FrameTamer — edit video by editing text

A desktop video editor in the spirit of Descript. A video is transcribed with word-level timestamps, and editing the transcript edits the video: delete a word and the corresponding fragment disappears from the cut. AI handles the tedious parts — finding filler words, generating chapters and applying edits described in plain language.

The UI ships under the working name ClipForge.

Contents

• Features
• How it works
• Architecture
• API
• Getting started
• Configuration

Features

Feature	Description
Projects	Upload a video, get a project with thumbnail, metadata and search.
Transcription	Local speech-to-text with faster-whisper and word-level timestamps — audio never leaves the machine.
Text-based editing	Delete or restore words in the transcript; the timeline is re-synced to the remaining words.
Waveform	wavesurfer.js waveform for precise navigation and range selection.
Filler words	Detects filler words in Russian and English speech so they can be cut in one pass.
Chapters	Generates chapters from the transcript.
Smart edit	Describe an edit in natural language ("remove the intro", "keep only the answers") and let Claude apply it.
AI agent tasks	Long-running Claude Code agent tasks per project with live log streaming and a kill switch.
Export	Cuts and concatenates the kept fragments with ffmpeg, with progress reporting and download.
Desktop app	Electron shell with native open/save dialogs, packaged with electron-builder.

How it works

flowchart LR
    V[Video file] --> U[Upload]
    U --> T["faster-whisper<br/>word-level timestamps"]
    T --> TR[(Transcript<br/>words + times)]
    TR --> E[Text editor<br/>delete / restore words]
    TR --> AI["Claude<br/>filler words · chapters · smart edit"]
    AI --> E
    E --> TL[(Timeline<br/>kept ranges)]
    TL --> X["ffmpeg<br/>cut & concatenate"]
    X --> O[Edited video]

Loading

sequenceDiagram
    participant UI as Frontend
    participant API as Backend
    participant W as faster-whisper
    participant C as Claude
    participant F as ffmpeg
    UI->>API: POST /api/media/upload
    UI->>API: POST /api/projects/{id}/transcribe
    API->>W: transcribe (word timestamps)
    W-->>API: words[]
    UI->>API: POST /api/projects/{id}/ai/filler-words
    API->>C: transcript + instructions
    C-->>API: ranges to remove
    UI->>API: POST /api/projects/{id}/editor/words/delete
    UI->>API: POST /api/projects/{id}/editor/sync-timeline
    UI->>API: POST /api/projects/{id}/export
    API->>F: cut & concatenate kept ranges
    UI->>API: GET /api/projects/{id}/export/progress
    UI->>API: GET /api/projects/{id}/export/download

Loading

Architecture

flowchart TB
    subgraph Desktop["Electron app"]
        FE["Frontend<br/>React 19 · Vite · Tailwind v4<br/>TanStack Query · wavesurfer.js"]
        NB["backend-node<br/>Express · Claude Code agent runner"]
    end
    PY["backend (Python)<br/>FastAPI · faster-whisper · Anthropic SDK<br/>ffmpeg-python · librosa"]
    FE <-->|REST + SSE| NB
    FE <-->|REST| PY
    NB -->|"spawn claude --output-format stream-json"| CC[Claude Code CLI]
    PY --> FF[ffmpeg]
    PY --> AN[Anthropic API]

Loading

frontend/       — React 19, Vite, TypeScript, Tailwind CSS v4, TanStack Query, React Router, wavesurfer.js
backend/        — FastAPI: projects, media, transcription, editor, AI and export routers
backend-node/   — Node/TypeScript server: projects, media and Claude Code agent tasks with SSE log streaming
electron-app/   — Electron shell; bundles the frontend and backend-node into a desktop app

API

Python backend (backend/):

Method	Path	Purpose
POST / GET	/api/projects	create / list projects
GET / PATCH / DELETE	/api/projects/{id}	read / update / delete a project
GET	/api/projects/{id}/thumbnail, /video	preview and source video
POST	/api/media/upload	upload media
POST	/api/projects/{id}/transcribe	run transcription
GET / PUT	/api/projects/{id}/transcript	read / save the transcript
GET	/api/projects/{id}/waveform	waveform peaks
POST	/api/projects/{id}/editor/words/delete, /restore	text-based cuts
POST	/api/projects/{id}/editor/sync-timeline	rebuild the timeline from the transcript
POST	/api/projects/{id}/ai/filler-words	detect filler words
POST	/api/projects/{id}/ai/chapters	generate chapters
POST	/api/projects/{id}/ai/smart-edit	natural-language edit
POST	/api/projects/{id}/export	start export
GET	/api/projects/{id}/export/progress, /download	export progress and result

Agent tasks (backend-node/): GET /tasks, GET /projects/:id/tasks, GET /tasks/:taskId, GET /tasks/:taskId/stream (SSE), POST /tasks/:taskId/kill, POST /projects/:id/transcribe.

Getting started

Requirements: Python 3, Node.js, ffmpeg in PATH, an Anthropic API key (and the Claude Code CLI for agent tasks).

# Python backend
cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python run.py --reload            # http://localhost:8000

# frontend
cd ../frontend
npm install
npm run dev

# desktop app (dev mode) and distributable build
cd ../electron-app
npm install
npm run dev
npm run build

Configuration

Variable	Description
ANTHROPIC_API_KEY	required for AI features
ANTHROPIC_MODEL	Claude model to use
WHISPER_MODEL	tiny … large-v3
WHISPER_DEVICE	auto / cpu / cuda
WHISPER_COMPUTE_TYPE	auto / float32 / float16 / int8
PROJECTS_DIR, MEDIA_DIR, TEMP_DIR	storage locations (default ~/ClipForge/…)
HOST, PORT, DEBUG	server settings