The AI Video Explosion

Luma AI has introduced its Dream Machine API, intensifying competition within the AI video generation space just hours after rival Runway launched its own API. This latest tool allows developers, startups, and enterprises to weave video generation into their applications, featuring text-to-video, image-to-video, keyframe control, video looping, and camera motion control.

Constructing on the success of the Dream Machine model, known for its striking realism and ease of use, Luma’s API is already survive Hugging Face, unlike Runway’s which stays on a waitlist.

Priced at $0.32 per million pixels, it offers an accessible option for smaller developers, while its Scale feature caters to larger enterprises with bespoke support. The API emphasizes responsible use, ensuring privacy and moderation controls. Read more.

Runway AI has launched an API for its Gen-3 Alpha Turbo video generation model, advancing its position within the AI video sector. This latest API enables organizations to embed Runway’s advanced video creation tools directly into their platforms and workflows.

The Gen-3 Alpha Turbo model enhances video fidelity and motion depiction, facilitating real-time video editing and automatic tasks similar to rotoscoping and motion tracking. Initially available via a waitlist, the API offers two subscription plans: one for people and small teams, and one other for enterprises.

Pricing is ready at one cent per credit, with five credits needed for a one-second video. Runway’s API goals to streamline video production, with initial feedback from early adopters like Omnicom Group. Read more.

On the Snap Partner Summit, Snapchat introduced a brand new AI video-generation tool for creators. This latest tool allows creators to supply videos from text prompts, with plans so as to add support for image-based prompts in the long run. Currently, the tool is in beta and available only to a select group of creators via the net. Snap has no immediate plans to increase access beyond creators. The tool is built on Snap’s own foundational video models, with transparency measures in place, including visible icons and watermarks on AI-generated content.

Snap has implemented safety protocols throughout the model’s workflow, from prompt generation to output, ensuring harmful content is avoided. While the announcement positions Snap ahead of competitors like TikTok and Instagram, which lack similar tools, Meta’s Make-A-Video, revealed in 2022, stays unreleased. The effectiveness of Snap’s tool compared to imminent offerings stays to be seen. Read more.

OpenAI has restructured its internal Safety and Security Committee into an independent board oversight committee tasked with delaying model releases over safety concerns. The choice follows a 90-day review of OpenAI’s safety protocols, revealing the necessity for greater independence in oversight.

Chaired by Zico Kolter and including Adam D’Angelo, Paul Nakasone, and Nicole Seligman, the committee will oversee major model releases and has the authority to halt launches until questions of safety are resolved. This latest structure parallels Meta’s Oversight Board, though OpenAI’s board members are involved in each bodies. CEO Sam Altman is not any longer a part of the committee, however the broader board will receive updates on safety matters. Read more.

In a related development, OpenAI has increased access to its AI models in response to user demand. For Plus and Team users, the speed limit for o1-mini has surged from 50 messages per week to 50 messages per day. Meanwhile, the speed limit for the more resource-intensive o1-preview has increased from 30 to 50 messages per week.