Skip to content

Image generation and vision

Updated May 20, 2026

The AI Superpower handles three kinds of AI work: text generation/analysis, image generation, and vision input. Text is the default. The other two are opt-in.


Image generation

Pick an image model in the AI Superpower section of the Add App modal. The app can then generate images from text prompts on demand. Students type a description, an illustration appears.


Two providers support image generation:

  1. OpenAI, DALL-E. Always available if you have an OpenAI key connected.
  2. Google Gemini, Imagen. Paid tier only. The free tier does NOT include image generation.


Anthropic Claude does not generate images. You'll see image-model options in the dropdown only for providers that support it.

Each generated image costs the creator (in Use My AI Key mode) or the guest (in Use Their AI Key mode), the same way a text call does. Keep your daily cap in mind.


Vision input

Your app can also send images to the AI for analysis. The AI can read text in a photo, classify it, describe it, or work with whatever's in the picture.


Use cases:

  1. Handwriting OCR ("read this handwritten note")
  2. Reading-level analysis of a photographed passage
  3. Photo classification ("what kind of leaf is this?")
  4. Math equation extraction from a whiteboard photo
  5. Description for accessibility ("describe this image in detail")


Vision works with OpenAI, Gemini (free and paid tiers), and Anthropic Claude. Pass up to two images per AI call.


When to use which

  1. Generate images when the app's job is to make pictures (story illustrations, concept art, decorative output).
  2. Vision input when the app's job is to understand pictures (OCR, classification, accessibility).
  3. Both are fine in the same app. A photo-tutoring app might receive a photo of student work AND generate an illustrated explanation back.


Privacy with photos

Photos are sent to the AI provider over an encrypted connection and not stored by Teacher Hive. The provider's policies apply. If guests upload photos of student work, the no-PII rule still applies. Cover student names and identifying marks before they hit the AI.

Back to Help