Anil-matcha /

vox-ai-motion-graphics-generator

趋势观察中

🎬 Turn any topic into a finished Vox-style paper-collage explainer / motion graphics video — script, collage keyframes, animation, voice-over, music & captions, all automated. An agent skill for Claude Code, Codex & other coding agents.

原始仓库简介 · 中文摘要待补充 · agent-skill / ai-motion-graphics / ai-video / claude-code / collage / explainer-video

GitHub 仓库 ↗

README / 项目原文

README.md ↗抓取于 2026/10/29 14:03:10

Vox AI Motion Graphics Generator

Turn any topic into a finished Vox-style paper-collage explainer video

📺 Watch the Vox AI Motion Graphics Generator in action →

🚨 The problem with making explainer videos today:

❌ Fragmented workflow — separate tools for scripting, image gen, animation, voice-over, music, and captions
❌ No narrative structure — raw text-to-video models don't understand story arcs, hooks, or pacing
❌ Inconsistent look — the collage aesthetic drifts from shot to shot
❌ Hours of manual editing — stitching, ducking music, and burning captions by hand

💡 The solution:

🎬 Screenwriter, Collage Artist, Animator, and Editor — all in one automated pipeline.

Type one topic. The agent writes the story beats, renders each beat as a torn-paper collage poster, animates it, adds a narrator voice-over, music, and burned-in captions — then stitches everything into a finished final.mp4. You stay in control with just two approval gates: the story beat map and the visual theme.

🎨 Explore 50+ more open-source AI apps →

🎥 Demo

https://github.com/user-attachments/assets/3db826d1-3271-4e8a-b4d0-a843013e67c7

One topic in, a fully narrated & captioned Vox-style collage video out. (download the demo)

"Make me a 15-second Vox-style collage video on the history of coffee." → a styled final.mp4, fully narrated and captioned.

📽️ Generated Showcase Videos (Playable Demo Videos)

Below are end-to-end videos generated using the pipeline:

https://github.com/user-attachments/assets/d9b3e85b-2f4a-4c64-9692-747f085dfd28

▶ "History of Coffee (Ethiopia 850 AD)" · 5s (B-roll Film)

https://github.com/user-attachments/assets/6628b678-78ff-481a-bc39-d7098ec1f2fe

▶ "Cold Brew Product Ad" · 5s (C-roll Film)

Mode	Film Title	Topic / Subject	Playable Video Link	Keyframe Poster	Motion Model	Narration Voice
B-roll	☕ History of Coffee	Ethiopia 850 AD	Play Video (`out/demo/final.mp4`)	`out/demo/keyframes/kf_1a.jpg`	`veo3.1-image-to-video`	MiniMax TTS (`Q19bea09caa6IRAeW7`)
C-roll	🍾 Cold Brew Product Ad	Artisanal Cold Brew Bottle	Play Video (`out/demo-croll/final.mp4`)	`out/demo-croll/keyframes/kf_1.jpg`	`veo3.1-image-to-video`	MiniMax TTS (`Q19bea09caa6IRAeW7`)

🎬 Generated Assets Summary:

☕ B-roll (History of Coffee):

📹 Video: Play Video (out/demo/final.mp4) | 🖼️ Poster: out/demo/keyframes/kf_1a.jpg | 📹 Clip: out/demo/clips/clip_1a.mp4 | 🎙️ Voice: out/demo/audio/narr_1.mp3 | 🎵 BGM: out/demo/audio/bgm.mp3

🍾 C-roll (Cold Brew Product Anchor):

📹 Video: Play Video (out/demo-croll/final.mp4) | 📸 Anchor Photo: out/demo-croll/anchor_photo.jpg | 🖼️ Poster: out/demo-croll/keyframes/kf_1.jpg | 📹 Clip: out/demo-croll/clips/clip_1.mp4 | 🎙️ Voice: out/demo-croll/audio/narr_1.mp3 | 🎵 BGM: out/demo-croll/audio/bgm.mp3

✨ Key Features

3 Input Modalities (A-roll, B-roll, C-roll):
- B-roll (Topic → Film) — a single one-line prompt produces a complete, captioned, narrated video
- A-roll (Talking-Head Video → Collage) — segment a presenter video with ASR (openai-whisper) and re-style into paper-collage while keeping real face, gestures, and lip-sync
- C-roll (Single Photo / Product → Anchored Collage) — cut out a selfie or product shot as a photographic sticker and build animated collage posters around it
Authentic Vox collage aesthetic — torn paper, cutouts, tape, halftone dots, newspaper clippings, bold flat color, big headlines
Story-first — picks a narrative arc (timeline, problem-agitate-solution, how-it-works…) and writes a hook-led beat map before generating anything
Style bake-off — renders the same beat in 3–4 themes so you pick the look before committing
Living-poster motion — animates each collage keyframe into dynamic motion with Google Veo 3.1
Voice + music + captions — narration (TTS or voice cloning), background music ducked under the voice, and burned-in captions, all automated
Two human gates, everything else automated — approve the beat map, pick the theme; the pipeline handles the rest
Agent-native — a self-contained skill any coding agent (Claude Code, Codex, Antigravity, …) can read and run
One API key + ffmpeg — no cluster of accounts to wire up

🎨 The Look

The aesthetic is the modern editorial paper-collage popularized by Vox explainers and creators like Stav Zilber and rom1trs: hand-cut paper cutouts, torn edges, tape, halftone dots, newspaper clippings, bold flat colors per beat, and big cutout headlines — brought to life with dynamic motion, a narrator voice-over, music, and burned-in captions.

The collage look is born in the image step (each beat is a finished collage poster), and the motion is added after — so the DNA of the style is locked in before anything moves.

🔄 How It Works

A single topic, video, or photo flows through the pipeline driven by beats.json under out/<project>/:

topic / video / photo
  │
  ├─ 1. Beat Map        Pick a narrative arc → write beats.json          ◀── GATE 1: approve the beat map
  ├─ 2. Style Bake-Off  Render the same beat in 3–4 themes               ◀── GATE 2: pick the look
  ├─ 3. Keyframes       One collage poster per beat (nano-banana-2 / flux-dev)
  ├─ 4. Motion          Animate each poster into a clip (veo3.1-image-to-video)
  ├─ 5. Voice & Music   Narration (minimax-speech-2.6) + BGM (suno-create-music)
  ├─ 6. Assemble        Stitch clips, duck music, burn captions (ffmpeg)
  └─ final.mp4

3 Modalities:

B-roll (Topic → Film): Topic in → script → keyframe posters → Veo 3.1 motion → TTS narration + Suno music → final.mp4.
A-roll (Talking-Head → Collage): Presenter video in → Whisper ASR beat segmentation → video-edit restyling → audio remuxing → final.mp4.
C-roll (Photo / Product → Collage): Product photo in → photographic sticker cutout → anchored collage posters → Veo 3.1 motion → cloned voice → final.mp4.

🧩 Models

Pipeline Job	Model Endpoint
Keyframes / Collage Posters	`nano-banana-2` / `flux-dev`
Motion / Animation	`veo3.1-image-to-video` / `veo3.1-fast-image-to-video` / `runway-image-to-video`
Re-style Talking-Head (A-roll)	`gemini-omni-video-edit` / `veo3.1-image-to-video`
Anchor Photo in Collage (C-roll)	`nano-banana-2` / `flux-dev`
Narration (TTS)	`minimax-speech-2.6-turbo`
Voice Cloning (C-roll)	`minimax-voice-clone`
Background Music	`suno-create-music`
Audio Transcription (A-roll ASR)	`openai-whisper`
Background Removal	`remove-background`

🚀 Quick Start

1. Install local dependencies

ffmpeg + ffprobe (brew install ffmpeg on macOS / choco install ffmpeg on Windows)
Python 3 with Pillow (pip install pillow)

2. Configure environment keys

export MUAPI_API_KEY="your-api-key"     # image / video / voice / music models — key from muapi.ai
export OPENAI_API_KEY="your-openai-key"  # story planning & ASR

3. Run Pipeline via Agent or CLI:

Mode 1: B-roll (Topic → Film)

Ask your coding agent:

"Make me a 15-second Vox-style collage video introducing the history of coffee."

Or run CLI scripts manually:

python scripts/style_bakeoff.py out/my-topic american-retro,swiss-modern,punk-zine
python scripts/keyframes.py out/my-topic
python scripts/clips.py out/my-topic
python scripts/audio.py out/my-topic
python scripts/assemble.py out/my-topic

Mode 2: A-roll (Talking-Head Video → Collage)

python scripts/asr_beats.py out/my-aroll source_presentation.mp4
python scripts/aroll_clips.py out/my-aroll
python scripts/aroll_assemble.py out/my-aroll

Mode 3: C-roll (Single Photo / Product → Anchored Collage)

python scripts/croll_keyframes.py out/my-croll
python scripts/clips.py out/my-croll
python scripts/audio.py out/my-croll
python scripts/assemble.py out/my-croll

🛠️ Project Structure

.
├── SKILL.md            # full agent workflow + approval gates
├── AGENTS.md           # entry point for coding agents (Claude Code, Codex, Antigravity)
├── README.zh.md        # Chinese documentation
├── scripts/            # one script per pipeline stage
│   ├── muapi_client.py    # MuAPI client wrapper
│   ├── provider.py        # task queue & status polling
│   ├── style_bakeoff.py   # multi-theme style candidates
│   ├── keyframes.py       # B-roll collage poster generator
│   ├── clips.py           # B-roll video animator (Veo 3.1)
│   ├── audio.py           # TTS narration & Suno music generator
│   ├── text_overlay.py    # subtitle captions & watermark generator
│   ├── assemble.py        # ffmpeg video assembly, ducking & composition
│   ├── asr_beats.py       # A-roll Whisper ASR beat cutter
│   ├── aroll_clips.py     # A-roll talking-head video restyler
│   ├── aroll_assemble.py  # A-roll clip & audio remuxer
│   └── croll_keyframes.py # C-roll photo sticker anchor generator
├── references/         # prompt guide, beat/story library, voices
└── examples/           # sample beats.json

This is an agent skill — Claude Code auto-loads it from SKILL.md; Codex and other agents follow SKILL.md via AGENTS.md. Just ask for a "vox video" or a "collage video."

MuAPI — Unified API for the image, video, speech, voice, and music models used by this pipeline.
MuAPI video generation docs — API guidance for the image-to-video and text-to-video stages.
MuAPI music and speech docs — API guidance for narration, voice cloning, and background-music generation.
Open-AI-Micro-Drama-Generator — agentic AI micro-drama video generator
AI-B-roll — auto-generate AI b-roll for your videos
Text-To-Video-AI — generate full videos from text
AI-Youtube-Shorts-Generator — auto-clip long videos into viral vertical shorts
awesome-ai-video-models — compare AI video models by API, price & speed

⭐ Star this repo if it helped you — and explore 50+ more open-source AI apps →

README.zh.md ↗抓取于 2026/10/29 14:03:10

English · 简体中文

🎬 Vox Director (拼贴动效导演) & Veo 3.1 ComfyUI 节点

一个选题、口播视频或照片进，一条成片出——脚本、拼贴关键帧、动效、旁白、配乐、字幕全流程自动化的 Vox 风格拼贴讲解/广告视频。

一个基于 MuAPI 平台 API 和本地 ffmpeg 的 Agent 技能 与 ComfyUI 节点套件，可供任何编码 Agent（Claude Code、Codex、Antigravity 等）或在 ComfyUI 中使用。给它一句话选题、口播视频或产品照片，直接输出成片 .mp4。

out/demo/final.mp4

▶《咖啡简史（埃塞俄比亚公元 850 年）》· 5 秒 (B-roll)


_{咖啡简史 · 5 秒 (B-roll)}	_{冷萃咖啡产品广告 · 5 秒 (C-roll)}

_{▶ 点击任意封面/路径播放生成视频}

📽️ 实测生成视频（在线播放）

以下是使用 Vox AI Motion Graphics Generator 全流程自动生成的示例视频：

https://github.com/user-attachments/assets/d9b3e85b-2f4a-4c64-9692-747f085dfd28

▶《咖啡简史（埃塞俄比亚公元 850 年）》· 5 秒 (B-roll Film)

https://github.com/user-attachments/assets/6628b678-78ff-481a-bc39-d7098ec1f2fe

▶《冷萃咖啡产品广告》· 5 秒 (C-roll Film)

模式	影片	选题 / 主体	播放视频	拼贴海报封面	动效模型	旁白音色
B-roll	☕ 咖啡简史	埃塞俄比亚公元 850 年	播放视频 (`out/demo/final.mp4`)	`out/demo/keyframes/kf_1a.jpg`	`veo3.1-image-to-video`	MiniMax TTS (`Q19bea09caa6IRAeW7`)
C-roll	🍾 冷萃咖啡产品广告	冷萃咖啡瓶照片锚定	播放视频 (`out/demo-croll/final.mp4`)	`out/demo-croll/keyframes/kf_1.jpg`	`veo3.1-image-to-video`	MiniMax TTS (`Q19bea09caa6IRAeW7`)

🎬 生成素材汇总：

☕ B-roll (咖啡简史):

📹 视频: 播放视频 (out/demo/final.mp4) | 🖼️ 海报: out/demo/keyframes/kf_1a.jpg | 📹 动效: out/demo/clips/clip_1a.mp4 | 🎙️ 旁白: out/demo/audio/narr_1.mp3 | 🎵 配乐: out/demo/audio/bgm.mp3

🍾 C-roll (冷萃咖啡产品锚定):

📹 视频: 播放视频 (out/demo-croll/final.mp4) | 📸 锚定原图: out/demo-croll/anchor_photo.jpg | 🖼️ 海报: out/demo-croll/keyframes/kf_1.jpg | 📹 动效: out/demo-croll/clips/clip_1.mp4 | 🎙️ 旁白: out/demo-croll/audio/narr_1.mp3 | 🎵 配乐: out/demo-croll/audio/bgm.mp3

这是什么

风格是 Vox 讲解片带火的现代编辑感纸质拼贴:手撕纸片、毛边、胶带、半调网点、报纸剪贴、每一拍一块大胆平涂色、大号剪纸标题——再配上动效、旁白、配乐和字幕，让整张海报活过来。

工作原理

一个选题依次流过每个阶段脚本，全程由每个项目一份 beats.json 驱动:

选题 / 视频 / 照片
  │
  ├─ 1. 分镜脚本   选叙事弧线 → 写 beats.json          ◀── 决策点 1:你确认分镜脚本
  ├─ 2. 风格试片   同一拍渲成 3–4 种主题               ◀── 决策点 2:你看图挑风格
  ├─ 3. 关键帧     每拍一张拼贴海报   (nano-banana-2 / flux-dev)
  ├─ 4. 动效       让每张海报动起来   (veo3.1-image-to-video)
  ├─ 5. 旁白+配乐  统一旁白 (minimax-speech-2.6) + 背景乐 (suno-create-music)
  ├─ 6. 合成       ffmpeg:拼接、配乐在旁白下自动闪避、烧字幕+水印
  └─ final.mp4

三种输入形态复用同一套引擎:

B-roll——一个选题进去，画面全靠生成。 它写脚本、生成每张拼贴海报、用 Veo 3.1 生成动效、配旁白与音乐，合成视频。
A-roll——你已经有一段口播视频。 它会被 ASR 自动切成段（openai-whisper），再整段套上拼贴风格，真人的脸、口型、手势逐帧保留（gemini-omni-video-edit / veo3.1）。
C-roll——你只有一张静态照片（自拍、产品图）。主体被抠成摄影质感的贴纸——绝不重绘——每一段的海报围着它生成（nano-banana-2 / flux-dev）。旁白支持声音克隆（minimax-voice-clone）。

模型（已在 MuAPI 上验证）

用途	模型
关键帧 / 拼贴海报	`nano-banana-2` / `flux-dev`
动效 / 视频生成	`veo3.1-image-to-video` / `veo3.1-fast-image-to-video`
口播视频转拼贴 (A-roll)	`gemini-omni-video-edit` / `veo3.1-image-to-video`
照片锚进拼贴 (C-roll)	`nano-banana-2` / `flux-dev`
旁白 TTS	`minimax-speech-2.6-turbo`
声音克隆	`minimax-voice-clone`
背景音乐	`suno-create-music`
语音识别 (A-roll ASR)	`openai-whisper`
抠图背景消除	`remove-background`

快速开始（三种模式）

export MUAPI_API_KEY="sk-..."

# 1. B-roll (选题成片)
python scripts/style_bakeoff.py out/my-topic american-retro,swiss-modern,punk-zine
python scripts/keyframes.py out/my-topic
python scripts/clips.py out/my-topic
python scripts/audio.py out/my-topic
python scripts/assemble.py out/my-topic

# 2. A-roll (口播视频转拼贴)
python scripts/asr_beats.py out/my-aroll source_presentation.mp4
python scripts/aroll_clips.py out/my-aroll
python scripts/aroll_assemble.py out/my-aroll

# 3. C-roll (单张照片/产品锚定拼贴)
python scripts/croll_keyframes.py out/my-croll
python scripts/clips.py out/my-croll
python scripts/audio.py out/my-croll
python scripts/assemble.py out/my-croll

Veo 3.1 ComfyUI 节点

ComfyUI custom nodes for generating videos with Google's Veo 3.1 model via the MuAPI platform.

Veo 3 on MuAPI — Model landing page for Veo generation.
Veo 3 text-to-video playground — Try the model directly in the browser.
veo4-video-generator — Ready-made Next.js SaaS for Veo — no ComfyUI needed
Veo-4-API — Python wrapper for Veo 4 API — use the latest Veo model in scripts
muapi-comfyui — ComfyUI nodes for 100+ MuAPI models including Veo
awesome-ai-video-models — compare AI video models by API, price & speed

Nodes

Node	Description
🎬 Veo 3.1 Text to Video	Generate 8-second video from a text prompt
🎬 Veo 3.1 Image to Video	Animate a static image; optionally anchor the last frame
🎬 Veo 3.1 Reference to Video	Generate video guided by up to 4 reference images
🎬 Veo 3.1 Extend Video	Continue a previous generation with a new prompt
🎬 Veo 3.1 4K Upscale	Upscale any previous Veo 3.1 generation to 4K
🎬 Veo 3.1 Save Video	Download & save generated video; returns frames tensor

All nodes live in the 🎬 Veo 3.1 category in the ComfyUI node menu.

Available Models

Text to Video

Model	Speed	Quality
`veo3.1-text-to-video`	Standard	Highest, with audio
`veo3.1-fast-text-to-video`	Fast	Good
`veo3.1-lite-text-to-video`	Fast	Lightweight

Image to Video

Model	Speed	Quality
`veo3.1-image-to-video`	Standard	Highest, with audio
`veo3.1-fast-image-to-video`	Fast	Good
`veo3.1-lite-image-to-video`	Fast	Lightweight

Other Variants

veo3.1-reference-to-video — multi-image reference generation
veo3.1-extend-video — extend a previous generation
veo3.1-4k-video — upscale a previous generation to 4K

All models output 8-second videos (Veo 3.1 fixed duration).

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/YOUR_USERNAME/muapi-veo31-comfyui
pip install -r muapi-veo31-comfyui/requirements.txt

Restart ComfyUI.

Setup

Get an API key from MuAPI
Paste it into the api_key field of any Veo 3.1 node

Parameters

Common

Parameter	Description
`api_key`	Your MuAPI API key
`prompt`	Text description of the video
`aspect_ratio`	`16:9` or `9:16`
`resolution`	`720p`, `1080p`, or `4k`
`extra_params_json`	Any additional model parameters as JSON

Image to Video extras

Parameter	Description
`image`	Start frame (IMAGE tensor)
`last_image`	Optional end frame for first–last mode

Reference to Video extras

Parameter	Description
`image_1` … `image_4`	Reference images (up to 4)
`generate_audio`	Whether to generate audio (default: true)

Extend / 4K Upscale

Parameter	Description
`request_id`	`request_id` output from a previous generation node

Example Workflows

File	Description
`MuAPI_Veo31_T2V_Example.json`	Text → Video → Save
`MuAPI_Veo31_I2V_Example.json`	Image → Video → Save
`MuAPI_Veo31_Reference_Example.json`	2 reference images → Video → Save

Load any workflow via ComfyUI → Load (drag & drop the JSON).

Chaining nodes

Veo31TextToVideo
  └─ video_url  ──► Veo31VideoSaver ──► frames ──► PreviewImage
  └─ first_frame──► PreviewImage
  └─ request_id ──► Veo31ExtendVideo
                       └─ request_id ──► Veo314KUpscale

Requirements

Python 3.8+
ComfyUI (any recent version)
requests, Pillow, numpy, torch, opencv-python

License

MIT

热度有据可查

累计 Star: 299
近 7 天净增: 观察中

已开始记录真实快照。
数据积累后展示趋势。

尚无约 7 天前的观测，不估算周净增。已有 1 天观测。

统计口径 ↗

项目档案

主要语言Python

许可证未提供

Fork52

最近推送2026/9/2 23:19:37

本站发现2026/10/29 13:59:35

数据更新2026/10/29 13:59:35

收录与分类依据

仓库简介：🎬 Turn any topic into a finished Vox-style paper-collage explainer / motion graphics video — script, collage keyframes, animation, voice-over, music & captions, all automated. An agent skill for Claude Code, Codex & other coding agents.

GitHub Topics：agent-skill, ai-motion-graphics, ai-video, claude-code, collage, explainer-video, ffmpeg, flux, image-to-video, motion-graphics, muapi, paper-collage

README 依据：已核实仓库中的 SKILL.md 文件或明确的技能安装命令

发现来源：分类检索 · 语音与音乐