Last updated 13 October 2026
GIFs and stickers are entirely local. When you drop in a photo to make a GIF or a sticker, the image is processed by your own browser. The AI models run on your device. The photo is not sent to us, we cannot see it, and no account is required.
3D depth is local too. The depth model runs in your browser on every account. The next section covers the one fallback where part of it runs on our server.
Publishing uploads. If you choose to publish a 3D scene so it has a shareable link and an embed, the finished files are uploaded to our storage, and this is the one point at which your photo itself is sent to us. In the depth fallback described next, intermediate data is sent earlier, but never the image.
Making a 3D scene needs a depth map. The model that produces it comes in two halves, and your browser downloads both and runs them on your device, like GIFs and stickers. No image data and no intermediate data reaches us, and once the model is cached it works offline. We record that a generation happened, for usage numbers.
The fallback. If your browser can't load or run the second half, it runs the first half on your photo and sends the result to our server, which finishes the depth map and sends it back. That result is a set of intermediate numbers (machine-learning “activations”), not an image, and your photo is never sent. We process them in memory and do not store or log them.
We should be straightforward about one caveat: intermediate activations are derived from your photo. They are not a picture and we do not attempt to reconstruct anything from them, but published research shows that approximate reconstruction from data of this kind is possible in principle. The fallback only happens when the local model can't run.
Published scenes are public. That is the point of them — they are made to be shared and embedded on other websites. Their files are served from public URLs, so anyone with the link, or with the page the scene is embedded in, can view them. They may also be cached by browsers and content networks for up to a year. Please do not publish a photo you would not put on a public web page.
At the moment you publish, your image and its subject mask are sent to our server once, so a model can paint in a clean background behind the subject. That data is processed in memory to produce the result and is not stored or logged by us. Together with the depth step described above and the upload of the finished scene, this is the complete list of times anything derived from your image leaves your device when you make a 3D photo. 3D models are covered next.
3D models are the one tool that needs a server: a GPU builds each model. From a photo, the background is removed in your browser first and only the resulting cut-out is sent; it is deleted once the model is made or the job fails. From text, your description is sent so the GPU can draw a reference image, and the image you approve is cut out in your browser the same way. Finished models, their thumbnails and reference images are stored privately in your account and shown only to you.
If you turn on a model's embed, a compressed copy of it and its thumbnail are served to anyone who views a page carrying the iframe. Embedded models aren't listed or indexed, and turning the embed off stops serving them within a few minutes.
You do not need an account for GIFs, stickers or 3D scenes. You need a free account to publish. If you sign up, we store your email address, your plan, and a count of the 3D scenes you have generated. We do not store the images those generations were made from. Authentication is handled by Supabase.
Payments are handled by Dodo Payments, which acts as the merchant of record. Your card details go to them, never to us — we never see or store them. What we keep is a record of each payment and what it bought (3D model credits, or the Pro plan, which we no longer sell), which is how we know what to give you.
We use two analytics services. Vercel Analytics counts page views and a few product events in aggregate. PostHog (hosted in the EU) counts how the site is used, so we can see where it gets in people's way: pages visited and steps such as adding a photo, making a GIF or a preview, building or downloading a model.
On your first visit we ask. If you accept, PostHog keeps an identifier in your browser, links your activity to your account once you sign in and keeps short replays of the page, which leave out what you type and every image. If you decline, or until you choose, visits are counted anonymously: nothing is stored in your browser and nothing is recorded. You can change your choice any time in Account.
For signed-in accounts our servers also log the main steps of the 3D service against your account, as part of running it: previews and models made, finished or failed, downloads, credits running out, checkout and plan changes. These record what happened, never your prompts or images. Your photos are never sent to analytics; they stay in your browser as they always have. None of this is used for advertising or to track you across other websites.
Published scenes stay up until you or we remove them. Because their files are served with long cache lifetimes, copies may persist in caches for a period after deletion.
Self-serve deletion of a published scene is not available yet. To delete a published scene, or your account and everything attached to it, email us at aman@metamemory.tech from the address on your account, and we will action it.
Gifsy is not directed at children under 13, and we do not knowingly collect their personal information.
If this policy changes in a way that affects you, we will update the date at the top of this page. Continuing to use Gifsy after a change means you accept the updated policy.
Questions, or a deletion request: aman@metamemory.tech. See also our Terms and Refund Policy.