OpenAI first previewed Sora in February 2024, releasing multiple demo clips that showcased the model's ability to generate high-definition videos up to one minute long. The team named it after the Japanese word for "sky," symbolizing its limitless creative potential.
After nearly ten months of testing with a small group of safety experts and creative professionals, OpenAI publicly released Sora Turbo in December 2024 as part of its "12 Days of OpenAI" holiday series, making it available to ChatGPT Plus and Pro subscribers.
In September 2025, OpenAI unveiled Sora 2, a significant upgrade with improved physics simulation and a dedicated iOS app featuring social features and character creation capabilities.
Sora System Card
OpenAI's outlines the safety measures, external red teaming efforts, and ongoing research implemented to deploy the model responsibly.
The model was trained on diverse datasets including publicly available data, proprietary data from partnerships (such as Shutterstock and Pond5), and custom in-house datasets, with pre-training filtering to exclude harmful content.
To prepare for launch, OpenAI worked with external red teamers from nine different countries who tested over 15,000 generations between September and December 2024, identifying weaknesses and providing feedback on potential risks.
The company also collaborated with hundreds of visual artists, designers, and filmmakers from more than 60 countries to gather feedback on creative applications.
Key safety measures include age-gating access to users 18 and older, restricting face uploads and likeness features, and applying stricter moderation thresholds for content involving minors.
Additionally, all Sora-generated videos include C2PA metadata for transparency and visible watermarks by default, while OpenAI actively blocks the generation of child sexual abuse materials and sexual deepfakes.
This information is important to understand the reasons why we will see that there are specific generative creation requests we can make to Sora, and others for which Sora is not the preferred model.
Sora 2 Prompting Guide
According to Sora 2 Prompting Guide, there are two valid approaches: detailed prompts give control and consistency, while lighter prompts open space for creative and unexpected outcomes. Using the same prompt multiple times will produce different results—this is intentional.
API Parameters (Cannot Be Changed via Prompt)
Model: sora-2 or sora-2-pro
Size: Resolution options depend on model (720p for sora-2; up to 1792x1024 for sora-2-pro)
Seconds: "4", "8", or "12" (default is 4)
Higher resolutions generate better detail and texture; shorter clips (4 seconds) follow instructions more reliably than longer ones.
Effective Prompt Structure
A good prompt should describe:
Camera framing (wide shot, close-up, angle)
Depth of field (shallow vs. deep focus)
Action in beats (specific, timed movements)
Lighting and color palette (quality, direction, anchors)
Weak prompt: "A beautiful street at night" Strong prompt: "Wet asphalt, zebra crosswalk, neon signs reflecting in puddles"
Motion and Timing
Keep it simple: one clear camera move and one clear subject action per shot. Describe actions in beats:
Weak: "Actor walks across the room"
Strong: "Actor takes four steps to the window, pauses, and pulls the curtain in the final second"
Using Image References
You can use an input image as the first frame anchor. The image must match the target video's resolution. This locks in character design, wardrobe, and aesthetic while the text prompt defines what happens next.
Dialogue and Audio
Place dialogue in a dedicated block below the visual description. Keep lines concise—a 4-second shot accommodates 1-2 short exchanges. For silent shots, suggest rhythm with ambient sounds like "distant traffic hiss."
Ultra-Detailed Prompts (For Cinematic Control)
For complex shots, you can specify: format/look, lenses/filtration, color grade, lighting setup, location/framing, wardrobe/props, sound design, and detailed shot lists with timing and camera notes.
Remix Functionality
Use remix for controlled, incremental changes—one at a time. Specify exactly what you're changing: "same shot, switch to 85mm" or "same lighting, new palette: teal, sand, rust."
When to Use Sora
Sora excels at generating animated and cartoon-style videos, particularly when working with 3D pixel art, Pixar-style animations, and stylized characters.
As demonstrated in the video below, Sora is highly effective for creating commercial ads featuring animated characters, especially when you need to generate multiple scenes with cuts and transitions within a single prompt using brackets with "cut" commands to create scene changes like:
"old man remembers" then "cut" to "old man building a Lego car."
The model performs exceptionally well with character consistency in animated styles, making it ideal for storytelling sequences where the same character appears across different scenes and emotional states.
Sora is also recommended for generating videos that require specific camera movements and angles, as you can instruct the AI to change perspectives between cuts.
The Sora Pro version is particularly suited for high-quality commercial productions where maximum visual fidelity is needed.
Additionally, Sora handles well
nostalgic and emotional narratives,
product placement integrations in animated environments,
and scenes requiring characters to perform actions like crying, jumping, screaming, or interacting with objects all while maintaining stylistic consistency throughout the entire video.
When NOT to Use Sora
Sora is NOT recommended for generating videos featuring real people or photorealistic human footage. For realistic human video content, alternative models like Kling, Sync, or Veo are better suited.
Sora also struggles with precise real product placement and logo reproduction. As you can see in the video, there were multiple failed attempts to correctly display the Lego logo and promotional posters at a racetrack, requiring three or four regenerations to achieve acceptable results.
Obviously, the number of attempts decreases as the quality of the image, the prompt and the logo we insert as a reference improves. A low-quality image will probably not be suitable for a commercial that we want to keep the details in every image, in every small detail. In general, higher resolutions generate detail, texture, and lighting transitions more accurately, while lower resolutions compress visual information, often introducing softness or artifacts.
Obviously, the number of attempts decreases as the quality of the image, the prompt, and the logo we insert as a reference improves. You can guide a generation with an input image, which acts as the first frame of your video. The image must match the target video's resolution.
For very long continuous scenes without cuts, Sora may introduce imperfections that require layering multiple generations together taking the best parts from different outputs and combining them manually in the video editor to achieve the desired quality.
