Over the past few years, VEED has established itself as a popular video editing platform powered by AI. They’re backed by Sequoia, trusted by 76% of Fortune 500 companies, and are used by millions of creators and businesses around the world to tell stories, create content, and more.

Recently, VEED partnered with Sieve to launch VEED Clips — a new tool that turns long videos into short, shareable clips in a few clicks. This has allowed VEED to deliver AI systems that are able to track different speakers and key subjects, resulting in “auto-edited” clips with proper framing.

Redefining AI-powered video creation

VEED believes AI can accelerate the video editing process by automating the tedious parts of common workflows, such as those involved in getting clips ready for social media. To quickly automate some of these workflows with AI, VEED tapped Sieve as the AI platform for developing several components required for Clips to work seamlessly.

While VEED has leveraged AI for years, integrating task-specific video models was a delicate challenge because it required working with heavy, unoptimized video models which could slow down the product experience.

VEED had to ensure high quality clips across a large domain of video content along with fast, cost-effective processing for a quick experience. Sieve is an AI toolkit built for developers working with video, and thus excels at solving a wide range of cases without compromising on performance.

Screenshots of the VEED Clips product

Building performant video AI pipelines with Sieve

Sieve’s infrastructure is purpose-built for video content due to its burst capacity allowing for parallel video processing, native GPU sharing especially useful for intertwined CPU/GPU operations, and performant video-focused building blocks perfect for key creative workflows.

VEED made use of key pipelines on Sieve such as active speaker detection and autocrop to power VEED Clips. These pipelines were used to efficiently:

1. ingest video content
2. cut videos into smaller chunks for parallel processing
3. detect key subjects such as people and other common objects
4. crop video chunks that consist of faces to track active speakers
5. auto-edit content to focus on key subjects and active speakers

The team at Sieve interviewed an engineer at VEED, Archie Edwards, who explained why they picked Sieve and how it’s been working with the team.

Collaboration that accelerates creative video

VEED’s in-house AI team knew it could deliver higher quality features quicker through direct collaboration with Sieve. They wanted to ensure they could do this while building products that embody safe and ethical use of AI technology.

Sieve ensures this by focusing on the last-mile of engineering AI models, rather than storing user-submitted content or training new models from scratch using it. It also has proprietary content moderation solutions designed to flag harmful content on the platform. VEED complements these efforts with additional guardrails, including algorithms that detect the misuse of their tech and verification methods that confirm the identities of the people before allowing things like voice cloning or avatar creation.

More progress coming

Sieve and VEED plan to launch much more together past VEED Clips, so stay tuned for futher updates!