Sora 2 vs Veo 3.1 vs Kling 2.5 vs Seedance 1.0: What Does Physics Best?

Greg Preece
15 Nov 202507:21

TLDRThis video compares the physics performance of four AI models: Sora 2, Veo 3.1, Kling Physics model comparison2.5, and Seedance 1.0. It discusses the strengths and weaknesses of each, highlighting Veo 3.1 and Kling 2.5 as top contenders, with Veo being preferred for its sound generation. The video also touches on the slow generation time of Sora 2 and the numerous mistakes made by Seedance. Viewers are encouraged to comment on which model they think performs best and suggest future comparison videos. The video concludes with a call to explore a tutorial on generating clips and turning them into long films.

Takeaways

  • 😀 The video compares four AI models—Sora 2 is an advanced AI model, and developers can access its capabilities through the Sora 2 API., Veo 3.1, Kling 2.5, and Seedance 1.0—based on their ability to generate accurate physics simulations.
  • 🎬 Sora 2 has a censorship policy that restricts the use of human images, which led to some clips looking different compared to others.
  • ⏱ Sora 2 is slower to generate clips, taking 5 to 10 times longer than the other models.
  • 🔊 Veo 3.1 and Kling 2.5 were rated as the best models for physics, with Veo 3.1 having the edge due to its ability to generate videos with sound.
  • ⚠️ Seedance 1.0 was found to make numerous mistakes in its generated clips, particularly in terms of physics accuracy.
  • 💻 The creator used an all-in-one website to generate clips quickly, which helped reduce the cost of video production with a 20% discount.
  • 💡 The creator suggests making comparisons between other models, such as sports or exercise body movements, in future videos.
  • 🖥 A tutorial is available for turning short clips into long-form films using the same platform, which may be helpful for creators.
  • 👥 There are links in the video description to all the prompts used to generate the clips, as well as the website used for generation.
  • 📢 The creator encourages viewers to leave comments on which model they think does physics best and suggests future content ideas.

Q & A

  • What is the main focus of the video titled 'Sora 2 vs Veo 3.1 vsPhysics model comparison Kling 2.5 vs Seedance 1.0: What Does Physics Best?'

    -The video compares the performance of different AI models (Sora 2, Veo 3.1, Kling 2.5, and Seedance 1.0) in generating realistic physics simulations, evaluating which model handles physics the best.

  • Why do some clips in the video look different from others?

    -Some clips look different because Sora 2's censorship policy prevented the use of personal images for scenes involving people. These clips were generated entirely from text descriptions, while others used a personal image as the base.

  • What was the biggest drawback of using the Sora 2 model for video generation?

    -The main issue with Sora 2 was that it took significantly longer (5 to 10 times longer) to generate clips compared to other models, making the process inefficient.

  • Which models performed the best in terms of physics simulations?

    -Veo 3.1 and Kling 2.5 were the top performers in physics simulations, with Veo 3.1 being slightly preferred due to its ability to generate videos with sound.

  • What was the biggest flaw with Seedance 1.0Physics model comparison?

    -Seedance 1.0 made numerous mistakes in its video generations, which were easy to miss on first viewing but quite apparent upon closer inspection.

  • Why does the creator prefer Veo 3.1 over Kling 2.5?

    -Veo 3.1 is preferred because, unlike Kling 2.5, it can generate videos with sound, which enhances the overall viewing experience, even though both models were nearly equal in terms of physics performance.

  • What platform did the creator use to quickly generate the clips?

    -The creator used an all-in-one website to quickly generate the clips, which offered a 20% discount at the time, making it one of the cheapest options for generating such content.

  • What future video ideas does the creator mention?

    -The creator suggests making a sports comparison or an exercise body movement comparison video, where different models could be tested to see how they handle specific physical scenarios.

  • How does the creator suggest turning short clips into longer films?

    -The creator recommends watching a tutorial on how to quickly generate clips and compile them into longer films using the same platform, or exploring advanced tools like the Sora 2 Pro API for enhanced capabilities.

  • What kind of clips does the creator suggest to generate for future comparisons?

    -The creator is open to suggestions and asks viewers to comment on what comparisons they'd like to see in future videos, such as testing different AI models for other types of simulations or activities.

Outlines

00:00

🔥 Heat, Physics Comparisons & Sora 2's Influence

In this paragraph, the speaker introduces a video comparing different physics models. They mention that all the prompts used to generate the clips are available in a Google document, linked in the description. The speaker explains the visual differences between clips, citing Sora 2's censorship policy as a factor that prevented them from using their own image in certain clips. While the clips with people were generated entirely from text prompts, the clips without people used the speaker's own image as a base. The speaker invites viewers to comment on which model they think performs best in terms of physics and promises to give their own verdict later in the video.

05:01

🎬 Model Comparisons, Cling 2.5 vs VO 3.1, and Future Video Ideas

In this paragraph, the speaker discusses the performance of different models for generating clips. They mention VO 3.1 and Cling 2.5 as the top two contenders, with VO being their personal favorite because it generates clips with sound, unlike Cling. The speaker also mentions that Sora takes significantly longer to create clips, and Seed dance made numerous mistakes, which viewers can spot by revisiting the video. The speaker solicits suggestions for future comparisons, including a potential sports or body movement video.Physics model comparison They also mention a tutorial for quickly generating clips and turning them into longer films, encouraging viewers to watch that video next. The paragraph ends with a call to subscribe to the channel.

Mindmap

Keywords

💡SJSON error correctionora 2

Sora 2 is a model used for generating video clips. In the video, the creator highlights Sora 2's censorship policy, which restricts the use of personal images in videos involving people. This results in the clips with people appearing different, as they must be generated entirely from text prompts, unlike other models that can use personal images as starting points. The video mentions that Sora 2 is slow in generating clips, taking much longer than other models like Veo 3.1 and Kling 2.5.

💡Veo 3.1

Veo 3.1 is another video generation model discussed in the video. It stands out for being one of the top contenders for accurately simulating physics in the clips. The creator notes that Veo 3.1 is preferred because it generates clips with sound, a feature that sets it apart from Kling 2.5. Despite both Veo 3.1 and Kling 2.5 being neck and neck in terms of physics, the sound capability of Veo makes it the best in the creator’s opinion.

💡Kling 2.5

Kling 2.5 is a video generation model that performs well in terms of simulating physics, as mentioned in the video. The creator compares Kling 2.5 to Veo 3.1, noting that both models are very similar in their physics generationJSON code correction, but Kling lacks the ability to generate sound. This makes Veo 3.1 the more preferred option overall, even though Kling 2.5 is still regarded as one of the top performers in terms of physics.

💡Seedance 1.0

Seedance 1.0 is another model discussed in the video, but it is critiqued heavily for making numerous mistakes. The creator explains that Seedance 1.0's flaws are subtle and easy to overlook on first viewing, but when observed closely, they become apparent. The creator implies that Seedance 1.0 doesn't perform well compared to the other models like Veo 3.1 and Kling 2.5, making it the least favorable choice.

💡Physics

Physics is a central theme in the video, where the creator compares how each model simulates physical interactions. The key question posed is which model handles physics the best, with particular focus on the generation of natural movements, realistic dynamics, and physical behaviors in the clips. The creator observes that both Veo 3.1 and Kling 2.5 excel in this area, but Veo 3.1 takes the lead due to its ability to generate sound, which adds another layer of realism to the generated clips.

💡Censorship policy

Sora 2's censorship policy is mentioned in the video as a restriction that impacts the way videos are generated. This policy prevents the use of personal images in videos that include people, which affects the visual consistency of the clips. The creator notes that this censorship results in clips with people looking different, as they must be generated from scratch rather than using real images as a base.

💡All-in-one website

The creator mentions using an all-in-one website to quickly generate video clips. This platform allows for the easy creation of multiple clips using the discussed models, with a special mention of a 20% discount that makes it a cost-effective choice. This tool is described as a way to streamline the process of generating content, especially when comparing models or creating a lot of clips in a short amount of time.

💡Sound generation

Sound generation is a feature highlighted in the video, specifically in reference to Veo 3.1. Unlike Kling 2.5, which does not generate sound, Veo 3.1's ability to incorporate sound is seen as a significant advantage. This feature enhances the realism of the generated clips, particularly when considering physical simulations that might benefit from sound effects like movement noises or environmental sounds.

💡Model comparison

Model comparison is the overall theme of the video. The creator is comparing different video generation models, specifically focusing on how they simulate physics, the speed of generation, and additional features like sound. By discussing the pros and cons of models like Sora 2, Veo 3.1, Kling 2.5, and Seedance 1.0, the creator aims to determine which one performs the best across different criteria, offering insights for viewers who may want to use these models.

💡Clip generation

Clip generation refers to the process of using AI models to create video content. In the context of the video, the creator uses this process to test how well each model simulates physics, handles sound, and generates realistic imagery. The speed and efficiency of generating clips are also important factors, with some models like Sora 2 being slower compared to others like Veo 3.1, which are able to produce clips much faster.

Highlights

The comparison between Sora 2, Veo 3.1, Kling 2.5, and Seedance 1.0 focuses on physics simulation capabilities.

Sora 2 has a censorship policy that led to different visual styles for clips with people in them.

Clips without people were generated using the user's own image as the starting frame for the models.

The video creator evaluates which model does physics best, with Veo 3.1 and Kling 2.5 being top contenders.

Veo 3.1 is preferred for generating videos with sound, which gives it an edge over Kling 2.5.

Sora 2 takes significantly longer to generate clips compared to other models, often 5-10 times slower.

Seedance 1.0 was criticized for making numerous mistakes in its generated clips, especially regarding physics.

The creator encourages viewers to choose the best model for physics and leave feedback in the comments.

A link to a platform that allows fast clip generation with a 20% discount is provided in the video description.

Future comparisons might include sports or exercise body movementPhysics model comparison videos to test model performance.

The creator offers a tutorial on how to quickly generate clips and turn them into long films using these tools.

Sora 2’s slow processing time was a major drawback, especially when compared to other faster models.

The creator expresses frustration with Seedance 1.0's lack of accuracy and frequent errors in generated content.

Veo 3.1 and Kling 2.5 are the top models for physics simulations, with similar results, but Veo's ability to include sound gives it an edge.

The video ends with a call to subscribe for more content and a teaser about future comparison videos.