Image understanding means adding a picture to the model's context

A multimodal model can process text together with an image and use both when generating a reply. Google's developer documentation, for example, describes passing images by public URL, inline data, or a file-upload API. Those are API options for developers; they should not be mistaken for a feature available in every consumer companion app.

For roleplay, a user might share a landscape as a scene reference, ask what objects are visible, or use a sketch to establish a character's look. The model's answer is still an interpretation of the pixels it received, not a verified record of what happened or who appears in the image.

Sources

Ask about visible details, not hidden intentions

A useful image prompt narrows the task: describe the room's visible objects, identify the colors in an outfit, or help turn a landscape into a fictional scene. If a detail matters, point to where it appears and verify the answer yourself. Image crops, resolution, lighting, and model choice can affect the output.

Do not ask an image model to determine a person's identity, private traits, or emotional state from appearance. A confident sentence can still be a guess. For story continuity, state which details are fictional canon and which are merely visual suggestions instead of assuming the model will preserve every observation.

Sources

Image privacy depends on the app, not just the AI model

A model API's ability to accept an image does not tell you whether a companion product stores uploads, includes them in chat history, shares them with a provider, or uses them under a particular account setting. Read the actual service's current privacy notice and upload controls. If those details are unclear, do not upload a sensitive image just to test a feature.

Before sharing, remove unnecessary personal documents, location details, faces, and other identifying information. Use images you own or have permission to share, and remember that screenshots may contain names or notifications outside the main subject. A deletion button is useful only if the service explains what it deletes and where copies may remain.

A practical checklist

  • Which model or provider receives the image?
  • Is the image retained with the conversation or for another purpose?
  • Can you delete the image and related chat data?
  • Does the app disclose when image understanding is active?
  • Can the scene be described in text instead of uploading a private image?

Visual context can enrich a scene without replacing character setup

A picture may help establish what a character is reacting to, but the roleplay still benefits from a clear prompt about setting, point of view, and who controls each character. Treat the image as one reference, then correct any details the model misreads. This keeps the exchange collaborative rather than letting the model silently turn uncertain observations into story facts.

Image input is one possible multimodal direction for companion software, not a feature claim about Jess. The linked Jess overview explains the current product experience; the broader privacy guide can help you review data handling before trying any companion service.

Sources and further reading

Research and product details were reviewed September 30, 2026 where applicable. Vendor policies, plans, and features can change; verify current details with the linked first-party source.