OpenAI Releases Sora AI Video Generator to ChatGPT Plus and Pro Subscribers

OpenAI has officially launched Sora, its advanced AI tool for generating videos from text prompts, signaling a new era for creative AI. Integrated into paid ChatGPT accounts, Sora allows users to animate still images, extend existing videos, and merge scenes into cohesive narratives.

The tool, now available globally with certain regional exceptions, represents OpenAI’s push into the competitive AI video generation space.

Sam Altman emphasized during the launch event that Sora balances creative possibilities with robust safeguards.
 

Combining AI Innovation and Usability

Sora builds on OpenAI’s existing models, employing diffusion modeling—a technique where noise is gradually removed from random inputs to generate refined videos. Users can create 20-second high-definition videos in 1080p, with customizable aspect ratios like widescreen, vertical, and square.

The version launching today, named Sora Turbo, offers advanced features for video generation. Users can create clips ranging from 5 to 22 seconds in a variety of aspect ratios, including widescreen, vertical, and square. Available resolutions include 480p, 720p, and 1080p, catering to diverse content needs.

To manage usage, OpenAI has implemented a credit-based system:

• ChatGPT Plus users receive 1,000 credits per month, supporting up to 50 priority videos at 720p and 5 seconds in length.
• ChatGPT Pro users are allocated 10,000 credits monthly, allowing up to 500 priority videos at 1080p and 20 seconds. Pro users also benefit from unlimited “relaxed” videos—low-priority outputs without watermarks.

 
If you use Re-cut, Remix, Blend, or Loop to create content that falls outside of the 5 second increments, credits will be spent as follows:

 
Watermarks are included by default but can be removed for relaxed videos generated by Pro users. OpenAI’s system automatically resets credits each month, with no rollover for unused credits.

Unique features include a storyboard tool for precise frame-by-frame adjustments and a “Blend” option for seamless transitions between scenes. These tools cater to creators ranging from casual users to professional filmmakers.

The model also integrates recaptioning, derived from DALL·E, which enables it to follow text prompts with exceptional accuracy. By combining these capabilities, OpenAI aims to provide a versatile and user-friendly platform for storytelling.

Safety and Moderation at the Core

To address ethical concerns, Sora includes multiple safety mechanisms. Every generated video contains metadata via the C2PA standard, ensuring transparency about its AI origin. Visible watermarks are enabled by default, while sophisticated classifiers block explicit or harmful content.

OpenAI’s red-teaming process—spanning 15,000 test scenarios—helped refine moderation protocols. This included blocking adversarial prompts designed to evade safeguards. Early testers identified vulnerabilities in generating content with complex themes like violence or misuse of likenesses, leading to significant updates before the tool’s public release.

“Our focus is to ensure Sora is used responsibly while empowering creators worldwide,” OpenAI said in a statement.

The Sora API Leak and Ethical Questions

Sora’s launch follows a controversial API leak in November, when a group of early testers uploaded the tool’s API to Hugging Face, granting public access to the experimental version. The testers accused OpenAI of exploiting unpaid labor and imposing overly restrictive conditions during the early access program.

In a statement accompanying the leak, the group wrote: “Hundreds of artists provide unpaid labor through bug testing, feedback, and experimental work for the program for a $150B valued company. This early access program appears to be less about creative expression and critique, and more about PR and advertisement.”

They also criticized OpenAI for requiring pre-approval for sharing Sora-generated content, calling the restrictions “stifling.”

OpenAI responded to the backlash, emphasizing the voluntary nature of participation. “Participation is voluntary, with no obligation to provide feedback or use the tool,” a company spokesperson said. “We’ve been excited to offer these artists free access and will continue supporting them through grants, events, and other programs.”

Sora’s development has faced significant hurdles, from technical limitations to organizational shifts. Early versions of the tool required over 10 minutes to generate a single minute of video, reflecting the computational demands of AI video generation.

OpenAI addressed this with the introduction of Sora Turbo, which improved processing times and output quality.

The Competitive Landscape in AI Video Generation

Sora’s launch places OpenAI in direct competition with leading players in the AI video generation market, each offering unique capabilities and approaches:

• Adobe Firefly Video Model: Integrated into Premiere Pro, Firefly enables users to generate and extend videos from text and images. Adobe’s ethical approach, including training models on licensed datasets and embedding Content Credentials, has set a benchmark for transparency.

• Runway ML’s Gen-3 Alpha Turbo: Known for its speed and quality, this tool delivers video outputs seven times faster than its predecessor, with growing adoption in the entertainment industry through partnerships like Lionsgate.

• Google Veo AI Generator: Still in development, Veo promises capabilities similar to Sora but has yet to release a public version.

Open-source alternatives like Pyramid Flow also compete for market share, offering developers flexibility and high-resolution outputs without licensing fees.

Despite the controversies, Sora represents a significant step forward in AI-powered video creation. OpenAI has emphasized its commitment to refining the tool through phased deployment and user feedback.

Features like C2PA metadata and visible watermarks aim to address concerns about deepfakes and misuse of likeness, ensuring transparency and safety.

Last Updated on December 11, 2024 10:07 am CET