Production Studio

A WISER plugin for producing and editing video, training videos first: scripts written for ear and eye, narration in your own voice clone, screen walkthroughs, animation, editing and publishing. Its first skills, experts and tools are being built; each family's AGENTS.md index lists what it holds.

Install

Point your harness at this repository as a plugin root, or add the repository as a marketplace and install production-studio@wiser-method-production-studio from its catalog.

Work lands in the working folder you attach, never in this repository.

What it needs

Production Studio loads beside the base plugin wiser, which carries the connectors and the tools it calls. What each part needs, as it stands today:

• Narration (skills/Narrate/): the root's route and the person's voice clone recorded by Set Up Voice Clone, Node 22 or later, and FFmpeg with ffprobe. ElevenLabs needs its speech module connected through Wiser's local gateway in Claude Code, Codex, Grok or Cursor. HeyGen needs its speech and account modules connected to your own API workspace through that local gateway, since narration is written as a WAV file. Your HeyGen API key is entered only on the provider's connect page through Connect Account, never in conversation. It bills a self-serve API wallet in US dollars, separate from any HeyGen web subscription, so subscribers pay again from that wallet. The free route runs locally through wiser's Chatterbox, with English word alignment through Transcribe Audio, which needs CPython 3.9 to 3.13 and its separate alignment packages and weights. All three routes check English takes through that tool's independent transcription, which needs Python 3.8 or newer and downloads its speech packages and model weights once, when you allow it. Packages stay in the tools' own folders and weights in the person-scoped models folder.
• A voice clone (skills/Set Up Voice Clone/): choose ElevenLabs on your own account and plan, HeyGen on your own API wallet, or the free local route. An existing ElevenLabs clone can be listed and confirmed; creating one uploads a measured MP3 through the local gateway and may need ElevenLabs' own verification. HeyGen needs its voices module connected through Connect Account; an existing private instant clone is listed and confirmed, while a new one uploads one measured MP3 or WAV of at most 3,000,000 bytes on the local gateway, with gateway approval. Creation acknowledges acceptance; setup waits for ACTIVE before recording the clone. A workspace's clone limit stops setup without deleting an existing clone. The free route needs an Apple silicon Mac on macOS 12.3 or later, native arm64 CPython 3.11 and FFmpeg. There is no RAM floor: Chatterbox's offline check and your accepted reference-recording trial decide, refusing more than 3.0 compute seconds per second of audio. Short scenes can be slower than the trial. Setup states its lower quality and Resemble's Perth watermark. The route takes about 4.7 GB including alignment; the tool conservatively checks 6.7 GiB free at each destination including an installation reserve. Downloads follow your route choice and install consent; no recording is used for cloning before the one consent answer is saved.
• Scripts (skills/Write Script/): stills of the product's steps, taken for the script. Nothing to install.
• Cutting recordings (tools/Timeline/): Python 3.11 to 3.13, an Apple Silicon Python on a Mac, and FFmpeg with libx264 and ffprobe. With your consent it installs its packages, about 95 MB, into its own folder.
• Take filing (tools/Narration Takes/): Node 22 or later and ffprobe. Nothing to install.
• Walkthrough scenes (tools/Scene Composer/): Node 22 or later, and the font files the scene's text uses. Nothing to install. Its scenes render through wiser's video-render 0.2.0 or later, which installs HyperFrames and its own Chrome once, when you allow it.
• Animated scenes (skills/Animate Scenes/): everything walkthrough scenes need, above, and FFmpeg for video-render's files. An explainer scene, such as a graphic of a process's steps, runs wiser's Animated Explainer, which draws through wiser's render and asks you to approve its words. With a HeyGen account, it can also show your existing avatar speaking the scene's current take, or in a bottom-right window for a named narration span over a walkthrough or explainer. It uploads that take's prepared narration and requests an Avatar IV render, showing the selected look's per-second rate and a two-render worst case before billing your separate API wallet. Uploads need their own gateway confirmation; renders need confirmation unless you grant a production standing approval, which never replaces the cost yes. This optional path needs HeyGen's avatars, assets, videos and account modules on Wiser's local gateway, Speech Preparation 0.2.0 and video-edit 0.3.0 or later; other scenes need no HeyGen account. It creates no new likeness and draws no avatar captions. Each scene is checked from frames and records by the Animator in a second context, which an app that can open one, such as Claude Code or Cowork, gives; in another app you open a new chat for that step.
• The finished video (skills/Assemble Video/): wiser's video-edit 0.2.0 or later, which needs Node and FFmpeg with ffprobe, libx264 and libopus; Node 22 or later for Scene Composer's caption join. Nothing to install. It does not yet cut a screen recording into scenes.
• Publishing (skills/Publish Video/): an app that can attach Wiser's local gateway, with wiser's YouTube connector connected to your own channel through your own Google app. It also needs FFmpeg with ffprobe for the thumbnail's frame, and wiser's render for the thumbnail. A custom thumbnail needs a verified channel, and chapters show once the channel has YouTube's Advanced features. Uploading is free; it uses your Google project's daily quota.
• Produce Video (skills/Produce Video/): a native macOS host for capture and the finished video's privacy check, wiser's Browser Control for a web product, and wiser's Screen Capture with its permissions for a macOS app, besides everything the skills it runs need, above.

ElevenLabs narration is billed by the account's plan; Narrate shows character-based estimates before billing, and every synthesis needs gateway confirmation unless you grant a standing approval for a production's narration, which lasts at most 7 days and never replaces the cost approval. HeyGen narration defaults to heygen-voice-1 with expressiveness_boost: 0.5; its rate is USD 15 per million submitted characters, read 2026-10-28 in wiser's HeyGen connector page. Narrate shows the selected scenes' cost and two-take worst case and asks once per production; bounded HeyGen synthesis needs no per-scene gateway confirmation or standing approval. The promotional heygen-voice-1-turbo-free model is used only on your explicit word and is refused from 2026-12-01 UTC, with no silent model change. A route or narration-setting change narrates every scene again, with cost shown first and earlier takes retained. The free route costs no money; Narrate shows expected compute time from your accepted trial. Its alignment supports English and a restricted token set; unsupported text stops before synthesis.