The advancements in AI video generation have been truly remarkable. One of the most impressive capabilities is the ability to maintain object permanence and persistence through occlusion. This means that the AI models can remember the appearance and position of objects, even when they are temporarily obscured or hidden from view.

For example, the video shows a shot of a burning building reflected in an eye. When the eye closes, the reflection of the building remains exactly as it was before, demonstrating the AI's ability to maintain the consistency of the scene. Similarly, in another shot, a clown waiting for a bus is temporarily obscured by a passing car, but the clown is still present when the car has passed.

These feats of object permanence and persistence were unimaginable just a few months ago, but the leading AI models, such as Runway Gen 4, Cling V2, and Higsfield, have now mastered these capabilities. The video showcases how the AI can seamlessly handle these complex scenarios, creating a sense of realism and coherence that was previously unattainable.

All the tools have gotten much better with emotions too. The generators are now able to showcase a range of emotions, with characters maintaining consistent expressions throughout the shots. Character persistence has also improved significantly, with the faces not changing into different people during the course of a shot, which used to be a common issue.

We can see examples of multiple characters laughing together, with their facial expressions and emotions remaining coherent. Even shots with a lot of movement, such as hands playing the piano, can now be generated with a high level of detail and realism.

The ability to combine different styles, like a realistic restaurant setting with a 2D cartoon waitress, further demonstrates the versatility and capabilities of these AI video generators. The smooth transitions and coherence between the various elements are quite impressive.

AI video generation has made remarkable progress, with models like Runway Gen 4, Cling V2, and Higsfield showcasing impressive capabilities. These tools can now handle complex tasks that were once unimaginable, such as maintaining object permanence, conveying emotions, and ensuring character persistence.

The examples highlighted demonstrate the versatility of these AI models. They can seamlessly blend realistic and stylized elements, as seen in the shot featuring a realistic restaurant setting with a 2D cartoon waitress. Additionally, the models excel at capturing diverse movements, from a paper airplane flying out a window to a focus shift between foreground and background characters.

Higsfield's preset shot types, such as the whip pan and action shot, have proven particularly effective in achieving complex camera movements and dynamic action sequences that were previously challenging for other generators. The ability to selectively color elements, like the woman's red lipstick, further showcases the models' precision and attention to detail.

While some tasks, like knocking down dominoes or realistic juggling, still pose difficulties, the AI models are continuously improving, and with some creative editing, these limitations can often be overcome. The examples of surreal and visually stunning shots from creators like Uncanny Harry, the Runway CEO, and Sway Molina demonstrate the immense potential of AI-generated video content.

As the technology continues to advance, the boundaries of what is possible with AI video generation will continue to expand, allowing creators to push the limits of their imagination and bring their visions to life in increasingly seamless and captivating ways.

The leading AI video generators have made significant advancements in their capabilities, allowing for the creation of complex camera movements, dynamic action sequences, and selective coloring effects that were previously unimaginable.

One impressive example is the whip pan shot, which was achieved using the new preset options in Higsfield's AI model. While the initial whip pan was a bit slow, it could be easily sped up in post-production, resulting in a smooth and seamless transition between the two scenes.

Similarly, the action shot preset in Higsfield was able to capture the fast-moving action of a woman running from a T-Rex with impressive results. The generator was also able to handle selective coloring, where a scene is predominantly black and white, with the exception of the subject's red lipstick, which was added convincingly.

Another noteworthy achievement was the AI's ability to accurately depict the action of blowing out candles, which was particularly challenging for the other models but was executed well by Runway Gen 4 after some rerolls.

While not every attempt was perfect, the AI models demonstrated their growing capabilities in handling complex camera movements, dynamic action sequences, and selective coloring effects, showcasing the rapid advancements in this field.

While AI video generators have made impressive strides, there are still some complex tasks that remain challenging for the current models. One such example is the attempt to have the AI generate a video of dominoes being knocked down. The creator tried to make the scene as simple as possible, but even the best-performing model, Runway Gen 4, was unable to perfectly recreate the physics and movement required for the dominoes to tip over correctly. The creator had to keep rerolling the generation to get a passable result, but it was still not quite perfect.

Another complex task that proved difficult for the AI models was generating a breakdancing sequence. The creator was surprised that the models were able to get somewhat close to this, with the best result being a quick cut to the best moment of the breakdancing. However, the overall sequence was not fully convincing.

Similarly, the attempt to have the AI generate a shot of someone tossing and catching their keys proved challenging. While the models were able to get close enough that the shot could be salvaged with some editing, any more complex throwing motions, such as shooting a basketball into a hoop, were almost impossible to achieve across all the tested models. The best result was still quite far from a realistic representation of the action.

These examples highlight the current limitations of AI video generators, even as they continue to push the boundaries of what is possible. The models struggle with precise, complex movements and physics-based interactions, particularly when involving multiple characters or actions. While the technology is rapidly advancing, there is still room for improvement in these areas.

AI video generation has come a long way, with new updates like Runway Gen 4 and Higsfield's camera controls enabling a wide range of complex movements, physics, and styles. The author has pushed various video generators to their limits with challenging prompts to explore their capabilities.

Some of the impressive feats achieved include:

• Object Permanence: The ability to maintain the appearance of objects even when they are occluded, as demonstrated by the reflection of a burning building in an eye that remains consistent even when the eye closes.
• Emotional Expressions: Generators have become much better at depicting realistic emotions and maintaining character persistence, with examples of multiple characters laughing together.
• Diverse Styles: The ability to combine different styles, such as a realistic restaurant setting with a 2D cartoon waitress.
• Camera Movements: Higsfield's new update introduced preset shot types, including a successful whip pan and an "action shot" for fast-moving scenes.
• Selective Coloring: Gen 4 was able to selectively color the lipstick of a character, while keeping the rest of the image in black and white.
• Text-to-Video: V2 demonstrated impressive cursive handwriting when generating text-to-video content.

However, the author also encountered some limitations, such as struggles with complex physics-based actions like knocking down dominoes or juggling. Precise movements and multi-character interactions also proved challenging for the current models.

Despite these limitations, the author showcases several impressive examples from various creators, demonstrating the rapid progress in AI-generated video capabilities, including bullet-time effects, realistic physics simulations, and fully animated short films.

The advancements in AI video generation have been truly remarkable. We've seen how the leading models, such as Runway Gen 4, Cling V2, and Higsfield, have overcome many of the previous limitations, delivering impressive results in areas like object permanence, emotions, character persistence, and complex camera movements.

While there are still challenges with certain advanced techniques like precise movements, multi-character interactions, and physics-based actions, the capabilities of these AI video generators continue to expand rapidly. The ability to create entire short films, indistinguishable from reality, is a testament to the incredible progress in this field.

As the technology continues to evolve, the potential for AI-generated video content to revolutionize various industries, from entertainment to marketing, is undeniable. The future holds exciting possibilities, and it will be fascinating to see how these tools continue to push the boundaries of what's possible in the world of video creation.