNES_CHAT_TIMEOUTChat 单次读超时(秒,默认 300;与多模态 chat_multimodal 对齐,消除双标准) stability_hardening P1:超时类错误最多重试 1 次,扩大成功窗口的同时压缩失败等待

    Chat 单次读超时(秒,默认 300;与多模态 chat_multimodal 对齐,消除双标准) stability_hardening P1:超时类错误最多重试 1 次,扩大成功窗口的同时压缩失败等待

  • AGNES_CONFIG_ID_HMAC_KEY多 Key 管理接口生成 Key id 所用的哈希盐,一般无需修改

    多 Key 管理接口生成 Key id 所用的哈希盐,一般无需修改

  • AGNES_CORS_ENABLED显式禁用(即使设置了 origins 也不启用中间件):

    显式禁用(即使设置了 origins 也不启用中间件):

  • AGNES_CORS_ORIGINS── 7. CORS 跨源白名单(可选,供独立本地伴侣工具调用本服务 API)────────── 场景:本地跑一个独立的简化前端(如 agnes-simple-ui,监听 :8787), 从浏览器直接调用本服务的 /api/* 接口。设置后服务仅允许列出的源跨源调用。 默认空 = 不启用 CORS(攻击面不变);同源页面不受影响。

    ── 7. CORS 跨源白名单(可选,供独立本地伴侣工具调用本服务 API)────────── 场景:本地跑一个独立的简化前端(如 agnes-simple-ui,监听 :8787), 从浏览器直接调用本服务的 /api/* 接口。设置后服务仅允许列出的源跨源调用。 默认空 = 不启用 CORS(攻击面不变);同源页面不受影响。

  • AGNES_IMAGE_I2I_MODEL── 5. 图生图模型(可选)─────────────────────────────────────────── i2i 默认与 t2i 同模型(当前默认 agnes-image-2.5-flash);如需为 i2i 固定旧版:

    ── 5. 图生图模型(可选)─────────────────────────────────────────── i2i 默认与 t2i 同模型(当前默认 agnes-image-2.5-flash);如需为 i2i 固定旧版:

  • AGNES_RATE_BURST共享桶容量(默认 = 4 × Key 数;仅需调节突发并发时使用)

    共享桶容量(默认 = 4 × Key 数;仅需调节突发并发时使用)

  • AGNES_RATE_LIMIT── 4. 限速(可选)───────────────────────────────────────────────── 共享桶:Chat / 图片生成 / 图片上传 / 结果轮询 共用 默认 = 20 × Key 数 × 0.8 安全系数;仍频繁 429 时可适当调低

    ── 4. 限速(可选)───────────────────────────────────────────────── 共享桶:Chat / 图片生成 / 图片上传 / 结果轮询 共用 默认 = 20 × Key 数 × 0.8 安全系数;仍频繁 429 时可适当调低

  • AGNES_SUBTITLE_ASS字幕合成:=0 关闭 ffmpeg ASS 单链(回退 moviepy 词级动效路径),默认开启

    字幕合成:=0 关闭 ffmpeg ASS 单链(回退 moviepy 词级动效路径),默认开启

  • AGNES_SWEEP_AGE_DAYS── 6. 运维(可选)───────────────────────────────────────────────── 启动时清理 N 天前的僵尸任务目录(不设置则不执行清理)

    ── 6. 运维(可选)───────────────────────────────────────────────── 启动时清理 N 天前的僵尸任务目录(不设置则不执行清理)

  • AGNES_VIDEO_POLL_TIMEOUT视频任务轮询总超时(秒,默认 1800)

    视频任务轮询总超时(秒,默认 1800)

  • AGNES_VIDEO_RATE_BURST
  • AGNES_VIDEO_RATE_LIMIT视频提交独立桶(默认 = 1 × Key 数) 若服务端对视频提交是「全局限 1/min」而非 per-Key,显式设为 1

    视频提交独立桶(默认 = 1 × Key 数) 若服务端对视频提交是「全局限 1/min」而非 per-Key,显式设为 1

  • HOST0.0.0.0── 3. 本地服务监听地址(非 Agnes API 地址)─────────────────────────

    ── 3. 本地服务监听地址(非 Agnes API 地址)─────────────────────────

  • PORT8765
  • PROMPT_LANGUAGE提示词语言(zh/en,影响 LLM meta-prompt 语言)

    提示词语言(zh/en,影响 LLM meta-prompt 语言)

Agnes Video Generator releases & version history

FAQ

What should I do when generation fails with 503 queue full?

The free video model has a shared inference queue that can stay saturated for 10+ minutes. The app handles this with a dedicated retry track (up to 15 retries, spaced 30–60s apart). Retry during off-peak hours or add more API keys.

Why do I get 401 / invalid token errors even though my API key looks correct?

A 401 usually means the API key does not match the domain it is being sent to (e.g., a global key used against the China endpoint). Each key can be bound to its own access domain in the Web config page, or use the Auto-detect domains button.

Can an AI agent operate this project for me?

Yes — it was designed for that. Install the standalone agnes-video skill once, then ask for outcomes — it installs the service, generates videos, diagnoses failures and can open a PR for a real defect.

Can I host this on my own server?

Absolutely. The project is designed for self-hosting. Just clone the repo, run ./start.sh, and the server starts on http://localhost:8765. No external dependencies, no cloud lock-in.

Can I run this with Docker?

Yes. Pre-built images are published to both GHCR and Docker Hub. Just pull the latest tag and run — no Python or ffmpeg installation needed.

What languages does the UI support?

The Web UI supports 22 languages including Chinese, English, German, French, Spanish, Japanese, Korean, Arabic, and more. Subtitles are generated in the source text language with font fallback for CJK, RTL, and Indic scripts.

Can I use my own images as references?

Yes. You can upload reference images for character or scene consistency across scenes, use custom end frames for precise visual transitions, or choose img2img to auto-generate end frames from your reference.

What video generation modes are supported?

Six task types: Simple Video, Creative Video, Manuscript Video, Digital Anchor, Poetry Video, and Simple Image. Additional options include text-to-video, image-to-video, keyframes animation, and image-to-image end frame generation.

How is this different from Runway, Pika, or Sora?

Unlike commercial tools that charge $10–$95/month, Agnes Video Generator is completely free and open-source (MIT). It offers built-in multi-scene pipelines, AI narration, auto subtitles, and digital anchor — features that require third-party tools or manual editing elsewhere.

Do I need a GPU to run this AI video generator?

No. All AI compute runs in the cloud via Agnes AI's free API. You just need a regular laptop or desktop that can run Python 3.10+ and ffmpeg.

Is Agnes Video Generator really free? Are there any hidden costs?

Yes, it is completely free. All AI model calls (Agnes Chat, Agnes Image, Agnes Video) are free with no trial period, no watermarks, and no usage limits. The only TTS integration (Microsoft Edge TTS) is also free and requires no extra API key.

More in Video Processing

Rate this project

Click to rate — no account needed.

0 comments

Posting as guest
Loading comments…