I recently experimented with five different outfits in a remarkably quick session—thanks to Doppl, an innovative app currently being tested by Google. This platform allows users to generate AI-enhanced videos of themselves dressed in various outfits sourced from the internet. While the technology shows promise, it encounters some notable challenges, such as mishandling the application of pants over shorts and occasionally rendering fictional feet in place of actual socks.
The app’s functionality is quite straightforward. Users simply need to upload a screenshot of the desired outfit from platforms like Pinterest or Instagram alongside a full-body image taken in good lighting, depicting them in a natural stance and without headwear. After both images are submitted, Doppl generates a still image of the user dressed in the chosen outfit. Though the process does take a moment, users can enhance their generated images with an animation feature that presents various fun poses, such as peace signs or friendly waves.
I selected a simple photo of myself in a T-shirt, shorts, and socks to kick off my outfit trials. My first attempt utilized one of Google’s example outfits. While the app accurately rendered the white and blue striped shirt, it bizarrely dressed me in red shorts instead of jeans, wrapping the supposed pants around my calves as if I were sporting oversized leg warmers.
In another instance, I opted for a screenshot of distressed jeans. Again, Doppl presented only the button portion of the pants while extending my shirt to an unusual length, cutting off at the spot where my real-life shorts ended. The oddity continued when I tried uploading an outfit featuring a person up to the knees wearing a striped button-down shirt and long shorts. Instead of creating a parallel outfit, the AI truncated the shorts and provided a pair of somewhat realistic looking artificial feet. Interestingly, even when shoes weren’t depicted in some uploaded outfits, Doppl generated random footwear, presumably based on what it believed would complement the look.
I discovered during testing that Doppl imposes restrictions on the types of images you can upload. Pictures depicting revealing attire, such as bikinis, cannot be processed. My colleague, Marina Galperina, ran into similar barriers when attempting to upload an image featuring former President Donald Trump. These limitations could potentially deter misuse of the tool for creating deceptive images involving public figures or inappropriate content.
Interestingly, when Marina and I uploaded mirror selfies, our digital counterparts were rendered significantly slimmer, resembling caricatures. This distortion did not occur with traditional full-body photos, suggesting a peculiar quirk in mirror selfies.
Although Google previously introduced a virtual try-on feature, it has taken a significant step forward this year by allowing users to upload personal photos for AI-generated fashion simulations. Doppl enhances this experience, enabling users to experiment with various garments from a wide range of sources, and even creating animated video outputs. With improvements addressing some current limitations, this app has potential as an engaging tool for visualizing outfits before making purchases.
