Connect the AI
The producer needs an AI engine to chat with you, propose the script and decide the cuts. You pick one of the two from Settings inside the app.
Option A — Claude Code (your own subscription)
If you already have a Claude subscription and use Claude Code in the terminal, this is the simplest option: no key to enter, the app detects it on its own.
| Step | What to do |
|---|---|
| 1. Install | Follow the official Claude Code installation guide: docs.claude.com — Claude Code setup. |
| 2. Sign in | Run claude once in Terminal and complete the login with your account. |
| 3. Done | Open (or reopen) Video Maker — the first run or Settings will show that Claude was found. |
Option B — OpenRouter API key
If you do not have Claude Code, OpenRouter gives you access to many models with a single key and prepaid credit.
| Step | What to do |
|---|---|
| 1. Create the key | Go to openrouter.ai/keys, sign up if needed and create an API key. |
| 2. Add credit | At openrouter.ai/settings/credits you load the prepaid credit you plan to use — it is not a subscription. |
| 3. Paste it in the app | Settings → Producer → Engine: OpenRouter → paste the key. The app only keeps a hint (for example "Key loaded ••••56f7") and never shows it in full. |
Recommended models and real cost
Prices checked live on the OpenRouter catalog (2026-09-26). The cost of a Tutorial recipe was measured by running the full video end to end (script, voice, music and render) with real material — 817,989 input tokens and 42,651 output tokens over 4 turns — and then recalculated at each model’s current price:
| Model | Price per 1M tokens | Cost of a full Tutorial recipe |
|---|---|---|
| DeepSeek V4.1 Flash (recommended, the app’s default) | US$0.14 input · US$0.42 output | ≈US$0.13 (recalculated at the current price) |
| Claude Haiku 4.5 (quality reference) | US$1.00 input · US$5.00 output | ≈US$1.03 (recalculated; the same video was not run with Haiku) |
We do not recommend OpenRouter’s free models. We tried them with the real producer and none of them finished the job: they get saturated at peak hours, have per-day and per-minute request caps, some only work inside certain apps and, depending on your account’s privacy settings, what you send them may be used for training. DeepSeek V4.1 Flash costs cents per video.
If you still want to try one, these are the free ones that currently accept tools and images (what the producer needs): google/gemma-4-31b-it:free, qwen/qwen3.8-27b:free, google/gemma-4-26b-a4b-it:free and nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free. Look for it in Settings → "All models" tab, with no guarantee it will finish the video. First, check openrouter.ai/settings/privacy.