Complex spoken explainer
A complete Chinese explainer combining 17 accepted performances, website demonstrations and eight project-owned animated scenes. Its 137-second Timeline includes an opening montage without speech, full-frame and circular presenter views, a half-frame handoff, covered narration, and a silent ending.
The source is an editable production, not a general-purpose explainer template. It demonstrates how to give each visual relationship an owner while exposing only the parameters the author actually needs.
Watch the finished film (137 seconds, approximately 42 MiB), or download the complete media archive (approximately 455 MiB) to open the editable production below.
Open the production
Use the Hypit checkout accompanying this example. The example has been validated against that checkout; it does not claim compatibility with an older published Distribution. Install the project-local packages from this directory:
npm ci --ignore-scripts
The code is stored in Git. Images, recordings, accepted Result resources and a finished video are supplied separately in the media archive. Download and extract the archive at this directory:
mkdir -p .hypit
curl --fail --location 'https://storage.googleapis.com/hypit-public-assets/assets/examples/complex-explainer/v1/20260914/media.tar.gz' --output .hypit/media.tar.gz
tar -xzf .hypit/media.tar.gz
The archive contains source assets, historical Result files retained for inspection, and the finished
film. The editable production uses project-owned assets rather than historical Build addresses;
private review notes, credentials and the original reference creator's footage are excluded. Media is
hosted on Google Cloud Storage and stays outside Git.
After extraction, .hypit/media.tar.gz can be removed to reclaim the archive's download space.
cd productions/explainer
hypit check runs/render.svrun
hypit plan runs/render.svrun --runtime ../../hypit.runtime.json
hypit studio --invocation runs/render.svrun --runtime ../../hypit.runtime.json
Open the printed Comments URL (#comments) to review the composition before export and leave
timestamped direction on the current picture. In the same session, switch to Studio to explore
the timeline and adjust exposed parameters. Browser review does not render a video file. Notes save to
this workspace's FEEDBACK.json; the Agent can read and resolve them while editing the owning Source
or scene code. Existing private review conversations are not included. The Inspector remains useful
for the selected Caption, Style and event choices without exposing every internal scene dimension.
The default saved Invocation is target-only. Accepted performance and supporting media are ordinary project assets; media preparation and speech alignment remain explicit Author relations. The included Profile has no generation account or credential store. A Build that reaches the authored WhisperX declarations therefore needs an explicitly configured transcription path, or the project must replace those declarations with project-owned transcript evidence first.
Submit runs/render.svrun to render the complete film. runs/export-parts.svrun instead renders five
consecutive ranges for limited temporary disk space. Its local Profile requests eight rendering
workers and one render request at a time. The production guide
explains joining exported parts. The archive also includes productions/explainer/output/final.mp4.
Read the work by responsibility
| Question | Owner |
|---|---|
| What is said, and which words motivate a change? | Script |
| Which raw resources, clock and canvas are used? | Assets |
| What directed the accepted performance? | Direction, performance Recipe |
| How are media occurrences, Uses, events and contributions assembled? | Main Source |
| How does the Caption look? | Composition Recipe |
| How does each coordinated scene behave? | Web scenes |
| How do the presenter and opening graphics move? | Opening system |
| How are cue boundaries and one-line captions handled? | Caption package |
| Where does the shared palette live without becoming another Track? | Visual language |
| Which existing material should a new Build use? | Assets and Main Source |
Generated and prepared-media Outputs remain in the entry Source, while accepted files are normal Author resources. Changing a Cue boundary or graphic event does not require replacing the performance. A rewrite of the spoken words is different: provide new performance media and matching evidence for that Script. The retained generation declarations show the original direction; configure an explicitly chosen Model/Provider path before requesting new material.
What this example teaches
A coordinated editor scene can own its folder, pointer, preview and timeline without publishing a control for every dimension. The presenter, Caption and countdown change for different reasons and remain independent. Shared palette and typing helpers are ordinary code dependencies, not extra Tracks. The scene's state is computed from frame time, making direct seeking and range rendering possible. Script Moments locate spoken changes; authored durations shape how those changes unfold.
Read the Treatment and craft notes for the practical reasoning: recovering behavior beyond a list of objects, the style change, scene boundaries, semantic events, framing and review. The presenter-led visual explainer playbook draws out the whole-work relationship for other productions. Components use public Author Package and Companion interfaces; Studio and Core contain no dispatch for this example.
Materials and editorial copy
The host image and voice are AI-generated. The competitor-case demonstrations were made by the project owner. The phone's drinking callback uses a newly generated fictional illustration, not the reference creator's footage. Website captures and product marks retain their sources; the diagrams, repeated marks and token counters are authored illustrations rather than measured product tests. See asset provenance.
This is the owner's authored explainer, including its comparisons and rhetorical claims. The reusable lesson is how the production is expressed, not a benchmark of other products.