v1.3.1: ready for npm

openvidstudio 1.3.1 is a packaging release with no code changes. It makes the four packages ready for npm.

What changed

• @openvidstudio/mcp-server, @openvidstudio/core and @openvidstudio/stem now each ship a README written for the npm page, and @openvidstudio/mcp-server and @openvidstudio/stem ship the Apache-2.0 LICENSE file they were missing. Before, the mcp-server npm page would have been blank.
• All five packages and the server's reported version are 1.3.1.

How it was found: all four packages were packed with pnpm pack before the first real publish and the tarballs inspected. The 1.3.0 tarballs were also installed into an empty directory; the server started and listed its 41 tools.

Install

From a clone, as before:

git clone https://github.com/AnayDhawan/openvidstudio.git
cd openvidstudio && pnpm install && pnpm build

The npm packages are published separately from this release (see the README for the current install). Until @openvidstudio/mcp-server 1.3.1 is on npm, npx -y @openvidstudio/mcp-server still installs 0.1.0 from September.

Full list of changes: CHANGELOG.md.

v1.3.0: one shape for every lesson

openvidstudio 1.3.0 gives Supernova lessons one shape, and teaches check_script two more lines of it. No lesson has been published yet, so treat the mode as built and tested, not proven in the wild.

What is new

• Every lesson is one concept in up to two minutes, in seven fixed beats: hook, picture, set up, derive, use it, aha, close. SCRIPT.md ("The arc") has the times for a 120 second lesson, the rules (the diagram starts inside 15 seconds, each formula is built line by line with a spoken form per line, the lesson ends on its check question) and a correctness routine: write the derivation, check the maths and the units with code, take every number on screen from that code, and have someone who knows the subject read it.
• check_script checks two more things. A lesson that runs over its budget (120 seconds unless the concept card sets maxSeconds) is an error. If the card has a reference such as a chapter code, the opening beat has to say its referenceSpoken (an error; a warning when referenceSpoken is missing, because then it cannot be checked). The concept card gained those three optional fields.
• The supernova skill is rewritten around the arc. It now has a derive-and-verify step, a step that sizes each beat from the measured voice, and the notes that save time on a first lesson.
• SUPERNOVA.md covers what building the first lesson taught: how to fit beats to the voice (it reads plain prose at about 3 words per second and number-heavy lines at about 2.2), how to caption (the word timings carry no punctuation, so take the text from the beat), that render_video needs skipBrandLock: true for a lesson, and that edge-tts 7.3 or newer is needed (7.2.8 gets a 403 from Microsoft).
• click.wav is a recorded double click (Mixkit) instead of a synthesized tick. public/sfx/CREDITS.md records where it came from and the licence.
• The doc map in OVERVIEW.md listed 6 of 13 docs. It lists all of them now.

Upgrading

• A lesson over 120 seconds now fails check_script. Set maxSeconds on its concept card if you really want longer.
• Existing projects keep their own copy of click.wav. Copy the new one from packages/core/public/sfx/ if you want it.

Install

Install from a clone. npx -y @openvidstudio/mcp-server still installs 0.1.0 from September, because publishing is blocked.

git clone https://github.com/AnayDhawan/openvidstudio.git
cd openvidstudio && pnpm install && pnpm build

What comes next

Ten issues are open for the follow-up work (#133 to #142): a component that shows a derivation step by step, a caption component, a tool that fits beats to the voice, three small fixes found while building a lesson, arc checks in validate_scenes, and more physics and computer science diagrams.

Full list of changes: CHANGELOG.md.

v1.2.0: Supernova mode, and a new tagline

openvidstudio 1.2.0 adds Supernova mode, the engine for short animated STEM lessons, built by @yakew7. It also makes tools reject inputs they do not take, and fixes a handful of Windows test problems.

Supernova is the engine only. No lesson video has been built and listened to yet, so treat it as ready to try, not proven. The first pilots are tracked in #107.

What is new

• Supernova mode. init_project takes mode: "supernova": a 1080x1920 project with the Void + Spark design system (a near-black canvas where only the thing you must look at glows, with an accent colour per subject), a new @openvidstudio/stem package and a supernova skill. A standard project is unchanged.
• STEM primitives. Spark (the pen, the pointer and the particle), Nova (the payoff), Axes, Plot and NumberLine, Equation (sans-serif maths through MathJax and the Fira font, with a required spoken form), FreeBody and ArrayViz.
• Script first. A new check_script tool lints a lesson script (sentence length, reading level, terms defined before use, the spoken form matching the narration). generate_narration writes word timings with the edge-tts engine, so a scene can point at the word being said. validate_scenes knows about lessons and counts how many things glow at once.
• CollageBroll. A halftone paper-collage b-roll primitive in the motion primitives catalog, with no paid API.
• plan_production's time estimate now includes the final render, a bundle per review round and capture retries, scaled by how heavy the production level is. It is calibrated on one recorded run, so treat it as a better estimate, not a promise.
• scaffold_scene follows the project's frame size, so a vertical project gets centred scenes. write_beats_file returns the validate_beats warnings. The API gallery has a still for every new primitive, and the DirectedShot still has sharp text.

Upgrading

• Tools now reject unknown inputs. A call like render_video { quality: "draft" } (the option is draft) used to run a full render silently. It now fails with an error that names the unknown input, lists the valid ones and suggests the closest. A client that sends extra keys will see errors it did not before.
• A fresh pnpm install now pulls MathJax and a font package for the new package, so the first install takes longer.
• plan_shots now reads a changelog correctly after a release cut, and the README's differences section was renamed so it keeps being read.

Install

Install from a clone. npx -y @openvidstudio/mcp-server still installs 0.1.0 from September, because publishing is blocked.

git clone https://github.com/AnayDhawan/openvidstudio.git
cd openvidstudio && pnpm install && pnpm build

Known limits

• Word timings need network access to Microsoft's service, so that test is skipped without it. If narration fails with a 403 from Microsoft, upgrade the engine: pip install -U edge-tts (7.2.8 was rejected, 7.3.1 works).
• The Windows test suite had portability bugs (a tsc shim, CRLF golden files, one text-fringe pixel); they are fixed in this release.

Full list of changes: CHANGELOG.md.

v1.1.0: director mode, a cinematic camera, native 9:16

openvidstudio 1.1.0 is the first release since the September launch. It adds a director mode, a cinematic camera, native vertical video, and a soundtrack that no longer ticks.

What is new

• Director mode. For a launch video or a trailer, the agent writes a creative brief for you to approve, builds scenes by hand from a design system, reviews contact grids of real frames before a full render, and finishes with a loudness target, a poster frame and a description with chapters and credits. Pick the peak or mid level when it plans the production.
• A cinematic camera. Shot moves (push in, pull out, pan, whip, orbit, dutch, hold, drift), handheld sway, an impact shake, real depth between layers, and motion blur that only runs while the camera moves. Depth of field is off by default now, so it no longer smears the text you are showing. Cursors arc and settle. Captions finish their reveal within 8 frames and can sit top, bottom or either side.
• Transitions that work. cut, whip, fade, push, wipe and light-leak are drawn over the cut, so beat timing and audio do not move. A page can also be captured as depth layers for parallax.
• Native 9:16. A project can be 1080x1920 from the start, the primitives follow the frame size, and export_rendition has a reels format.
• No ticking. The original score has no drums by default, because evenly repeated hits sound like a ticking clock under a voiceover. finalize_video now warns when a finished mix still ticks, and the click-track bed (pulse-bed.mp3) is gone.
• Review tools. render_stills crops to a region for checking small text, and validate_beats warns when every shot is the same length.

Upgrading

• Depth of field, the per-word title blip and the old drum arrangement are now opt-in (depthOfField, sfx, percussion: "full").
• Layer orders by depth, not by order in the file, and must be a direct child of the camera stage. Scenes that relied on the old parallax or on paint order may look slightly different.
• export_rendition with an unknown format is an error instead of quietly producing store frames.
• preflight asks for Node 20, which is what the packages already declared.
• stitch_composition only picks up music-bed.mp3 on its own. A leftover pulse-bed.mp3 is ignored unless you name it.

Install

Install from a clone. npx -y @openvidstudio/mcp-server still installs 0.1.0 from September (about 27 tools, no director mode), because publishing a new version is blocked for now.

git clone https://github.com/AnayDhawan/openvidstudio.git
cd openvidstudio && pnpm install && pnpm build

Then point your MCP client at packages/mcp-server/dist/stdio.js. The README has the config.

Not in this release

Lesson videos (a mode for short STEM explainers) are not here. They are tracked in #107 and do not work yet.

Full list of changes: CHANGELOG.md.