Shpritz AI Review

We spend a lot of these reviews proving a negative. A platform says it has no upload feature, and we go into its JavaScript to check, because saying so costs nothing and building it that way costs something.

Shpritz AI is the first one in weeks where both halves lined up on the first pass. Its policy says users cannot upload content. Its code says the same. That is worth more than either statement alone.

What is Shpritz AI?

Shpritz AI is an AI companion platform with an unusually complete generation toolkit attached. You chat with characters, create your own, and generate images, short videos and voice notes. Beyond that it offers three things most competitors in this bracket do not: a prompt-driven image editor, a 4K upscaler, and an image-to-video route that turns a still you already made into a clip.

The domain to use is www.shpritz.ai. There is a decoy: shpritzai.com is a parked domain that bounces you to an unrelated affiliate comparison site, carrying a campaign parameter that reads "parkedomains" in plain text. We found the same squatting pattern on the .com variants of Nemora and SecretCrush. At this point it is a category-wide phenomenon, and the rule is simple: for these brands, the .ai is the product and the .com is somebody's advertising.

The operator is Social Sync Ltd, of Limassol 46E, Strovolos, Nicosia, Cyprus. That is the sixth Cypriot company we have documented in this run of reviews. We have stopped treating it as a coincidence and started treating it as the industry's default incorporation.

The policy question, confirmed twice

Shpritz has an edit-image feature, an upscaler and an image-to-video pipeline. On a platform that accepted uploads, that combination would be an undress tool with a friendlier name, and this review would end here. So we checked it properly.

The platform answers first, in its own Blocked Content Policy: "Although users cannot upload or share independent content, the Services allow private conversations with AI Companions and access to AI-generated media."

Then the code answers. We pulled all 42 of its JavaScript bundles, about 3 MB, and traced the editing pipeline. Every one of those three features is keyed on a source image ID, not a file: setEditImageSourceImageId, setSourceImageId for the upscaler, and a source_image_id column that the client resolves by querying the platform's own media table. Turning an image into a video navigates to /create/video?imageId= with an ID from your existing library.

There is no file input in the product, no image-accepting field, no presigned upload URL and no application upload endpoint. The handful of multipart references belong to the HTTP library, not to the app. There is no image-to-image, no reference image, no identity adapter and no face swap. The 37 hits for "mask" that first caught our attention are all CSS mask properties, SVG mask elements and a process.umask shim, with no inpainting mask anywhere. We also checked for user-trainable models, since the video engine supports LoRAs, and found none: LoRAs here are picked from a platform list, not trained by you on your own pictures.

So the editor operates on a closed library. It can only touch pictures the platform itself made. That is the same reasoning that cleared CreatePorn's inpainting beta, and it is the strongest form of the argument, because it does not depend on anyone keeping a promise.

On the written rules, one clause deserves real credit. The community guidelines list, among prohibited actions, "Create, attempt to create, publish, or share images resembling real persons". No deceptive-or-harmful qualifier, no consent carve-out, no "without authorization". That is the unconditional version we have been asking platforms for since SpicyMate.

The blocked content policy, awkwardly, does not match it. Its impersonation clause forbids "impersonating real individuals, public figures, or celebrities in a deceptive or harmful manner", which is the weak formulation we criticised at SpicyMate, FLIRTcam and Cherrypop. Two documents from the same company on the same question, one strict and one loose. The architecture makes the point mostly academic, but a company should not have to be saved from its own drafting by its engineering.

The features that actually matter

Prompt-based image editing. Pick a generated image, describe the change, get a revision. This is the single most requested capability in generator communities, because the standard frustration is an image that is nearly right.

A 4K upscaler. Separate from the editor, applied to your own results. Rare at this level.

Image to video. Any still in your library can become a clip. The video engine is a WAN model with selectable LoRAs, which is more technical detail than most platforms in this category will admit to.

Voice. Custom voice notes from your companion, on a dedicated route.

Ten legal documents. Terms, privacy notice, cookies notice, underage policy, content removal policy, blocked content policy, DMCA policy, complaint policy, 2257 exemption and community guidelines. Whatever else is true here, somebody did the compliance work.

Pricing, without the bullshit

Nothing, again.

Shpritz publishes no price anywhere we could reach. The upgrade page and the tokens page both render behind the application, and there is no usable third-party source to cross-check against, because almost nothing independent has been written about this platform.

One directory describes a pay-as-you-go structure: you buy token bundles and spend them per generation, larger bundles cost less per credit, and subscription tiers exist only as a plan rather than a product. If that is accurate it would actually be the friendlier model, since you are not renting an allowance you might not use. But it is a single source of the kind we have caught inventing features before, so we are not putting a number in this review and neither should anyone quoting it.

What we can say without a price: this is a token economy, so the meter is the cost. And the toolkit here is exactly the sort that burns credits fastest, because upscaling and video are the expensive operations everywhere else we have measured them.

What users say

Almost nothing verifiable. No Trustpilot profile, no meaningful forum discussion, and the pages ranking for the brand are AI-tool directories and affiliate listings rather than user reports. A domain reputation scanner rates the site as probably legitimate, which is close to the weakest possible endorsement and we mention it only because it is the sole external check that exists.

Absent, and worth stating: no breach, no scam allegations, no billing complaints. Also absent: any evidence of a user base. Treat the silence as missing information rather than a clean record.

One gap we will name, because we praised another platform for closing it yesterday. Shpritz has an underage policy and a 2257 exemption statement, but no age verification beyond the usual self-declaration. Hallufox runs a camera-based identity check. Doing the paperwork is not the same as doing the check.

Final verdict: who is Shpritz AI for?

Get Shpritz if: you want a generator with real post-processing, an editor and an upscaler rather than a one-shot prompt box, and you want that on a platform whose no-upload boundary is verifiable rather than merely claimed. On architecture it is one of the safest we have reviewed.

Skip it if: you refuse to sign up before seeing a price, which is entirely reasonable and which Nemora lets you avoid, or if you want a platform with a track record. This one has none.

The formula we would pick: register, go straight to the tokens page before generating anything, and price a video and an upscale specifically. Those two operations will decide what this actually costs you, and they are the two the marketing will not put a number on.

Rating: 7.3/10. Safer than it needs to be, quieter than it should be about the bill.