Summary
Highlights
Sora 2 introduces automatic scene cuts within generated videos, a significant improvement over Google V3, where users previously had to manually prompt for cuts. Sora 2 allows for simple prompts and image uploads to generate complex video sequences with varied shots, like property walkthroughs featuring a real estate agent. A custom GPT for optimal Sora 2 prompting will also be shared.
Sora 2 offers a free model with 100 generations per day. To access it outside the United States, a VPN (like NordVPN) and a ChatGPT account are required for login. An invite code is also necessary, which can often be found in comments or through platforms like Higsfield, which offers unlimited generations for pro subscribers (up to 12 seconds, wide or portrait aspect ratio).
Using Sora's native platform offers faster generation times compared to third-party sites like Higsfield. Sora also includes a 'Cameo' feature, allowing users to create videos with specific characters, such as Sam Altman or a custom avatar created via the mobile app (requires an American iTunes account). This enables unique storytelling with consistent characters.
The video introduces a custom GPT designed for Sora 2, which facilitates 'timeline prompting.' This method breaks down video generation into structured parts (e.g., first 3 seconds for hook, 3-6 seconds for context), detailing camera motion, sound effects (SVX), and voice-overs (VO). This technique ensures highly specific and detailed video outputs, significantly improving workflow and quality.
A demonstration using a Clash Royale prompt highlights Sora 2's advanced context understanding. The AI accurately generates specific game elements, sound effects, and even a natural-sounding voice-over. This showcases the model's ability to maintain high consistency in character and environment throughout the video, which has been a major challenge in previous AI video models.
Sora 2 can be used for practical applications like real estate property showcases. By uploading an image of a property and providing a detailed prompt, Sora 2 can generate videos featuring a real estate agent giving a tour with various camera shots. The model also excels at integrating specific products into scenes, as demonstrated with a Moroccan tile and Hawkston beer, even if they lack extensive pre-existing training data.
Sora 2 can be prone to content violations due to high traffic; users can adjust prompts using ChatGPT to resolve these issues. The video also explains how to access higher quality video generations (1080p, 15 seconds long) and adjust aspect ratios (portrait/landscape) through a paid pro subscription ($200/month) within Sora's settings.
Sora 2 videos generated directly on the platform include watermarks. These can be removed using advanced AI tools like enhancer.ai's content-aware inpainting feature. The process involves uploading the video and masking out the watermark, which can also be done manually in video editing software like Premiere Pro for precise control.
Sometimes, giving Sora 2 creative freedom with a generic prompt can yield impressive results, especially for longer videos. Examples include a 15-second beer advert focusing on the English countryside, which showcases its ability to create compelling shots based on broad descriptions. This flexibility allows users to balance precise control with AI's creative interpretations.
The video showcases several impressive Sora 2 generations, including user-generated content (UGC) style videos with realistic animal interactions, CGI-style trailers inspired by animation movies like 'Robots,' and cinematic military trailers. These examples highlight Sora 2's capacity to produce high-quality visuals, complex narratives, and detailed UI elements from simple text prompts, rivaling professional animation quality.
A more advanced prompting structure for the Sora 2 Pro version is introduced: 'timelight structured cinematic prompting.' This syntax uses headers, markdown bullets, and time cues for semantic layers, allowing for precise control over art direction, continuity, pacing, and movement. This method is ideal for users with the Pro version, as it bypasses character limits and offers a reliable way to achieve highly consistent and cinematic results across multiple shots.
Sora 2 allows for highly creative and artistic storytelling, such as generating a fire and water helix transitioning into a fire department scene, demonstrating its versatility in creating unique visual concepts. The ability to integrate specific products into these custom scenes, like the Hawkston beer into a cinematic horror-themed prompt, further expands its potential for advertising and branding.
A review of the generated videos, including a real estate agent property tour and updated Hawkston beer advertisements, shows mixed but largely impressive results. The AI's ability to maintain consistency, generate scripts, and create visuals from a basic prompt is highlighted as mind-blowing. The video concludes by emphasizing the endless possibilities with correct prompting and providing resources for users to explore and create their own compelling content.