# Education and language learning

Match teaching materials, bilingual voice and playback costs to the way learners use your product.

## Identify who uses the output

A teacher preparing classroom illustrations and a developer adding voice playback to a language-learning product have different workflows. Define the learning material, the learner's action and the review standard before selecting a model.

For illustrations, record the concept being taught and the details that must remain correct. For speech, record languages, pronunciation requirements, text length and how often each passage is played.

## Prepare teaching visuals

Use [image model selection](https://seedrouter.ai/image-generation) to find suitable reference and editing capabilities. Keep the approved instructional content separate from visual styling. Review diagrams, written labels and depicted objects for accuracy; a polished image is not evidence that the teaching content is correct.

The [reference prompting guide](https://seedrouter.ai/blog/nano-banana-pro-prompt-guide) can help structure visual instructions. Your course authoring tool remains responsible for lesson structure and publication.

## Design future voice playback

Bilingual learning can require switching languages within one explanation while preserving pronunciation and continuity. Compare generating audio on demand with keeping reusable audio for frequently played passages. Cache validity should follow the text and voice settings, not only a lesson identifier.

Those are requirements for future TTS integration, not a claim that SeedRouter currently supplies speech APIs. Long-form reading also requires chapter boundaries, pronunciation review and resumable playback in the application.

## Evaluate the complete learning interaction

Review visual clarity, factual correctness and whether the learner can use the result at the intended device size. For future audio, include rare terms, mixed-language sentences and repeated playback in the evaluation. Your product controls learner accounts, content access and playback behavior; model access is one component of that system.

## Related resources

- https://seedrouter.ai/blog/nano-banana-pro-prompt-guide