cutmaster panel

CutMaster panel

The CutMaster panel is the AI editing product that lives inside the same package as the MCP server. Point it at a raw or assembled timeline; pick a preset; click through four screens; get a new timeline with the cut already assembled, ready to review and adjust.

Requires DaVinci Resolve Studio + GEMINI_API_KEY. The free Resolve edition will not work. The panel needs a Gemini API key (free tier is fine) — get one at aistudio.google.com/apikey.

What it does

Works on raw footage — drop a folder into a bin, the panel transcribes, detects themes, builds a cut.
Works on assembled cuts too — Clip Hunter turns a finished YouTube edit into 3 TikTok shorts in one run.
Speaker-aware — multi-speaker transcripts stay attributed.
Non-destructive — every run creates a new _AI_Cut_N timeline; your source is never touched.
Snapshot-based safety — auto-snapshots before building; rollback is instant.

The four screens

Preset — pick what you're making. Today: Clip Hunter (YouTube → Shorts), Tightener (clean a cut), Sequencing (raw → narrative), Curated (take selection).
Analyze — transcribes the audio + detects themes. ~1 minute per 10 minutes of footage.
Configure — tweak: number of clips, target length, themes to prioritise, excludes, output format.
Review — inspect the proposed cut, hit Build, and a new timeline appears in Resolve.

Status — where the panel runs today

The in-Resolve dock panel (via Workflow Integration plugin) is not shipping yet — scheduled for v3-6. Until then, the panel runs as a local web app in your browser, sitting next to Resolve. Identical features, no plugin install required.

The browser-based panel and the future docked panel share the same React app and the same FastAPI backend. When v3-6 ships, you'll move from a browser tab to a Resolve dock pane without losing any state.

Setup

For the full editor-facing setup (5 minutes), see the canonical guide on GitHub:

docs/CUTMASTER_SETUP.md ↗ — clone, set the Gemini key, start the backend, open the panel.

The short version:

bash

$ git clone https://github.com/CelaviiHQ/cutmaster-ai.git
$ cd cutmaster-ai
$ bash scripts/setup.sh --panel    # installs backend deps
$ cp .env.example .env             # set GEMINI_API_KEY
$ uv run cutmaster-ai-panel        # starts backend
$ open http://127.0.0.1:8765/      # opens the panel

How it relates to the MCP server

The panel and the MCP server share a Python package. They're different surfaces over the same code:

The MCP server exposes the atomic / workflow / intelligence tools to Claude over stdio.
The panel exposes the stateful CutMaster product to humans over HTTP / SSE.
Both can run at the same time. They don't conflict.

If you only want Claude to drive your timeline, use the MCP server (see Install). If you want a finished-cut-from-footage product with no terminal, use the panel.

What's coming

v3-1 — Preset Pick screen revamp
v3-2 — Analyze screen polish + Cancel button
v3-3 — Configure screen per-preset split
v3-4 — Review screen reorg + "Open in Resolve" button
v3-6 — Resolve Workflow Integration plugin — the panel docks inside Resolve itself
v3-8 — Live progress feedback during LLM calls (stage events, elapsed counters, token-streaming indicator)

Deeper reading

docs/CUTMASTER_SETUP.md ↗ — full setup with troubleshooting.
docs/CUTMASTER_ARCHITECTURE.md ↗ — pipeline structure: STT → analysis → director → execute.
Architecture — where the panel fits in the four-bucket model.