Smart Share allows the AI to autonomously share photos with Users during the conversation.
The Persona can select and send images on its own, making interactions more dynamic and engaging without manual prompts.
FEATURE ACTIVATION
To use this feature, you first need to activate it.
This is done in your project settings.
PROVIDING THE LIST OF SHAREABLE PHOTOS
Once the feature is enabled, you need to provide the list of photos that can be shared.
For this, with each reply request, you must send an array with the URLs of the shareable photos, using the replyParameters.shareablePictures parameter.
To prevent the AI from sharing the same photo multiple times with a User, you must keep an up-to-date list of photos that can still be shared with them.We recommend you track, in the Answer object, the presence of the pictureUrl property, which contains the URL of the photo just shared by the AI.You should then remove the corresponding photo from the list of shareable photos for this conversation. answer.created event documentation.
OPTIMIZATION WITH PHOTO DESCRIPTIONS
This is optional, but we strongly recommend providing our model with a description for each shareable photo, so the AI can reply better and understand what it shares.
We can handle this automatically for you.
To do so, simply set replyParameters.vision.shareablePicturesAnalysis to true and do not provide any descriptions.
The standard fee of €0.015 per unique photo URL applies, full details here.
You can also provide your own descriptions by filling replyParameters.shareablePictures.caption for each photo in the array.
Finally, you can choose not to provide descriptions and not to let us analyze the photos.
To do this, set replyParameters.vision.shareablePicturesAnalysis to false.
The AI will then select a photo at random from those available.
This method works BUT the AI has no understanding of what it shares. This can lead to misunderstandings or mistakes that wouldn’t otherwise occur.
LIMITATIONS OF THE CURRENT VERSION
In its current version, autonomous photo sharing with our photo analysis model works very well.
The AI accurately understands what it shares and only shares content that fits the conversation context (for example, no explicit photos in a serious discussion).
It also avoids sharing too many photos, even if the User insists.
The only limitation for now is that it works reactively. It does not share anything unless the User asks for it.