# Bria AI — Full Reference

Summary version: https://bria.ai/llms.txt

> Enterprise visual AI platform built on 100% licensed training data. Deterministic, reproducible outputs through Visual GenAI Language (VGL). Full IP indemnification for enterprise customers.

---

## Company Overview

Bria AI is an enterprise visual generative AI company headquartered at 1412 Broadway, Manhattan, NY, with R&D operations in Tel Aviv, Israel. Founded by Dr. Yair Adato (CEO), Bria was built on a core conviction: generative AI can only be trusted at enterprise scale when its training data, outputs, and attribution are fully transparent and legally sound. The company's technology powers visual workflows for enterprise teams at Getty Images, imgix, Elementor, Microsoft, AWS, NVIDIA, Envato, and Alamy, among others.

Bria's platform is differentiated by two interlinked innovations: Visual GenAI Language (VGL), a structured JSON specification language that makes generative outputs deterministic and inspectable, and a patented attribution engine that traces training data influence back to individual creators and compensates them based on revenue generated from their contribution. This "Spotify model for visual AI" — licensed data plus creator revenue-sharing — eliminates the IP exposure that characterizes most generative AI deployments and enables Bria to offer full indemnification to enterprise customers.

Unlike prompt-based systems that treat generation as sampling from a probability distribution, Bria treats it as rendering a fully specified visual state. The Fibo model family is trained natively on VGL, meaning developers express intent in structured JSON rather than natural language, and receive structured JSON back describing exactly what was generated. This makes Bria the only visual AI platform where outputs are auditable, reproducible, and suitable for agentic automation at scale.

Bria's compliance posture reflects its enterprise positioning: ISO 27001 certified, SOC 2 attested, C2PA compliant, EU AI Act pilot signatory, and Fairly Trained certified. Deployment is flexible — managed cloud, bring-your-own-cloud (AWS/Azure/GCP), on-premise, and on-device (edge) — with open-source model weights available on Hugging Face and GitHub for teams requiring full data sovereignty.

---

## Machine-readable metadata (Schema.org)

Public pages ship **JSON-LD** in the HTML body: `<script type="application/ld+json" id="json-ld-site-graph">`. The payload is one `@graph` linking Organization (`/#organization`), WebSite (`/#website`), SoftwareApplication + WebApplication (`/#software-application`), and Dataset (`/#primary-dataset`) on the canonical site origin. Use it for the same structured facts as Google Rich Results / Search Console. Each page also includes page-specific JSON-LD: WebPage + BreadcrumbList on all inner pages, FAQPage on pages with FAQ modules, and BlogPosting on article pages.

---

## Technology Differentiators

### What VGL Is and Why It Matters

Visual GenAI Language (VGL) is Bria's structured specification language for generative visual AI. It is analogous to what HTML is to web layout or what SQL is to databases — a formal, machine-readable language that encodes intent precisely, eliminating ambiguity.

Standard prompt-based generation suffers from an information mismatch: a natural language sentence encodes far less information than an image contains. This gap manifests as unpredictability — small prompt changes produce large, unexpected output shifts, and identical prompts produce different results across runs. For enterprise automation, this is a fundamental blocker.

VGL solves this by providing a formal control surface covering every visual dimension: geometry (spatial coordinates, bounding boxes, depth maps), color (RGB/HSL values, palettes), camera (angle, FOV, focal length, distance), lighting (direction, intensity, color temperature), composition (object relationships, layering, focus hierarchy), and attributes (material, texture, style parameters). Because the language matches the dimensionality of the image space, generation becomes rendering rather than sampling.

The Fibo model family is trained natively on VGL — not as a post-processing wrapper, but as the model's internal representation. When an image is generated, Bria returns the complete VGL specification used, enabling inspection, modification, and deterministic reproduction. The same VGL input always produces the same output. This property makes Fibo uniquely suited for agentic workflows, compliance-sensitive production, and brand-locked creative automation.

**Page**: https://bria.ai/the-vgl-paradigm

### How Licensed Data Differs from Web-Scraped Training

Most generative AI models are trained on data scraped from the open web — images and text collected without permission from their creators. This creates two enterprise-level risks: (1) legal exposure to copyright infringement claims, since scraped data frequently includes protected work; (2) output unpredictability, since models trained on inconsistent, unlicensed data cannot make strong guarantees about what they will generate.

Bria operates exclusively on licensed data from 30+ premium content partners including Getty Images, Envato, and Alamy. Each dataset element is covered by a commercial license that explicitly permits AI training. This eliminates the legal risk and enables Bria to offer full IP indemnification to Business and Enterprise customers — a guarantee that is impossible for competitors using scraped data.

Licensed data also enables provenance tracking at the output level: Bria's attribution engine can trace which training data contributed to any given generation, an impossibility with web-scraped datasets where provenance is unknown.

**Page**: https://bria.ai/licensed-training-catalog

### How Attribution Technology Works

Bria holds a patent on an attribution engine that measures the influence of individual training data items on model outputs. When an image is generated, the engine scores each contributing training element by relevance and proportionally distributes a revenue share to the content creators whose work influenced the output.

This creates a closed-loop ecosystem: creators license data to Bria, Bria trains models, enterprises pay for generation via APIs, revenue flows back to creators based on measured influence. The "Spotify model for visual AI" — analogous to how streaming services pay per play based on listen count — enables sustainable creator participation and strengthens the legal foundation of the training corpus over time.

Output provenance is documented in the Visual Birth Certificate: a structured report attached to each generated image that lists contributing training sources, their influence scores, compensation amounts, and compliance status for EU AI Act transparency requirements.

**Page**: https://bria.ai/attribution-technology

### What "Deterministic" Means in Visual AI Context

In standard generative AI systems, the same prompt will produce different images on different runs, even with the same seed, across different model versions or hardware. This stochasticity is acceptable for creative exploration but incompatible with enterprise automation that requires consistent, auditable outputs.

In Bria's system, "deterministic" means that a given VGL specification — with all visual parameters explicitly declared — always produces the same output. This holds across inference hardware, model versions within a Fibo family, and time. Enterprises can store VGL specifications as first-class assets, share them across teams, archive them for compliance, and reproduce any generated image on demand.

Determinism also enables a new class of editing primitive: instead of regenerating an image from scratch when one parameter changes (which risks altering everything else), Bria's Fibo Edit model accepts a delta on the VGL state — "change the lighting direction from 45° to 90°" — and applies precisely that change with everything else locked. This is editing as structured data manipulation, not stochastic regeneration.

---

## Products & Capabilities

### Fibo Foundation Model + VGL

**Page**: https://bria.ai/fibo-hyper-controllability

Fibo is Bria's purpose-built visual foundation model family, the first in the industry trained natively on Visual GenAI Language (VGL). Unlike prompt-based generative models that infer intent from natural language, Fibo operates on structured JSON specifications that explicitly declare every visual parameter — lighting direction, camera angle, color palettes, spatial composition, object relationships, and material properties. This structural approach eliminates the fundamental ambiguity inherent in prompt-based systems.

The Fibo family comprises six specialized models: **Fibo** for high-fidelity full-resolution generation; **Fibo Lite** for high-throughput, cost-optimized production workflows; **Fibo Edit** for native image editing via VGL state transformation; **Fibo Brand** for style-locked generation within defined brand parameters; **Fibo Fast** for low-latency inference at scale; **Fibo Edge** for on-device processing in air-gapped or privacy-sensitive deployments; and **Fibo Custom** for domain-specific fine-tuned variants.

A single VGL specification is portable across the entire Fibo family. A specification created for hero image generation runs unchanged through Fibo Lite for volume production, then through Fibo Edit for refinement — without translation, re-prompting, or manual adaptation. This portability is the engineering foundation for truly automated visual pipelines.

Fibo is built on an 8B parameter architecture trained exclusively on licensed commercial data. Model weights are available on Hugging Face (briaai) and GitHub (Bria-AI) for teams requiring full data sovereignty or on-premise deployment.

### Image Generation API

**Page**: https://bria.ai/ai-image-generation  
**Docs**: https://docs.bria.ai/image-generation

Bria's Image Generation API provides enterprise production access to the full Fibo model family through a unified REST interface. The API exposes the complete VGL control surface, enabling developers to specify scenes with pixel-level precision — no prompt engineering, no negative prompt hacks, no regeneration loops.

Core capabilities include **Text-to-Image (Fibo)**: standard-quality generation with full VGL parameter control, ControlNet (structural, depth, and recoloring guidance), and IP Adapter for image-prompt style transfer. **Text-to-Image Lite (Fibo Lite)** delivers cost-optimized generation for high-volume production, with a self-hostable variant for regulated industries requiring total data isolation. **Reimagine / Structure Reference** uses a reference image's structural depth to guide generation while introducing new materials and textures. **Prompt Enhancement** automatically expands simple prompts into rich, detailed VGL specifications.

API responses include both the generated image and the complete VGL specification used, enabling inspection, archiving for compliance, and programmatic modification. The API supports multilingual prompts, seed-based reproduction, 1-to-many variations, and negative constraints. All outputs are commercially safe under the standard indemnification terms.

Pricing (Development plan): Fibo $0.03/image · Fibo Lite $0.02/image · Structured Prompt $0.02/call. Business and Enterprise plans offer volume pricing and dedicated SLAs.

### AI Editing Suite

**Page**: https://bria.ai/ai-image-editing

Bria's AI Editing Suite provides production-grade image and video manipulation APIs built on the same licensed, commercially-safe models as the generation pipeline. Every tool is designed for automation at scale, with commercially clean outputs covered by Bria's indemnification terms on Business and Enterprise plans.

**Image editing**: Background Removal uses RMBG 2.0, Bria's open-source model available on Hugging Face with 25M+ downloads, delivering state-of-the-art foreground extraction. Background Generation replaces removed backgrounds with AI-generated scenes contextually matched to the subject. Generative Fill (inpainting) fills masked regions with context-aware content. Eraser removes unwanted objects with realistic reconstruction of the occluded region. Image Expansion (outpainting) extends images beyond original boundaries in any direction. Increase Resolution upscales images up to 8× with detail synthesis preserving sharpness at the target resolution.

**Video editing**: Video Background Removal provides frame-accurate, temporally consistent background separation for video clips. Video Increase Resolution upscales video to 8K. Video Eraser removes objects from video with temporal consistency across frames.

Pricing (Development plan): Background Removal $0.018/image · Fibo Edit/Gen Fill/Background Generation $0.03/image · Eraser/Image Expansion/Increase Resolution $0.02/image · Video Background Removal $0.01/sec · Video Increase Resolution/Eraser $0.02/sec.

### Tailored Generation (LoRA Fine-Tuning)

**Page**: https://bria.ai/tailored-generation

Bria's Tailored Generation allows enterprises to create custom visual AI models fine-tuned on their brand assets, product catalogs, mascots, and style guidelines. This enables brand consistency at generation time — not as a post-processing filter, but as a model-level guarantee — making every output compliant by definition rather than by review.

Three fine-tuning paths are available. **Self-service via Playground**: upload brand assets, configure training parameters, and deploy a custom Fibo variant within hours — no ML engineering required. This is available on all plans. **API-based fine-tuning**: programmatic training pipeline for automated model refresh when brand guidelines evolve or new campaigns launch, available on Business and Enterprise plans. **Expert-guided fine-tuning**: Bria's research team provides hands-on training for complex brand requirements, character consistency, and domain-specific specialization.

The fine-tuning technology is based on LoRA (Low-Rank Adaptation) applied to the Fibo foundation model. Effective adaptation requires as few as 20–50 curated images for style transfer; full fine-tuning for domain specialization uses larger, more diverse datasets. Custom model weights are available for download under open-source licensing on Business/Enterprise plans, enabling on-premise deployment and full sovereignty over the resulting model.

Reported outcomes: 93% faster time-to-production versus traditional brand photography workflows, 90% budget reduction in campaign production. Mascot and character fine-tuning is particularly effective for gaming, entertainment, and CPG brands.

### Product Shots

**Page**: https://bria.ai/product-shots

Bria's Product Shots suite automates the full lifecycle of e-commerce product imagery — from raw product photo to publication-ready catalog asset — without photography studios, stylists, or post-production teams. Every API call generates commercially safe, brand-consistent output at per-image economics that make traditional photography workflows uneconomical by comparison.

**Lifestyle by Text**: Generate lifestyle product photography from a cutout image and a text description of the desired scene ("ceramic mug, coffee shop, morning light, bokeh background"). The model places the product in a fully generated scene with accurate perspective, scale, lighting, and shadows. **Lifestyle by Image**: Use a reference scene photograph to place the product in a specific environment with precise lighting match and color harmony. **Product Cut Out**: Professional-grade background removal optimized for product shapes including transparent materials, complex packaging edges, and reflective surfaces. **Product Shadow**: Generate realistic drop shadows, contact shadows, and scene reflections consistent with the target environment's lighting. **Product Embedding**: Full scene generation around the product with accurate perspective integration, ambient occlusion, and environmental lighting.

The suite integrates with PIM (Product Information Management) and DAM (Digital Asset Management) systems for automated catalog updates. Shopify, Magento, and Salesforce Commerce Cloud integrations enable direct pipeline connection from product upload to live storefront.

Reported business outcomes: 80–90% cost reduction versus traditional studio photography, 15–20% reduction in return rates attributed to more contextually accurate product imagery, same-day response to trend shifts (versus 3–6 week traditional production timelines).

Pricing: Lifestyle by Text/Image $0.03/image · Product Cut Out/Shadow $0.018/image · Product Embedding $0.03/image.

### Visual PaaS

**Page**: https://bria.ai/visual-paas

Bria's Visual Platform-as-a-Service provides the infrastructure, SLAs, and deployment flexibility required for enterprise visual AI at production scale. The platform's 99.9% uptime SLA covers all managed cloud deployments, with dedicated throughput tiers at 10, 60, or 150 actions/minute (customized for Enterprise).

Deployment options: managed cloud (multi-tenant, lowest operational cost), BYOC — bring-your-own-cloud (Bria-managed software running in your AWS, Azure, or GCP account), on-premise (full on-site deployment for regulated industries), and on-device/edge (Fibo Edge for air-gapped or latency-critical deployments). Source code and model weights are available to Business and Enterprise customers, enabling fork-and-own deployments.

Bria is available natively on AWS Marketplace and Azure Marketplace for procurement through existing cloud commitments. Live video streaming support is available on Business/Enterprise plans.

---

## Customer Success Stories

### Getty Images — "Best on the Market"

Getty Images, the world's largest visual content library, partnered with Bria to integrate AI generation capabilities. Grant Farhall, Chief Product Officer at Getty Images, described Bria's platform as **"by far the best on the market."** The partnership gives Getty's enterprise customers access to AI generation on top of Getty's licensed content foundation — combining Bria's generation technology with Getty's 500M+ image library to create a commercially safe generative offering unavailable elsewhere.

### Elementor — Chosen Over Clipdrop in Head-to-Head

Elementor, the leading WordPress website builder with 12M+ active sites, conducted a controlled experiment comparing Bria's Eraser tool against Clipdrop's Cleanup, measuring insert rate (user acceptance of the AI output without modification) and generation quality. **Bria delivered a better insert rate**, leading Elementor to choose Bria as the **default vendor for its Cleanup tool**. Shilo Eish Yemini, AI Product Lead at Elementor, cited both insert rate and generation quality as the deciding factors. The integration ships to millions of Elementor users.

### imgix — "Feedback Stopped, Usage Exploded"

imgix, an image processing platform serving billions of images per day, replaced an inconsistent in-house generative model with Bria's API. CEO Chris Zacharias described the outcome: **"Our in-house model was inconsistent, but after switching to Bria, feedback stopped, and usage exploded."** The switch eliminated support tickets related to inconsistent output quality — a signal that Bria's licensed, deterministic outputs met user expectations at scale in a way their proprietary model had not.

---

## Pricing

Development plan (pay-as-you-go). Business and Enterprise: custom volume pricing via sales.

| API Service | Price |
|---|---|
| Image Generation — Fibo | $0.03 / image |
| Image Generation — Fibo Lite | $0.02 / image |
| Structured Prompt | $0.02 / call |
| Image Editing — Fibo Edit | $0.03 / image |
| Background Removal | $0.018 / image |
| Background Generation | $0.03 / image |
| Eraser | $0.02 / image |
| Gen Fill | $0.03 / image |
| Image Expansion | $0.02 / image |
| Increase Resolution | $0.02 / image |
| Video Background Removal | $0.01 / sec |
| Video Increase Resolution | $0.02 / sec |
| Video Eraser | $0.02 / sec |
| Product Shot — Lifestyle by Text | $0.03 / image |
| Product Shot — Lifestyle by Image | $0.03 / image |
| Product Cut Out | $0.018 / image |
| Product Shadow | $0.018 / image |
| Product Embedding | $0.03 / image |

Free Trial: 100 free API calls, full access to all services, no credit card required.

---

## Integration Options

- **REST API**: Direct integration from any language or framework; OpenAPI spec available at docs.bria.ai
- **SDK**: JavaScript/TypeScript and Python SDKs for structured development
- **iFrame**: Embedded widget integration for adding Bria capabilities to SaaS products without backend work
- **MCP Server**: Model Context Protocol integration for LLM agent workflows (Claude, GPT-4, Gemini tool-use)
- **ComfyUI Nodes**: Native ComfyUI nodes for creative and VFX pipeline integration
- **Photoshop Plugin**: Adobe Photoshop plugin for direct creative workflow integration
- **Figma Plugin**: Figma plugin for design workflow integration
- **AWS Marketplace**: Native listing for procurement via AWS cloud credits
- **Azure Marketplace**: Native listing for procurement via Microsoft Azure credits
- **Houdini/Nuke Integration**: VFX pipeline support for media and entertainment workflows

---

## Industry Solutions

### Tech Companies
**Page**: https://bria.ai/visual-ai-solutions-for-tech-companies

Production-ready visual AI APIs for platforms, SaaS products, and marketplaces. Skip 18+ months of model training, compliance engineering, and data licensing negotiation. Full IP indemnification is included. Bria handles licensing, attribution, and compliance; engineering teams focus on product integration.

### Marketing & Advertising
**Page**: https://bria.ai/marketing-teams-and-advertising-agencies

Campaign templates that scale to millions of brand-consistent variations. Mascot and character fine-tuning for consistent IP representation across markets. Reported 90% budget reduction on campaign production versus traditional methods. ControlNet and VGL ensure brand guidelines are encoded as parameters rather than reviewed as outputs.

### E-Commerce & Retail
**Page**: https://bria.ai/ecommerce-and-retail

80–90% cost reduction versus traditional product photography. 15–20% reduction in return rates with context-driven lifestyle imagery. Same-day response to trend shifts versus 3–6 week traditional production timelines. PIM/DAM integration for automated catalog updates. Shopify, Magento, and Salesforce Commerce Cloud compatible.

### Media & Entertainment
**Page**: https://bria.ai/media-entertainment-and-gaming

Concept art, storyboard automation, VFX and post-production support, matte painting, and set extensions. Character-consistent generative workflows for gaming asset production. Native ComfyUI, Houdini, and Nuke integrations for pipeline-ready deployment.

---

## Licensed AI & Compliance

### Licensed Training Catalog
**Page**: https://bria.ai/licensed-training-catalog

Bria licenses training data from 30+ premium content partners including Getty Images, Envato, and Alamy. Each data element is covered by an explicit commercial license permitting AI training. No web-scraped data. No unlicensed material. Provenance is tracked from source to model output via the attribution engine.

### Visual Birth Certificate
**Page**: https://bria.ai/birth-certificate

Structured provenance report attached per generated image (available on Business/Enterprise plans). Documents training data sources and their influence scores on a given generation, creator compensation amounts and recipient details, EU AI Act compliance status, and a full audit trail from prompt/VGL spec through output. Enables downstream copyright registration support and regulatory compliance documentation.

### Copyrightability Report
**Page**: https://bria.ai/copyrightability-report

Documentation supporting copyright registration for AI-assisted works (Enterprise plan). Includes complete generation history and VGL specifications, human edit logging and decision trail, selection and curation records, and the full licensed provenance chain. Designed for legal teams and intellectual property filing processes.

### Responsible AI
**Page**: https://bria.ai/responsible-ai

Six operational pillars: (1) Licensed data only — no web scraping or unlicensed training material; (2) Fair creator compensation — patented attribution engine distributes revenue proportionally; (3) Privacy by design — no generation of public figures without explicit consent; (4) Safe outputs — three-layer content moderation covering generation, output review, and policy enforcement; (5) Bias mitigation — active monitoring and dataset balancing across demographic and aesthetic dimensions; (6) Regulation readiness — EU AI Act pilot signatory, active engagement with NIST AI RMF and ISO/IEC 42001.

### Legal Lobby
**Page**: https://bria.ai/legal-lobby

Bria's IP policy framework and legal position on AI training data, attribution rights, and enterprise indemnification. Covers the contractual basis of Bria's training data licenses, the mechanics of indemnification coverage under each plan tier, and Bria's public position on AI copyright law, creator rights legislation, and regulatory compliance obligations.

---

## Security & Certifications

**Page**: https://bria.ai/security-and-compliance

| Certification / Standard | Status |
|---|---|
| ISO 27001 | Certified |
| SOC 2 Type II | Attested |
| C2PA (Coalition for Content Provenance and Authenticity) | Compliant |
| EU AI Act | Pilot signatory |
| Fairly Trained | Certified |

Bria's infrastructure supports private cloud deployment (BYOC on AWS, Azure, or GCP), on-premise installation, and air-gapped edge deployment for organizations with strict data residency requirements. All API traffic is encrypted in transit (TLS 1.3). Dedicated customer success and premium platform security are available on Business/Enterprise plans.

---

## Company

- **About**: https://bria.ai/about-bria — Leadership team, mission, and company history
- **Careers**: https://bria.ai/career — Open roles in New York and Tel Aviv (engineering, research, sales, product)
- **Contact**: https://bria.ai/contact-us — Sales inquiries and support
- **Legal**: https://bria.ai/legal-lobby — IP policy and legal framework
- **Blog**: https://bria.ai/blog — Technical updates, case studies, and company news
- **Architecture**: https://bria.ai/architecture — Platform technical architecture overview

**Leadership**: Dr. Yair Adato (Founder & CEO), Amit Goldenberg (President & CBO), Misha Feinstein (CTO), Vered Horesh (Chief AI Strategy Officer), Ron Mokady (VP Research)

**Investors and Partners**: Microsoft, AWS, NVIDIA, Getty Images, iStock, Envato, Alamy

---

## Contact Information

- **Address**: 1412 Broadway, Manhattan, NY, USA
- **Sales**: https://bria.ai/contact-us
- **Platform**: https://platform.bria.ai
- **API Docs**: https://docs.bria.ai
- **GitHub**: https://github.com/Bria-AI
- **Hugging Face**: https://huggingface.co/briaai
- **LinkedIn**: https://linkedin.com/company/bria-ai
- **Twitter/X**: https://twitter.com/bria_ai_
- **Discord**: https://discord.gg/bria