Sora's Grand Entrance Disables OpenAI Systems; Altman Halts New Sign-Ups Temporarily

OpenAI's 12-day event had mixed reactions from users, with the first two days leaving some unimpressed and craving more. However, the anticipation for the third day, which was heavily promoted, resulted in the much-awaited introduction of Sora, a feature that has Altman, OpenAI's CEO, extremely excited.

Sora's Launch and Features

Sora was officially launched in the United States and "most countries," offering its services to ChatGPT Plus and ChatGPT Pro users on Sora.com. Alongside, Sora Turbo was released, boasting enhanced capabilities such as faster processor times. The Sora model is capable of generating complex scenarios with multiple characters, specific actions, and accurate thematic and background details. It not only understands the user's request in the prompt but also comprehends how these elements exist in the real physical world.

OpenAI also unveiled the Sora Video Editor, capable of producing videos up to 20 seconds long while maintaining visual quality and adhering to prompt requirements. Users can describe the video through text in the input field at the bottom of the screen or upload images or video files in the initial prompt. Uploading any content not owned by the user or without all necessary rights is prohibited, as is uploading any images or videos of others without explicit written consent.

Video Generation and Editing

After submitting a prompt, Sora may take up to a minute to generate a video, during which users can adjust settings such as aspect ratio, resolution, duration, and the number of variants to create. However, changing these settings affects the points spent on video generation.

Once the video is generated, users can preview all videos by hovering over the library previews, comparing results. Users can click on individual videos to view, edit, and build upon the generated videos in a larger view. The Sora Video Editor includes features such as Re-cut for trimming and extending videos in a new storyboard, Remix for describing changes and generating new videos based on these, Blend for transitioning elements from one video to another, and Loop for creating seamless loops based on any part of the video. The storyboard feature allows users to create videos by selecting individual frames to generate in the video based on timestamps.

OpenAI Demonstrates Sora's Remix Feature

OpenAI showcased Sora's Remix feature, replacing a mammoth with a robot, which was quite impressive. However, unlike the first generation, which could complete the task instantly, it took a few minutes.

Users can also add videos to specific folders, report videos, or delete them, but once deleted, the content cannot be restored.

System Overload

Due to the overwhelming response, Sora's research vice president, Aditya Ramesh, and former CTO Mira Murati, also congratulated the team on the new video feature. OpenAI researcher Will DePue expressed his excitement about the product's interesting and intuitive nature, sharing his preference for generating fake historical footage.

ChatGPT Plus user Lance Ulanoff eagerly tried Sora by entering a prompt for a scene of a middle-aged man building a rocket by the sea. He described a tranquil scene with a moon, gentle waves, a bonfire, and a friendly dog, with default settings of 5 seconds and 480p. Based on his description, two video options appeared in about a minute: one with a dog but its tail on its head, and another with the dog correct and showing a man building a small rocket model.

System Overload and User Feedback

The system was quickly overwhelmed, and despite some features like the storyboard still functioning, no new AI videos could be generated due to the system's full capacity. Ulanoff noted that the system might take some time to settle, but once it does, Sora would be worth trying.

Altman later posted that due to higher-than-expected demand, they paused new registrations for the video generator Sora, and video generation speeds would temporarily slow down.

Concerns and Limitations

Some users expressed concerns about the realistic effects of the videos, fearing that children might develop inaccurate impressions of natural physical phenomena. There are also restrictions on the use of Sora, as it is currently only accessible to ChatGPT Plus and ChatGPT Pro users, with ChatGPT Free, Enterprise, and Edu accounts unable to access Sora. ChatGPT Plus users can create up to 50 priority videos (consuming 1,000 points) with a resolution up to 720p and a duration of 5 seconds, while ChatGPT Pro users can create up to 500 priority videos (consuming 10,000 points), generate unlimited relaxed videos (not consuming points but potentially queued during low website traffic), support up to 1080p resolution, 20 seconds duration, and create up to 5 videos without watermarks.

Video creation through Sora requires points, with different resolutions and durations consuming varying points. Generating multiple videos in one request is charged at the same rate as running two separate generation requests.

OpenAI has not yet disclosed when Sora will be launched in Europe and the UK. The list of countries supported by Sora can be found at the provided link. OpenAI may use user data for model training, and users can opt-out by selecting "Settings" > "Data Control" and disabling the "Improve models for everyone" feature in their profile.

Sora's Anticipation and Controversy

Sora has been a topic of great anticipation and attention due to its delayed official release. Recently, LontVideo's CEO Ruud van der Linden revealed that OpenAI's Chad Nelson announced the upcoming release of Sora v2 in a C21Media keynote in London. According to leaked video content, Sora v2 can generate dynamic videos up to one minute long, supporting text-to-video, text+image-to-video, text+video-to-video, and video-to-video. However, this release did not meet the mentioned expectations.

Sora was also maliciously leaked on huggingface last month, opening access for about an hour. The reason was a significant conflict between artists granted early access to test Sora and OpenAI, who felt exploited for free labor and overly controlled in their early testing plans. In response, these artists leaked Sora's video generator access to express their dissatisfaction with OpenAI.

The leaked information indicated that this version of Sora could generate videos with a maximum resolution of 1080P and a duration of 10 seconds. Some users found that the leaked API was actually Sora Turbo. However, an OpenAI spokesperson responded by stating that Sora remains in the "research preview" phase.

With Sora's official release, the question remains whether it has met the expectations of its eager audience.