Draft for counsel review. Not final.
How we use AI
Draft, final text to be approved by Madrivo's counsel.
Draft notes for counsel (preview only)
- Whether to name the model providers on the page, or describe them by role. This draft names the models, because the page's job is to show the ground is known.
- The parse model is chosen in phase V-A and is not named here yet.
- California SB 942 as amended by AB 853 binds a covered provider of a generative AI system with over a million monthly users. We are a downstream user of a third party model, so the content-credential duty sits with the provider. Confirm, and revisit near a million monthly users.
Added by the second legal seat, 2026-09-17
- DMCA designated agent, registered with the US Copyright Office. A prerequisite to the safe harbour for user uploaded photographs, and a filing rather than a clause.
- Verification. The Privacy Policy matches one email address. Confirm that is enough for a deletion asked for by someone with no account.
- California wiretapping on the consent certification recording of the quote form.
- Whether continued use carries assent to a material change of terms.
- The enrichment seams. Whether a litigator scrub or a do not call lookup needs disclosure beyond the Privacy Policy's contact checks row.
Build note. The same absolute structure claim this page just lost also sits in an internal reference about.aboutPictures. It is outside this task's write scope and needs the same rewrite.
Every picture in the studio is made by artificial intelligence from a photo you upload. None is a photograph of finished work.
What the models do. One checks that your photo is a bathroom before anything else runs. One reads what you type in the chat and turns it into a structured change, such as "walls, colour, sage green". One makes the picture. Today those are google/gemini-2.5-flash-lite for the check and openai/gpt-image-2 for the picture, with google/gemini-3.1-flash-image as a fallback. The chat reading model is being selected and will be named here before launch. We run none of them ourselves, reaching them through an API provider under contract.
How you can tell. Every version carries the badge "AI concept. Not your actual result." burned into the picture itself, a caption below it, a line saying which model made it and when, and a list of what was kept and changed. Those are the disclosure. Please leave them on if you pass the image on.
About content-credential metadata. Some models attach C2PA content credentials to what they make. We tested ours. The credentials do not survive the resizing, re-encoding and badge burn-in every version goes through, so we do not claim to carry them, and you will not find them in a file you download. We chose a badge in the pixels instead, because it survives what a metadata block does not: screenshots, messaging apps, printing. On our side we keep the model, date, prompt version and exact instruction for every version.
What the pictures get right, and what they do not. The studio is built to keep your room's shape, window, door and proportions and to change only the finishes and fixtures your budget range allows. It gets that right most of the time, and some versions come out better than others. It works from one photograph, so it does not know your plumbing, wiring, building code or local prices. A version is something to think with and show people. It is not a plan, a drawing, an appraisal or a quote, and not a promise of any result.
What a version does not cost or promise. A version does not function as a construction drawing or a permit set. What your project actually costs, and whether it can be built as shown, depends on your home's structure, code and permit requirements, and on the company you hire to do the work. Individual results vary.
What we do not make. No people, text, logos or brand names. Ask for any of those and the studio will refuse.
Your information. We send your photo and your chat messages to the AI provider to do the work above. The provider may not use them to train AI models and may not use them for anything else. On every text request we set the provider flag that refuses data collection, and we record each image provider's published API data use terms before saying this on the site. Our Privacy Policy sets out the arrangement, including what happens if a provider's terms change.
If something looks wrong. If a version misrepresents your room, ignores structure it was meant to keep, or produces something it should have refused, tell us at support@remodelize.ai with the project and version. We read every report and use them to fix the prompts and the rules.