Openai plans to release a new version of its flagship AI video model Sora someday this quarter. The revolutionary thing at launch is that Sora has lost his position as a competitor, and Google's VEO 3 sets the gold standard for AI video generation.
SORA 2 is expected to arrive in the coming weeks given the rapid release of the GPT-5. Like the GPT-4O, the GPT-5 is natively multimodal, processing input or output types (including video) while performing complex inference tasks similar to the “O” series models.
Sora remains a solid platform. Its storyboarding features break new ground, allowing ChatGpt Pro subscribers to generate clips up to 20 seconds. However, the underlying model indicates its age. The output, unlike Veo 3, Kling 2.1, or Minimax 2, suffers from motion control issues, no sound generation, and complex physics rendering.
Even in the social video field, Openai is currently facing competition from almost every AI platform, including Meta, Grok, and Midjourney. But Openai is the world's largest AI lab with important resources, and despite Meta's recent talent raid, it is a formidable engineering team. Don't count yet.
What Openai needs to make SORA competitive
To compete with Google's video models or emerging Chinese competitors, Openai must take advantage of multimodal features while expanding the Sora feature set. It's not painful to narrow ChatGpt integration either. There are five important improvements to Sora 2.
1. Native sound generation is not negotiable
Please take a look
If Openai wants to compete with VEO 3, SORA 2 must handle both video and audio natively. Models without sound generation start with disadvantages.
Currently, SORA generates only silent clips. This is a major weakness when the VEO 3 generates sound effects, ambient noise, and even dialogs as a core feature. This is not just about tacking audio as an afterthought. It's about true integration.
VEO 3 can generate lip-synced character utterances in multiple languages. SORA 2 requires the same built-in audio features, from atmospheric soundscapes to spoken language.
It's not just catching the Veo 3 when Openai offers full multimodal generation (video + audio) while maintaining clips of over 20 seconds. You could jump completely.
2. Physics simulation needs to be dramatically improved
Visual realism is beyond resolution – it's basically about physics. Current SORA outputs often show unnatural motion and distorted physics. It is a water that defies gravity, a predictively deforming object, or a movement that feels fundamentally wrong.
Google explicitly prioritizes real-world physics with VEO 3, and the results speak for itself. Their videos are excellent at simulating realistic physics and dynamic movement with minimal glitches. On the other hand, SORA's older models produce unstable movement and inconsistent object interactions, shattering the immersion.
For SORA 2 to compete, its model needs to better understand real-world behavior, from natural human walking to smoke dynamics to fluid mechanics, to ball bouncing. Openai essentially needs to integrate the physics engine into SORA. Incredible movement and interaction (no more limbs or fusion backgrounds) bridge the significant gap with competitors.
3. Conversation prompts must be standard
An ace of open in the hole? ChatGpt has already trained millions of people to communicate conversations with AI. SORA 2 needs to take advantage of this by making video creation feel like dialogue, not programming.
Rather than requesting full prompts or complex interface navigation, the system should support natural before and after improvements. Google is already moving in this direction. The flow tool uses Gemini AI to enable intuitive, everyday language prompts.
Runway does this brilliantly in chat mode, using a new Aleph tool that allows Gen-4 to cleverly refine a single element. Luma's Dream Machine has this concept built from the ground up.
Imagine this workflow: simply type “Medieval Knight on a Mountain,” receive the draft video, say “Add Dragon to Sunrise,” and Sora will update the scene instantly. This conversational approach reduces barriers for newcomers and accelerates the workflow of professionals.
Technology exists. CHATGPT already interprets follow-up requests and dynamically adjusts the output (as demonstrated by GPT-4OS native image integration). Fully integrated with ChatGpt, let's talk about the path to great videos. Its user experience exceeds the technical prompts that most competitors still need.
It also allows native image generation first and then animation using Sora, similar to how Google works with Veo 3 for Gemini or New Grok Imagine features.
4. Character consistency and customization are essential
Character and scene consistency represents another important area of improvement. Now, generating two clips of “The Girl in a Red Dress” could potentially create two completely different people. The SORA output drifts across generations of styles and details, making consistent multi-scene stories or repetitive characters almost impossible.
SORA 2 requires that you enable consistent characters, objects and art styles for long videos and clip series. Competitors already offer this – Kling 2.1 boasts “consistent characters and film lighting directly from the text prompt.” Google's flow goes further, allowing custom assets (character images, specific art styles) as “intensities” across multiple scenes.
OpenAI should provide similar features such as uploading reference images, tweaking styles, or persisting characters throughout the scene. If SORA 2 can maintain a consistent character look throughout the video, creators can actually tell the story instead of creating a disconnected clip. Especially if you have native audio integrations that exceed clips of more than 20 seconds.
Consistency and customization require Sora 2 to provide that control, whether it's an artist who maintains a signature style or a filmmaker who needs character continuity.
5. Deep chat integration and universal access
Finally, Openai needs to maximize the benefits of its ecosystem by ensuring wide accessibility while deep integration with Sora 2 into ChatGpt. Google's VEO connects to a wider toolkit (Gemini integration, API access, Flow app), and Meta inevitably embeds AI video throughout the product.
Openai can be distinguished by making Sora 2 a seamless ChatGPT feature. Applying this approach to SORA 2 will instantly provide AI video studios without switching apps for millions of ChatGPT users. They can follow Google and have a low limit on videos per day using Unlimited Access's premium plan.
Mobile optimization is extremely important. Today's creators are filming, editing and posting entirely from their mobile phones. If SORA 2 has the ability to quickly generate within ChatGPT's mobile app (or dedicated SORA app), you can capture Tiktok and Reels Creator Market. Imagine your phone instantly, “Just Chat Gupto, making my 15-second video as a cartoon astronaut landing on Mars.”
By making SORA 2 ubiquitous, OpenAI can quickly build a user base while gathering essential improvement feedback through ChatGPT, the developer API and mobile platforms.
Platforms like Leonardo, Freepik and Higgsfield already use Google's VEO 3 and Hailuo's Minimax 2. This is impressive and fast available via the API, so Openai is lagging behind the creative AI space by not updating SORA.
Conclusion
Openai has a real opportunity to regain leadership by learning from the successes of its competitors. Google's VEO 3 currently benchmarks with native audio, realistic physics and powerful rapid adherence, but newer models like the Kling 2.1 and Minimax 2 keep pushing the boundaries.
The runway has more features, along with new improvements to the Gen-4 model, which is similar in quality to physics and Sora, with others like Pika focusing on the creator market, further pushing Openai out of its precious space.
SORA 2 cannot be just an incremental upgrade. You need to be surprised.
Encouraging news? Openai has basic elements such as a powerful language model, first-generation video model to build, and a large user base for ChatGpt. If Openai offers native sound generation, realistic physics, dialogue ease of use, character consistency, and seamless product integration, SORA 2 could have broken the entire field very well with VEO 3, Kling, and its own game.
If it all comes together, don't be surprised if the next virus AI video in the feed is created in SORA 2.
