When on trips, users often take near duplicate photos at the same place. Multiple photos with little variation among them can consume a large amount of the limited storage space on the user device and/or cloud storage services where users may choose to store their photos. Retaining only a few of the photos can reduce storage use but makes the photos permanently unavailable. This disclosure describes techniques that enable users to recreate close approximations of discarded photos from corresponding text descriptions. A suitable image-to-text conversion model is applied to photos prior to discarding the photos and the generated text description is stored. The text description, context, and any related photos are used to prompt a suitable image generation Artificial Intelligence (AI) model to recreate the photo. The techniques can additionally support generating variations of photos that did not originally exist. The accuracy, faithfulness, and fidelity of the generated photo can be improved with prompts that contain more detail as well as context from other related photos.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.