
The rapid advancement of image-to-video generative AI models offers powerful new opportunities for storytellers, filmmakers, and animators. In this post, we'll provide a practical comparison of three notable models from Kling AI, Luma AI, and Runway——to help you identify each model's strengths, weaknesses, and ideal use cases.

<div class="video-container" style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden; max-width: 100%; margin: 2rem 0;">
  <iframe style="position: absolute; top: 0; left: 0; width: 100%; height: 100%;" src="https://www.youtube.com/embed/7lg7S2nz3KM?rel=0" title="Comparing Image-to-Video Models" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div>

### Overview

We've selected **Kling 1.6 (Standard)** and **Luma Ray2 Flash**, as run on <a href="https://fal.ai" target="_blank">Fal</a>, as well as **Runway Gen-3 Alpha Turbo** as operated directly by <a href="https://runwayml.com/" target="_blank">Runway</a>. These are all accessed by API, using the Possible Studio's interface, with image inputs, all generated by the Flux Dev or Flux Schnell generative image models. These video models balance strong performance on dimensions creators care about, while being reasonably efficient and fast to run.

Our evaluation covers key attributes creators care about, such as physics, behavior, and coherence, using a variety of benchmark scenarios ranging from realistic live-action scenes to vivid animations. This is intended as a practical overview rather than a deeply technical breakdown, making it easier to choose the right model for your storytelling needs.

### Evaluation Criteria

We'll assess each model based on the following categories:

- **Physics**: Including object permanence, fire and particles, liquid dynamics, and animation physics.
- **Behavior**: Focusing on people, animals, character animation, and detailed actions (like hands and cooking).
- **Coherence**: Emphasizing consistency in style, sci-fi and fantasy scenarios, and natural phenomena.

### Model Performance Comparison

#### Physics

All models handle fire effects reasonably well. However, they struggle noticeably with object physics, especially in sports and animated sequences. For instance, skateboard scenes are amusingly unrealistic across all three.

<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-skateboarder.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

- **Liquid Dynamics**: Luma excels, accurately simulating a wine pour, though Kling shows color inconsistencies. Runway struggles significantly, with water moving through glass.

<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-winepour.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>

#### Behavior

Character behavior and realism vary across the models:

- **Human Interaction**: Kling is strongest in producing believable reactions and expressions. Luma shows uncontrolled motion at times, while Runway often defaults to overly dramatic, slow-motion shots with uncanny effects.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-womansurprise.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

- **Animals**: Kling and Luma both handle animals well, whereas Runway tends to produce visible artifacts, such as turning dog whiskers into unrealistic water streams.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-golden.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

- **Machines**: Luma performs best with mechanical objects like cars, while Kling is acceptable. Runway notably struggles, sometimes animating scenes with incorrect directional movement.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-redconvertible.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>

#### Coherence

Consistency and style coherence significantly differ between models:

- **Object Permanence**: Runway faces substantial challenges, causing unrealistic appearances and disappearances of characters. Kling is adequate, though still imperfect, while Luma occasionally introduces noticeable artifacts.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-animedinner.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-airballet.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

- **Stylistic Consistency**: Runway struggles most in this campfire scene, particularly with maintaining black-and-white styling, inadvertently introducing color. Kling completely colorizes the shot, but Luma performs well.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-campfire.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

- **Fantasy and Natural Phenomena**: All three handle fantasy scenarios well, with Runway uniquely excelling in dynamic wizard animations and holographic effects. Kling and Luma convincingly animate natural phenomena like auroras, while Runway's results are less realistic.
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-bluewizard.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-holographic.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />
<video autoPlay loop muted playsInline style="max-width: 100%; height: auto;">
  <source src="/blog/media/watermarked-aurora.mp4" type="video/mp4" />
  Your browser does not support the video tag.
</video>
<br />

### Summary of Results

| Criterion       | Kling 1.6 (Standard) | Luma Ray2 Flash  | Runway Gen-3 Alpha Turbo |
| --------------- | -------------------- | ---------------- | ------------------------ |
| Physics         | Good, minor issues   | Very good        | Struggles significantly  |
| Behavior        | Best overall         | Mixed            | Often dramatic/artifacts |
| Coherence       | Consistently good    | Generally good   | Frequent inconsistencies |
| Rendering Speed | ~3 min per shot      | ~40 sec per shot | ~40 sec per shot         |

### Conclusion: Choosing the Right Model

Selecting the right generative video model depends heavily on your specific project needs:

- **Kling 1.6** offers reliable overall performance, particularly excelling in behavioral realism, making it ideal for character-driven storytelling. Takes the longest.
- **Luma Ray2 Flash** shines in speed and accurate physics rendering, suitable for detailed, fast-turnaround projects where fluid and mechanical accuracy are crucial. Relatively fast.
- **Runway Gen-3 Alpha Turbo** provides distinctive results with stylistic flair, best suited for experimental and fantasy-oriented animations despite its frequent inconsistencies.

We recommend experimenting with each model to get a feel for how they fit your project's needs and aesthetics. The evolving landscape of generative video models continues to offer exciting possibilities, and these tools promise to become even more powerful for creators in the near future.

We're always on the lookout to incorporate promising models into the [Possible Studio](https://app.thepossible.io). If there's a model you like that is missing, let us know!
