- Specialization
- Multimodal Models
- Business
Data Provisionand 2 more business valuesLess
- AI Research
- Model Evaluation
- Market
- Software Development
- Ecosystem
- Not classified
- Customers
Enterprisesand 1 more customers valuesLess
- Researchers
More details
- Organization type
- Commercial Company
- Ecosystem
- Not classified
- Business
- Data Provision, AI Research, Model Evaluation
- Specialization
- Multimodal Models
- Market
- Software Development
- Customers
- Enterprises, Researchers
- Delivery models
- Professional Services
- Revenue models
- Services Fees
- Listing status
- Not classified
- Latest equity round
- Not classified
- Reported funding amount
- Not classified
About
Velvet is a multimodal-model R&D company that creates audiovisual datasets and evaluations for AI labs and enterprises. Its focus on interactive capabilities, including spatial reasoning and real-world understanding, differentiates it from general-purpose data providers.
Market
Velvet competes at the intersection of B2B generative-AI video creation and multimodal data infrastructure. Its video offering is positioned as a unified production layer that combines multiple generation models, editing, brand-safety checks, collaboration, and managed services, rather than requiring customers to stitch together separate AI tools and professional editors.
Velvet targets B2B marketing and creative teams at brands and enterprises, particularly organizations producing high-volume external marketing content or lacking in-house video-production capabilities. It also serves AI labs and enterprises that need multimodal datasets and evaluations.
At a Glance
Problem
Velvet addresses a realism gap in frontier AI development: multimodal models need to understand and act within the kinds of interactive workflows that humans navigate, while conventional datasets and evaluations do not adequately capture those conditions. The company’s stated “killer use case” is helping AI labs and enterprises create data and tests that reflect real-world workflows and thereby improve models’ interactive capabilities.
The economic pain is strategically important but not publicly quantified in the available evidence. For model developers, better workflow-grounded datasets and evaluations are intended to make model development and measurement more realistic and effective; Velvet does not publish pricing, quantified customer savings, or a specific cost comparison.
Product / Service
Velvet is positioned as a multimodal model R&D and data company rather than primarily as a consumer video editor. It works with AI labs and enterprises to create multimodal datasets and evaluations, simulating how people navigate real workflows. Its official site also presents dataset-sample submission and meeting-request flows, suggesting a B2B, research-led engagement model rather than a fully self-serve public product.
The benefit is specialized data and evaluation infrastructure for frontier models: customers can use workflow-realistic material to train, test, and advance systems with interactive capabilities. The public site is deliberately high level, so the exact dataset formats, evaluation methodology, pricing, and degree of managed service are not disclosed.
Market
Velvet competes in the emerging multimodal AI data, evaluation, and world-model R&D market serving frontier AI labs and enterprises. The clearest documented adjacent competitors are Scale AI, whose platform supports multimodal evaluations, and Surge AI, which provides data-labeling infrastructure across AI data workflows. Velvet’s differentiation is its stated emphasis on realism in human workflows and on producing both datasets and evals, rather than only labeling data or providing a general-purpose evaluation tool.
Public evidence indicates an early-stage company, not a scaled commercial business: YC lists Velvet as founded in 2025, active, and part of its Fall 2025 batch, with four employees and open roles. Sacra reports approximately $500,000 in pre-seed funding. No customer, revenue, or usage metrics are disclosed in the evidence, so Velvet is best characterized as pre-revenue or traction-undisclosed. The corpus also contains a conflicting third-party description of Velvet as an all-in-one AI video-creation platform, but the current official site and YC profile consistently identify the company as a multimodal data lab.
Founders & Leadership
Funding History
Y Combinator
Recent News
A StartupIntros profile describes Velvet as a video-creation platform launched in 2025 to help marketing teams generate AI-powered video ads. The profile provides recent company and founder coverage but does not identify a specific partnership or integration.
Y Combinator’s company profile identifies Velvet as a multimodal data lab founded in 2025 by Lucas Mantovani and Lucas Tucker, with a San Francisco-based team. This is the clearest public profile coverage aligned with velvet.video’s stated focus on multimodal datasets and evals.
Active Roles
3Business Model
Velvet operates as a B2B AI-data company, working with AI labs and enterprises to create custom multimodal datasets and evaluations. Public materials do not disclose specific pricing, so its revenue appears to come primarily from customer engagements for these data and evaluation services.