FervorCreative AI
Live Latest 18.11.26 · morning 401 tools tracked 992 workflows indexed 626 topics Hot: Gaussian splatting, ComfyUI, ZONOS3-Mini

Open Explainer makes the spoken phrase, not the timestamp, the unit of editing for explainer videos, which fixes the timing problem that makes most of them feel cheap, and its built-in checks matter more than its animations.

Open ExplainerClaude CodeChatGPTvideo-gencreative-workflowslocal-creative-aivoice-cloneprompt-craft

Make an AI Explainer Video Timed to Your Own Voice: Open Explainer's Free Word-Cued Workflow

Watch a cheap explainer video with the sound off and it looks fine. Turn the sound on and something feels wrong within ten seconds. The bullet point appears a beat before the narrator mentions it. The arrow draws while she is still on the previous idea. The picture and the voice are telling the same story at slightly different speeds.

That gap is the real craft of explainer video, and it is the part most tools leave to you and a timeline. Open Explainer, a free toolkit released in public beta on GitHub on November 14, 2026, makes a different bet. It ties every cut, every diagram step and every gesture of an optional cartoon presenter to a spoken phrase in your recording. You never type a timestamp.

I think that is the right unit of editing for this kind of film. The animations are pleasant. The idea underneath them is what makes the toolkit worth an afternoon.

Your voice is the edit

In a normal edit, you place a graphic at 0:14.3 because that is when you said "satellite." Re-record the line a little faster and every graphic after it is now early. You nudge them all.

Open Explainer works from a storyboard file where each scene carries two things: the exact words of narration it covers, and what should appear on which phrase. A scene from the project's own guide looks like this:

{"id": "s02", "type": "bullets",
 "say": "Each satellite repeats a simple message. It says who it is, exactly where it is, and the precise time.",
 "heading": "Every message says",
 "items": [{"text": "Who it is", "at": "who it is"},
           {"text": "Exactly where it is", "at": "exactly where"},
           {"text": "The precise time", "at": "precise time"}]}

Each at is a phrase from the narration. A speech-to-text step works out when you said every word in your recording, and the toolkit pins each item to its phrase. The README puts it plainly: storyboards name phrases, not timestamps, and the toolkit aligns them to the audio.

The practical payoff: as long as the wording stays the same, a new take calls for re-timing, not a new edit. Bring the recording in with oex use-narration, which the user guide says "re-times everything," and your cues ride along with your voice.

What it actually makes

Out of the box you get 1080p MP4 files in 16:9 or 9:16, from about 30 seconds to five minutes and longer, with captions, gentle music ducked under the voice and sound effects. Scene types cover titles, bullet lists, comparisons, flows, big numbers, images and animated diagrams.

The diagrams are the newest piece, added in version 0.3.0. Boxes appear as they are named, connectors draw, little tokens travel along them like a request and its answer, and states switch to done or error on the words that describe them. There is a flat look and a 3D isometric look with 17 built-in illustrated objects, from a robot to a shield to a plane. The project's one-minute showcase, "How AI Agents Work," is on YouTube if you want to see the style before installing anything.

It renders on a laptop processor. No graphics card is needed.

The part that will save you

Here is what I would point a nervous first-timer toward: the checks.

When the final render finishes, Open Explainer measures the finished MP4 for frame count, length, how far the audio and picture drift apart, whether each cue landed on its word, loudness, peaks, black frames and captions. A separate review command flags stretches of text with nothing moving, diagrams that never animate or are crammed too close together, and music that is too loud or too quiet.

It also writes a credits file listing the license of every sound and asset used, with a publish_safe field. For anyone who has had a video flagged for a sound effect they downloaded years ago, that one file might be worth the whole install.

The project is honest about what the checks are for. In its words, neither the report nor the review "replaces watching the film."

Your own cartoon presenter, optionally

You can present the film as a cartoon version of yourself. You make the images in your own image tool, such as ChatGPT, with the project's ready-made prompts, and your photo never touches Open Explainer.

The faster route uses sheets: a master character, then a 3x3 grid of poses, a 3x3 grid of expressions, the same expressions with mouths removed, and a sheet of seven mouth shapes for speech sounds. A command cuts each sheet into separate transparent images, finds the real gaps between drawings instead of assuming a neat grid, and produces a review picture so you can spot problems before you spend a render.

Be ready to look closely here. The guide says a generated sheet "is not animation-ready by itself." The mouth sheet has no labels, so you check every mouth's name by eye, and the "fv" mouth often comes out looking like a smile. You also measure where the mouth sits on the face yourself, in any image viewer that shows pixel positions.

Put this into practice

Plan on an afternoon for your first film. You will want a Mac, the only tested platform (more on that below), plus FFmpeg, a free video tool, and comfort with a few typed commands, or Claude Code to type them for you.

  1. Install and check. Follow the README quick start: install with uv, then run oex doctor to confirm your setup.
  2. Watch the demo render. oex new demo --demo followed by oex -C demo render makes a short film with no recording, so you know what you are aiming for.
  3. Size your script before you record. oex estimate script.txt --target 3:00 tells you if the script fits your target length. Rewriting text is cheaper than re-recording.
  4. Record in a quiet room. The getting-started guide is right that a quiet room matters more than the microphone. Your phone's voice memo app is fine.
  5. Write the storyboard, or let an agent write it. With the skills installed, Claude Code can draft it from your script. Keep cue phrases short and distinctive: "exactly where" beats "it".
  6. Preview, then look at the contact sheet. render --preview and sheet give you a quick render and a page of still frames per scene. Fix wording or cues in the storyboard; only changed scenes render again.
  7. Final render, then read the reports. Check the QA output, run the review, and open the credits file before you publish anything.
  8. Watch it once, start to finish, with sound. The tool asks you to. Do it.

Add the mascot on your second film, not your first. Get the timing right with no presenter, then bring the character in.

Where it breaks

Mac only, for now. It was built and tested on Apple Silicon Macs. It is plain Python and FFmpeg and may work on Linux and Windows, but the project says those are untested, and its automated test setup has not passed yet.

It is brand new. Version 0.3.0, public beta, a single star on GitHub when this was written. Expect bugs and report them.

The free draft voice is not for publishing. The built-in macOS voice is for private drafts only; Apple's license limits its system voices to personal, non-commercial use. Record yourself, or use a voice service whose terms allow publishing. The showcase narration was the maintainer's ElevenLabs voice clone.

Lip sync is cartoon-grade. It follows the sound of your voice, not the exact speech sounds, and lip sync from generated sheets is tested only on a head-and-shoulders presenter.

The visuals have a fixed vocabulary. Diagrams are built from boxes, pills, circles, diamonds, connectors and labels plus the 17 objects. Other pictures appear whole, without step-by-step animation. If your topic needs a custom animated illustration, this will not draw it.

Agent support is uneven. End-to-end runs were done with Claude Code. Codex can find the skills, but a complete Codex run has not been verified.

Timing is the craft

Most AI video tools compete on how good a single frame looks. Open Explainer competes on when each thing appears, and for teaching that matters more. A plain box that shows up exactly when you say its name teaches better than a beautiful one that arrives late.

Take the next idea you keep explaining out loud to people, a process at work, a tool you love, a concept your students trip on. Write it as a three-minute script, record it, and let your own voice cut the film. Then watch it with the sound on and see if it finally feels right.


From the briefing