TMochiLearn

Create a complete interactive educational video from one lesson brief.

TMochiLearn is a one-shot creator for educational, technical, and training content. A creator supplies one lesson brief, selects the generation models and choice depth, and TMochiLearn builds the narrative, every learning path, the rendered media, and an interactive player in one workflow.

Learners watch a continuous lesson and choose what to explore at each decision point. A lesson can contain one to three binary choice levels, producing two, four, or eight complete learning paths.

Watch the app demo

What one generation creates

• A lesson narrative expanded into a complete binary learning tree
• A deduplicated catalog of shared and path-specific scene layers
• Rendered video, narration, music, and supporting audio for every route
• A live learning-path map with scene and full-path previews
• A resumable Creator Studio session at /creator/[sessionId]
• A downloadable ZIP containing the interactive manifest and generated artifacts
• An interactive publication that can be shared at /watch/[publicationId]

Published lessons appear in the public learning library at /learn.

Creator workflow

1. Sign in or register with a Samsar account. TMochiLearn creates a non-billable draft session immediately.
2. Enter a lesson brief of up to 4,000 characters and choose a target duration from 30 to 180 seconds.
3. Select the inference, image, and motion models, then choose one to three levels of learner decisions.
4. Submit once. The app creates the lesson narrative, generates every route, and renders the unique media layers.
5. Follow live progress, inspect the learning-path tree, and preview individual scenes or complete routes.
6. Download the artifact package or add publication details and publish the finished lesson.

The default form uses a 30-second lesson and two choice levels, which creates four final paths. TMochiLearn uses a 16:9 interactive-video layout.

Interactive generation model matrix

TMochiLearn does not use the account-level default model list. It loads the current Express catalog from GET /video/supported_models and only displays models advertised as compatible with interactive generation. The generation route fetches the catalog again and validates every selection before submitting the request.

Stage	Model	Request value
Inference	gpt-5.6-sol	gpt-5.6-sol with effort: high (default) or effort: xhigh
Image	GPT Image 2	GPTIMAGE2
Image	Nano Banana Pro	NANOBANANAPRO
Video	Nvidia Cosmos 3	COSMOS3SUPERI2V
Video	Veo 3.1	VEO3.1I2V
Video	Veo 3.1 Fast	VEO3.1I2VFAST
Video	Seedance 2.0	SEEDANCE2.0I2V

TMochiLearn always submits the selected inference model explicitly and ignores the user's account default. For direct API clients, omitting the inference model from the unified interactive request uses gpt-5.6-sol with effort: high. Use effort: xhigh for deeper technical analysis; legacy suffixed Sol model keys remain compatible.

The table is the supported matrix, not a guarantee that every model is available in every deployment. The runtime catalog is the source of truth.

Render pricing

Production render credits use one rule:

render credits = total unique rendered layer-seconds × video model rate

Video model	Rate
Nvidia Cosmos 3	20 credits/second
Veo 3.1	60 credits/second
Veo 3.1 Fast	36 credits/second
Seedance 2.0	40 credits/second

The total includes every distinct scene layer created for the interactive lesson. Shared scenes count once, and each choice-specific scene counts once. The selected rate covers the complete render pipeline; inference and image model selections do not add separate render charges.

For example, 150 unique rendered layer-seconds with Nvidia Cosmos 3 costs 150 × 20 = 3,000 credits.

Before generation, TMochiLearn may show a conservative maximum because it does not yet know which scenes will be shared. The final charge uses the actual unique rendered duration and can be lower. Pipeline stages are settled individually, so fractional seconds may produce a small rounding difference. Only completed stages are charged, which can reduce the final charge if rendering stops early.

Production and standalone deployments

Behavior	Production	Standalone
Model catalog	Full supported interactive catalog	Only interactive-compatible models available through configured providers
Samsar credit admission	Requires an available credit balance	Bypassed
Generation credit settlement	Stages debit Samsar credits	Zero Samsar credits; stage receipts are waived
Provider cost	Covered by the Samsar credit rate	Paid directly by the deployment operator under each provider's pricing
Creator availability	Requires at least one model for every pipeline stage	Same requirement; fails closed when a stage has no configured model

In standalone, provider credentials and adapters belong to the Samsar processor, not this frontend. TMochiLearn disables submission unless the runtime catalog contains at least one compatible inference model, one compatible image model, and one compatible video model. The production credit rates above are not standalone provider-price quotes. If the Creator displays the production comparison estimate in standalone, that number is informational and is not deducted.

Run locally

Requirements:

• Node.js >=22.13.0
• npm
• A reachable Samsar API deployment
• A Samsar account/session for Creator Studio

cp .env.example .env.local
npm ci
npm run dev

Open http://localhost:3000. The default configuration connects to the public Samsar production API.

Environment variables

Variable	Required	Default	Purpose
SAMSAR_API_BASE_URL	No	https://api.samsar.one/v1	Samsar API base URL, including the /v1 path
SAMSAR_ARTIFACT_HOSTS	No	Empty	Comma-separated HTTPS hostnames added to the artifact-download proxy allowlist

Provider API keys are configured on the Samsar processor deployment and must not be added to the TMochiLearn frontend environment.

Development commands

Command	Purpose
npm run dev	Start the Next.js development server
npm run dev:production-catalog	Run locally against the public production catalog
npm run build	Create the Next.js production build
npm start	Serve the Next.js production build
npm run lint	Run ESLint
npm test	Build the Cloudflare Worker target and run the Node integration suite
npm run dev:worker	Start the local vinext/Cloudflare Worker target
npm run build:worker	Build the Cloudflare Worker target

Deployment

The Next.js build uses standalone output and includes server rendering, route handlers, authentication, and request-derived social metadata. Vercel can use the Next.js framework preset with npm run build.

Build and run the production container with:

docker build -t tmochi-learn:latest .
docker run --rm -p 3001:3000 \
  -e SAMSAR_API_BASE_URL=https://api.samsar.one/v1 \
  tmochi-learn:latest

The container runs as an unprivileged user, listens on port 3000, and exposes GET /api/health. When deploying behind nginx, use nginx.conf.example and forward Host, X-Forwarded-Host, and X-Forwarded-Proto so canonical and social URLs are generated correctly.

Cloudflare Worker builds are available through npm run build:worker and npm run start:worker.

License

TMochiLearn is available under the MIT License.