Product / Model terms and laws
Can you use it? Model terms and laws.
Each file shows the usage terms of the model behind it: commercial use, training on your work, who owns the output, where it may run. Each project lists the rules for the markets it will run in.
Answer a client's lawyer in minutes, with the clause behind each answer.
Per model
Two sets of terms. Both on the record.
You rarely reach a model directly. You use a service - Higgsfield, fal, Comfy Cloud, Runway - that runs a model someone else made. Two sets of terms can apply, so Archibal reads both and says which answer comes from which.
generate a shot, a voice, a still
The service you used
Higgsfield, fal, Comfy Cloud, Runway... Their terms govern your contract: what you may do with the output and what you agreed to when you signed up.
The company that made the model
Google, ByteDance, Black Forest Labs... Their terms can still govern what happens to your data. Where the two disagree, the stricter reading is shown and the other is noted.
What each answer covers
Commercial use
Can the output go into paid work? Conditions such as a revenue cap, a paid plan or own-use-only are shown next to the answer.
Training on your work
Does the service or the model maker train on your inputs, prompts or outputs - and can you opt out?
Who owns the output
You, the provider, or something in between, as the terms put it.
Indemnity
Will they defend you if the output infringes someone's copyright - and does that promise reach you, or only their direct API customer?
Marking and disclosure
What the model marks on its own (a watermark, Content Credentials) and what the terms ask you to disclose when you publish.
Territory
Where the outputs may be used, and where the service is offered at all.
Every answer shows where it came from: the source document, the clause it rests on, and the date it was read. A question the terms do not answer is marked as not on record, never guessed.
Try it
Look up a model you used this week.
The same record your projects use, open to anyone. Each model has a page per service it runs on, with both sets of terms side by side.
Commercial use, output ownership, training on your inputs, indemnity and marking, from the providers' documents with a link to the clause - per model, and per service you use it through (fal, Higgsfield...).
Per project
Where it will run decides which rules apply.
The same video can need an AI label in one market and nothing in another. Tell the project where the work will run, and it lists the rules for those markets and what to do about each.
- 01
Say where it will run
The countries and regions, the channels it will run on (social, broadcast, outdoor...) and what it is for - a commercial, a political ad, a film.
- 02
See which rules apply
The project lists the rules for those markets, and keeps what is yours to do apart from duties that fall on the AI provider or the platform - many of these laws bind them, not you.
- 03
Know what to do
Each rule comes with what to do, in plain words - add an audience label where it runs, get consent before using a real person's voice or likeness.
Rules on record include
Advisory, not a gate. The list informs your decision and your lawyer's; it never blocks a delivery. If your team wants a lawyer to sign off before anything ships, you can turn that on per project.
Read this part
A record, not legal advice.
Each answer is what the source document said on the day it was read. Terms change, so the record keeps the date, the changelog lists every change we recorded, and a delivery freezes the terms as they stood when you delivered - later changes never rewrite it.
None of it tells you what a law requires of you in your situation. Read the clause, and ask your lawyer before you rely on it. What Archibal saves you is the hunt: when a client's lawyer asks which models made the work and on what terms, the answer and its source are already in one place.
Frequently asked
What about the models' own terms - can I use the output commercially?
Every model in your project carries its usage terms on the record: whether commercial use is allowed, who owns the output, whether your prompts and outputs are used for training. Both layers are covered - the service you used (fal, Higgsfield, Comfy Cloud) and the company that made the model - taken from their own documents, shown plainly with the clause and the date it was read, and kept in the delivery as they stood that day.
Why are there two sets of terms?
When you reach a model through a service, you sign up to that service's terms, and the company that made the model has terms of its own. The service's terms govern your contract; the model maker's terms can still govern what happens to your data. Each answer says which one it comes from. Where both speak and disagree, the stricter reading is shown and the other is noted.
What if the terms say nothing about a question?
Then the record says it is not on record. Silence is shown as silence, never filled in with a guess.
Does an Archibal record cover EU AI Act Article 50 for me?
No. Article 50(2) machine-readable marking binds the AI model providers, and Article 50(4) audience disclosure binds deployers - whoever publishes deepfakes or certain AI content. What the record gives you is documentation: signed Content Credentials (a technique Recital 133 names) and a record of how the work was made, which brands and platforms increasingly ask agencies for.
Can the rules for my markets stop a delivery?
No. They are advisory: the project lists the rules that apply where the work will run and what to do about each, and you decide what ships. The only legal step that can hold a delivery is an optional lawyer sign-off, and only when your team turns it on.
Is this legal advice?
No. It is a dated record of what the source documents said, with a link to each clause. Read the source and ask your lawyer before you rely on it. Terms change: the changelog lists every change we recorded, and each model page lets you ask for an email when its terms change.