Bria
Tool · this deep dive is built from the official pages of the platform itself and of the tools that support it.
Overview
Bria is a visual AI platform built for companies. The composition of the work comes from the pricing page and the integrations block: image generation on its own Fibo models, image editing, product photography (Product Shot) and video editing.
It is not a site with a "make me a picture" button but a set of ways to connect: API, SDK, iFrame, MCP and plugins. Generation is embedded into another product or workflow rather than opened in a tab.
The prompt language is not stipulated on any page read — not the pricing, not the terms of use, not the page about training data. From this it does not follow that only English works: the published documents say nothing about language.
Which platforms it works with
Word for word from the homepage: "SDKs, iFrames, APIs, MCP, Skills, PS and Figma plugins". The pricing page adds "Integrations: Figma, DCC and more". PS is, by context, the Photoshop plugin, but the tool gives no expansion, so the abbreviations stay as written.
These are places where the generation is embedded: a layout editor, a graphics program, other code through API or MCP. Publishing to social networks is not announced; the finished file is posted by hand.
Pricing
Prices from the pricing page, currency US dollars. The structure matters more than the sums: monthly subscription steps are not offered, the count runs per operation.
Pay-as-you-go — "Starting from $0.018 per generation", no ceiling on generations, 60 requests per minute, priority queue. Enterprise — price on request.
Per operation: image generation 0.02–0.03 USD by model (Fibo Lite or Fibo); image editing 0.018–0.03 USD; Product Shot 0.018–0.03 USD; video editing 0.02–0.0225 USD per second.
The minimum is 0.018 USD for one generation — the price of an operation, not of entry. No fixed monthly step is named, so "how much per month" does not follow from what is published.
What is free
Free of charge: 100 generations, access to all models through API, MCP, Skills and SDK, integrations (Figma, DCC and others), cloud-only deployment, 10 requests per minute, ordinary queue, Discord support.
One line deserves quoting: "Full data attribution on every image". Attribution is declared as part of the free offer rather than as a restriction — what that means is clearer in the section on rights.
What the free 100 generations lack is not about volume: commercial use rights are tied to a paid subscription.
Restrictions
The line between steps runs through speed and queue, not picture count.
Free — 100 generations, 10 requests per minute, ordinary queue. Pay-as-you-go — no limit on generations, 60 requests per minute, priority queue. Enterprise — deployment on-prem or in the customer's own cloud.
Enterprise is the only step where computation leaves the vendor's servers — the free plan is marked cloud and cloud only — and the sixfold gap in request rate is what a batch job meets first.
Rights to the output
Bria has a clause no other AI tool analysed in the catalogue has.
Attribution is embedded into the output, and removing it is forbidden. The terms forbid, word for word, to "remove, obscure, or alter any attribution, watermarks, or metadata embedded in Outputs by Company's attribution technology". The tool embeds attribution, watermarks or metadata into the delivered file, and the user may not delete, obscure or alter them. No differences by plan are stipulated. The wording lists both watermarks and metadata but does not say which applies in each case; C2PA is not named. The output reaches a client carrying a mark of origin, and removing it breaches the terms.
Commercial rights are tied to a paid subscription: "Certain Services, including but not limited to API access, advanced AI model features, and commercial use rights, are available exclusively to users with active paid subscriptions or valid licensing agreements". Commercial use sits in the same row as API access; on the free 100 generations it is absent. For Enterprise the pricing page adds "Commercial use rights for open-source models" and "Unlimited IP indemnification".
What the terms never state is who owns what is generated. No assignment of rights, as some other tools make, is described: the document speaks of the right to use, not of ownership.
Do they train on your data
The gap between what is said and what is passed over is the whole of it.
About its own training data the tool writes in detail: "Every asset in our training pipeline comes from commercial agreements with 30+ data partners", then "No scraping. No synthetic data. No gray areas" — from the licensed training catalogue page. That is the origin of the material the Fibo models were trained on.
About your uploads exactly one sentence appears, and it concerns fine-tuning: "our architecture keeps your data isolated and auditable".
What the source lacked: no direct statement that user uploads and prompts feed the general models — and no direct denial either. It answers where the training data came from, not whether yours joins it.
An opt-out mechanism is not described; that fact is marked absent. There is a ban in the opposite direction: the terms forbid using Bria's outputs to train other models.
How you earn with it
What bears on earning is the line between free and paid. A picture made on the free 100 generations may not, by the letter of the terms, go into commercial work; one made on Pay-as-you-go or under an Enterprise contract may. Enterprise adds "Unlimited IP indemnification", absent from both other steps. Over all of it the attribution ban stands: work sold to a client stays marked.
What it will not do
Let you strip the attribution it will not: removing, obscuring or altering it is forbidden outright, and the ban is not divided by plan.
Grant commercial rights on the free plan it will not: those are for paid subscriptions and licensing agreements only.
Let its outputs train other models it will not: using them "to train, develop, or improve any other artificial intelligence or machine learning models" is forbidden.
Accept unlawful material it will not: storing or transmitting "content which is unlawful, infringing, harmful or which violates any person's rights, including privacy rights and intellectual property rights" is forbidden.
Publish for you it cannot — social accounts are not announced — and leave the vendor's servers on the free plan it will not: deployment there is cloud only, on-prem is named for Enterprise.
Verified data
The checked data this deep dive rests on.
About the platform
A visual AI platform for companies: image generation on its own Fibo models, image editing, product photography with Product Shot and video editing — through an API, SDK, iFrame, MCP and plugins
The composition is taken from the pricing page and the integrations section of the official site
source, checked 2026-08-07
Platforms
Verbatim from the official site: «SDKs, iFrames, APIs, MCP, Skills, PS and Figma plugins»; the pricing page adds «Integrations: Figma, DCC and more»
PS means the Photoshop plugin judging by the context of the phrase; the source gives no expansion, so it is left as written
source, checked 2026-08-07
Pricing
Free: 100 generations with access to all models through the API, MCP, Skills and SDK, integrations (Figma, DCC and others), verbatim «Full data attribution on every image», cloud deployment only, 10 requests a minute, a queue and support on Discord
source, checked 2026-08-07
Pay-as-you-go — verbatim «Starting from $0.018 per generation», with no ceiling on the number of generations, 60 requests a minute, a priority queue; Enterprise — price on request. Per-operation price list: image generation 0.02–0.03 USD (Fibo Lite / Fibo), image editing 0.018–0.03 USD, Product Shot 0.018–0.03 USD, video editing 0.02–0.0225 USD per second
Billing is per unit; the pricing page offers no monthly tiers
source, checked 2026-08-07
0.018 USD per generation (Pay-as-you-go); the pricing page names no fixed monthly tier
The minimum price here is the price of an operation, not of a subscription
source, checked 2026-08-07
Limits and restrictions
Free: 100 generations and 10 requests a minute with a standard queue; Pay-as-you-go — no limit on the number of generations, 60 requests a minute and a priority queue; Enterprise — deployment on-prem or in your own cloud
source, checked 2026-08-07
Restrictions
The terms forbid storing and transmitting, verbatim, «content which is unlawful, infringing, harmful or which violates any person's rights, including privacy rights and intellectual property rights» and separately using the output «to train, develop, or improve any other artificial intelligence or machine learning models» (verbatim); there is no topical list (images of real people, politics, explicit content) on the pages read
Topical prohibitions are silence of the source; the general ban on unlawful content is confirmed
source, checked 2026-08-07
Legal
Differs by plan, and this is written in the terms verbatim: «Certain Services, including but not limited to API access, advanced AI model features, and commercial use rights, are available exclusively to users with active paid subscriptions or valid licensing agreements». The pricing page adds for Enterprise «Commercial use rights for open-source models» and «Unlimited IP indemnification» (verbatim). Who owns the output the terms do not state
The right to commercial use is expressly tied to a paid subscription — the free 100 generations do not carry it
source, checked 2026-08-07
The service embeds attribution in the output: the terms verbatim forbid you to «remove, obscure, or alter any attribution, watermarks, or metadata embedded in Outputs by Company's attribution technology»; no difference between plans is stipulated in this clause, and C2PA is not mentioned by name
The wording lists both watermarks and metadata but does not specify which is applied in each case
source, checked 2026-08-07
About its own models the service writes verbatim «Every asset in our training pipeline comes from commercial agreements with 30+ data partners» and «No scraping. No synthetic data. No gray areas»; about user uploads it says only that during fine-tuning «our architecture keeps your data isolated and auditable» (verbatim). There is no direct statement on the pages read that user uploads and prompts go into training the general models
The source describes the provenance of the training data in detail but does not answer the question about user input — that is exactly what is recorded here
source, checked 2026-08-07
Catalogue section: all similar See also: catalogue index · scheduler comparison · find by situation · platform restrictions