# Current optional free local mode · 0.5.0

Use [free-local.md](free-local.md) for local planning, Qwen narration, forced word alignment and the complete pipeline. Its models need about 6.7 GB plus environment/cache/output space. Tested on Apple M1 Pro / 32 GB macOS. No hosted planning, voice or rendering service fee after setup. Hardware, electricity, installation and optional assistant/premium media costs remain separate.

The basic renderer instructions below remain applicable. Older voice instructions describe preserved examples.

# Setup, requirements, and costs

Verified 23 October 2026. The core workflow is local after its tools and assets are installed. No video rendering API key or per-video cloud charge is required.

## Required renderer

1. Install a supported Node.js release (Node 20.9+; tested 24.19.0) and FFmpeg with `libx264`, plus `ffprobe`, from sources you trust. See [Node downloads](https://nodejs.org/en/download) and [FFmpeg downloads](https://ffmpeg.org/download.html). Put both executables on your PATH. The kit does not redistribute these binaries.
2. Open a terminal in the kit folder and run `npm ci`. The pinned direct dependency is `sharp` 0.35.5. Its initial installation downloads platform-specific native libraries. See [sharp installation](https://sharp.pixelplumbing.com/install/).
3. Run `node scripts/render.cjs doctor` and `npm test`. A successful doctor is a dependency check, not a guarantee of all exports.
4. Render a supplied sample. Keep several gigabytes of working storage available for dependencies, optional voice models, and your outputs.

If executable paths are unusual, pass `--ffmpeg /path/to/ffmpeg --ffprobe /path/to/ffprobe`, or set `FFMPEG_PATH` / `FFPROBE_PATH`. Paths are local configuration; do not bake them into shared plans. On Windows, use your shell's quoting and a current compatible FFmpeg distribution. Windows/Linux rendering and font behavior remain untested in this prototype.

The renderer runs on the CPU. No discrete GPU is required for the tested sample. 8 GB RAM is a practical starting recommendation, not a tested minimum; larger images, 4K frames, other open applications, and optional voice models raise memory needs. The package was tested on Apple M1 Pro with 32 GB RAM. It streams frames into the encoder and does not keep every frame on disk.

## Costs customers should understand

| Component | What the kit requires | Per-video service fee |
|---|---|---|
| Local scene renderer | Included code + locally installed Node, sharp, FFmpeg | None charged by the kit |
| Supplied narration/recording | Audio file you have rights to use | None charged by the kit |
| Optional local planning, voice and alignment | Separately downloaded pinned Qwen/MLX models and Python setup; see free-local.md | No paid API call after setup |
| AI assistant for planning | Optional; use your own supported assistant/account or edit JSON manually | Depends on the chosen assistant |
| Computer, electricity, storage | Customer-provided | Real running costs, not included |
| Stock assets, music, premium voices, image/video generation | Optional, not bundled | Provider/license dependent |

No model downloads or account purchases occur automatically. A paid AI subscription is not included in the kit price. The optional local planner uses separately downloaded Qwen weights and the included helper. Model weights and runtime packages are not bundled or installed automatically. See free-local.md for exact setup and tested hardware. Avoid claiming “all videos free forever” or “any computer, no setup.” The precise claim supported here is **no per-video rendering fees for this local workflow**.

The tested optional voice environment used Python 3.12 and pinned packages in `requirements-voice.txt`. The sample ships its WAV, so reproducing the supplied video does not need that environment. A fresh project-directory npm ci install with a separate empty cache was verified on the same Mac. Package availability may change; recreate a clean environment and test before offering support on another platform. Dependency security updates and installer maintenance remain a product-owner responsibility.

A real npm lockfile is included. A separate fresh folder was installed with official npm 12.2.1, then reinstalled using `npm ci` with an empty cache. The doctor, 29 core checks, 10 overlay checks, and a 1080p narrated render passed with network access denied after installation. This is an independent clean-directory dependency test on the same existing Mac and FFmpeg toolchain, not a fresh operating-system installation or a Windows/Linux guarantee.

Current toolkit price: $67 once. Local planning, narration and rendering have no per-video service fee after setup. ChatGPT/Codex or Claude usage is optional, consumes your own plan limits or credits, and is billed separately by that provider. Premium media inputs are separate. Rendering supplied audio/scripts needs no AI model.