CubricStudio
Privacy policy

Your work stays on your computer.

Last updated 8 November 2026

Cubric Studio has no account, no sign-up, no analytics and no automatic crash reporting. It uploads only these things to us, and only when you ask it to: an anonymous benchmark result, if you tick a box to share it, a reference video you attach to a cloud video generation, which we delete within the hour, a bug report with the app's log, when you press Send report, and, if you join our beta testers, your computer's hardware details (never your files or names). This page lists every time the app or this website talks to someone else, what goes with it, and why.

Who we are

Cubric Studio and cubric.studio are made by Fabio André Rosa Gonçalves, trading as Mad Pony Interactive, in the United Kingdom. Under UK data protection law (the UK GDPR and the Data Protection Act 2018) we are responsible for the small amount of personal data described in "If you support us or get in touch". Contact us at contact@madponyinteractive.com.

The short version

What stays on your computer

To remove everything, delete your Cubric Studio folder and your project folders.

When the app goes online

Your licence key

Cubric Studio 2.2 and later needs a licence key from your Gumroad purchase. When you activate it, the app sends the key and a hashed computer ID to licence.cubric.studio, our licence server, hosted by Cloudflare. The computer ID is your operating system's own machine ID, hashed with SHA-256 so it cannot be turned back into the original; it is how one key counts up to 3 computers. Our server asks Gumroad whether the key is a real, unrefunded purchase. After that the app checks in about once a day while you are online, and works offline in between.

Our server keeps the key only as a SHA-256 hash, the hashed computer ID, and when that computer last checked in. It keeps no IP address or email, and no name except the Discord name of a beta tester or someone applying to be one (see "Beta testers"). "Sign out of this computer" in the app frees that computer's place straight away; a computer that has not checked in for 30 days frees its place on its own.

Downloads and updates

Installing the engine, downloading models and updating the app connects to: models.cubric.studio (our file store, hosted by Cloudflare), Hugging Face, GitHub, the Comfy node registry, Python package servers, and ollama.com if you install Ollama. The app asks our licence server for the newest version and downloads updates from it, using the licence token from your key, and it may check that Hugging Face and RunPod are reachable. These services see your IP address and which file you asked for, as with any download. Apart from the licence token on our own update server, the app adds no identifier of its own.

Beta testers (only if you join)

Some people on our Discord help test new versions before everyone else. If you are invited, you apply on a private page of licence.cubric.studio with your licence key and your Discord name. Our server checks the key with Gumroad and keeps your application (the key only as a hash, plus your Discord name) until the developer approves it, which marks your key as a beta tester, or turns it down, which deletes it. The app itself sends a Discord name only with a bug report, and only if you type one into it. While "Beta updates" is on in Settings, the app gets new versions early from licence.cubric.studio, and once each time it starts and checks for updates it sends that server your computer's hardware details: your graphics card and its maker, video memory and system memory (your Pod's, when you generate on RunPod), processor, operating system version, app version, and whether you generate on this computer or on RunPod. The server also records when they were last sent. It never sends your files, projects, prompts, keys or computer account name. Your bug reports carry the same details and say you are a beta tester, in the private staff channel only; the public copy of a report never does. We use them to see which computers a new version was tested on. They are kept until you leave the beta (ask on Discord or by email, or the developer removes you); your Discord name and details are then deleted from our server straight away. Turning "Beta updates" off stops sending them and offers you the stable version back.

Cloud generation (optional, your DeepInfra key)

When you generate with a cloud model, your prompt, settings and any reference images, audio or videos you added go to DeepInfra. DeepInfra may pass the request to the company that made the model (for example Google, ByteDance or Black Forest Labs) to run it. The result comes back into your project. The app also reads your DeepInfra balance with your key so it can show what a generation costs. DeepInfra's privacy policy applies.

Reference videos for cloud video (optional, your DeepInfra key)

Some cloud video models, such as Seedance 2.0, only take a reference video as a web link, not as a file. When you add a video to one of these generations, the app uploads that clip to our own service at relay.cubric.studio, which runs on Cloudflare, and sends DeepInfra the link instead of the file. DeepInfra, and the company that made the model, fetch the clip from that link to make your video. The link is a long random address that is never listed or published. The app deletes the clip as soon as the generation ends, whether it worked or not. If the app cannot (for example, it closes mid-run), our service stops answering the link after 55 minutes and deletes the clip within the hour. Clips can be at most 30 seconds long. We do not look at, copy or keep them, and nothing records who uploaded one. Like any web service, the relay sees your IP address. We use it only to limit uploads to 10 a minute, and it is never written down. Reference images and audio do not come to us: they go straight to DeepInfra with your request, as described above. Cloudflare's privacy policy applies to the hosting.

Dictation (optional, your DeepInfra key)

When you press the microphone button and speak, the recording of your voice goes to DeepInfra, which runs a speech model (Whisper) to write it out as text. The text comes back into the box you were typing in. The app does not save the recording, on your computer or anywhere else. DeepInfra's privacy policy applies.

Remote GPU (optional, your RunPod key)

The app creates GPU machines ("Pods") and storage on your own RunPod account. Your prompts, input images and videos go to your Pod, and results come back. Anything on a Pod or storage volume stays there until you delete it, from the app or from RunPod's console. RunPod's privacy policy applies.

The assistant and prompt tools

With Ollama, these run on your own computer and nothing leaves it (unless you choose one of Ollama's cloud models, which Ollama runs on its own servers). If you connect a remote provider instead, such as DeepInfra, OpenRouter, OpenAI or an address you enter, the text you send and any images you ask it to look at go to that provider, under your key and their policy.

The same model runs a child-safety check. Most prompts are checked on your computer alone, by a word list. When a prompt names a child together with swimwear, or is written in an alphabet the word list cannot read (Chinese or Russian, for example), the app asks this model to check it before it generates. With a remote provider, that prompt goes to them too.

When a generation sends a picture and its words ask for nudity, underwear or sexual content, the app also asks the image describer you picked whether anyone in the picture could be under 18. With a remote describer, that picture goes to them too.

When the assistant looks at a video that has sound, it also listens to it, through the speech to text you picked for dictation. With DeepInfra, the video's sound goes to them in pieces of up to a minute, under your key and their policy. With Ollama or the app's own engine, it stays on your computer. The words it heard are kept with the card in your project, beside the picture's description.

Agent benchmark sharing (only if you tick Share)

In Settings > Remote, "Benchmark this model" runs our 28 agent tests on a model. If you tick "Share the result anonymously" and the run finishes in full, without being stopped and without errors, the result goes to our own service at bench.cubric.studio, which runs on Cloudflare. What goes: the provider (DeepInfra, OpenRouter, OpenAI or Ollama, never a custom endpoint and never its address), the model id, which of the 28 tests passed, the cost per chat, the app version, the seconds each test took and, for a local Ollama model, your graphics card name and its VRAM. Never your prompts, the model's replies, your keys, file paths or projects, and no user id. A short confirmation or error message comes back. The anonymous result is kept and shown publicly on bench.cubric.studio. It cannot be deleted afterwards, because nothing ties it to you. Like any web service, it sees your IP address. We use that only to limit uploads to 10 a day, keep it only as a hash made with a secret that changes every day, and delete those hashes after 2 days. Cloudflare's privacy policy applies to the hosting.

Community benchmark scores

When the app lists agent models in Settings > Remote (at most once a day), it asks bench.cubric.studio for the scores of the current test suite. This needs no tick box, and it sends no prompts, keys or project data. As with any web request, the service sees your IP address. What comes back is the community median score for each model. The app keeps nothing on your computer beyond a one-day cache. Cloudflare's privacy policy applies to the hosting.

Dev updates

The bell on the home screen shows posts from the developer's #dev-updates channel on Discord. To fill it, the app asks our own service at updates.cubric.studio, which runs on Cloudflare, for the latest posts when it starts and every 30 minutes while it is open. It sends nothing about you, your projects or your keys. As with any web request, the service sees your IP address, and it is never written down. The images in the posts come through the same service, so the app never contacts Discord. Which posts you have read is kept only on your computer. Discord sees nothing unless you click Join the Discord or Open in Discord, which opens your browser. Cloudflare's privacy policy applies to the hosting.

Model licences (a few models only)

Some model makers grant their licence to each person on Hugging Face (FLUX.2 Klein 9B, for example). To check you have it, the app asks Hugging Face, using an access token you paste. Once the check passes, the token is saved on your computer, encrypted like your API keys above, so the next licence that needs it does not ask again; it is only ever sent to Hugging Face, and you can delete it in Settings. The app also keeps a note, on your computer, that you accepted the licence and when.

Bug reports (only when you click)

The report button shows you the report before it sends anything: the error, what you wrote, your operating system, app version and graphics card, and a Discord name if you add one. Nothing is sent until you press Send report. The report also carries the app's log (the end of it, at most about 512 KB), so we can find the cause without having to ask you for it; the dialog says so under the button. The log records what the app did and can include your prompts, file and model names, and error messages. It goes to our licence server, which posts the report, with the log as a file, in a private Discord channel only the developer can read, and the same error from several people joins one thread. If you have a licence key, the server checks it so we know the report comes from a buyer; the key itself is never posted. Before anything is sent, API keys, tokens and licence keys are taken out of the report and the log, and so are your home folder's path, your computer account name and any email address. A file path outside your home folder can still show, so read the error first. If you are a beta tester, the report also carries your hardware details (see "Beta testers").

Hearing back. When a report is sent, our server gives the app a random receipt number that says nothing about you. The app keeps it, with the report's title, only on your computer, and while it holds one it asks licence.cubric.studio once an hour whether that report has been fixed, sending nothing but the receipt numbers. When the developer marks it fixed, the bell tells you which version fixes it. Dismissing that note deletes the receipt. Our server keeps a receipt for at most 180 days, linked only to the report's Discord thread. The Discord name you typed is remembered on your computer for your next report, and is sent only with a report you send.

Using Cubric Studio from other AI apps (MCP)

AI apps such as Claude, Codex or Gemini can drive Cubric Studio through MCP. That connection never leaves your computer: the app only listens on 127.0.0.1, turns away requests from web pages, and the Claude Desktop extension talks to nothing but the app.

But whatever the AI app reads through Cubric Studio becomes part of your conversation with it. Project names, file paths, model lists, prompts and results go to that AI's provider (Anthropic, OpenAI, Google or whoever runs it) under their privacy policy, not ours. Only connect an AI app you are happy to share your projects with.

This website

cubric.studio is hosted on GitHub Pages. It sets no cookies and has no analytics, and its fonts and videos are served from the site itself. GitHub receives your IP address and browser details to deliver the pages, under GitHub's privacy statement. The home page asks GitHub, from your browser, for the latest release, so the download buttons point at the right files. Links to GitHub Sponsors, PayPal, Gumroad and Discord take you to those sites, under their own policies.

If you support us or get in touch

Apart from the IP addresses our services see, the reference videos that pass through the relay, the hashed key and computer IDs on our licence server, the Discord name of a beta tester or applicant and a tester's hardware details, and the bug reports you choose to send (all described above), this is the only personal data we receive:

Why we may use it: to give you what you paid or pledged for (contract), to answer you, keep our downloads working, pass a reference video you attached to the model you chose, see which computers a beta version was tested on, and limit uploads to the benchmark service and the relay (legitimate interests), and to keep tax records (legal obligation).

How long: payment and pledge records for as long as UK tax law requires, usually six years. Emails until the matter is closed and no longer than we need them. The hashed IP addresses that limit benchmark uploads, 2 days. Reference videos, less than an hour. A computer's licence record, until it signs out or has not checked in for 30 days. Bug reports, until the problem is fixed. Bug report receipts, at most 180 days. A beta application, until it is approved or turned down. A beta tester's Discord name and hardware details, until they leave the beta.

Where it is held: most of the services named on this page are based in the United States. Where we use them, UK law requires the transfer to be protected (for example by the UK-US data bridge or standard contract clauses), and they provide that. Where you use them with your own account, your agreement with them applies.

Your rights

You can ask us for a copy of your personal data, and ask us to correct it, delete it, limit or object to how we use it, or hand it over in a portable form. Email contact@madponyinteractive.com and we will reply within one month. Everything the app makes is already on your own computer, under your control.

If you are unhappy with how we handled your data, you can complain to the Information Commissioner's Office at ico.org.uk.

Children

Cubric Studio is not made for children under 13, and we do not knowingly collect their data.

Changes

When the app or this site starts sending something new, we update this page and the date at the top.