AI Video for animating still product images

Last week, I had the pleasure of experimenting with various image-to-video models. Honestly, I did not have much interest in video models prior to my experimentation, but I’m glad that I had the opportunity to do this, because - WOW. These models are GOOD. I tried a variety of models, and I think there are some obvious tradeoffs between them in terms of quality and speed, so let’s dive into some of these tools and their outputs.

Task

We’re going to be experimenting with three different models:

  1. Luma Dream Machine

  2. Hailuo AI (Minimax)

  3. Kling AI

Luma Dream Machine

Hailuo AI (Minimax)

Kling AI

For the input, we will use these still product images. The aim is to use AI to convert them to videos, to make them more dynamic and life-like.

Philips Air Fryer 5000

Breville AquaStation Hot

For each image, we’ll try all three video models with similar prompts and compare their outputs.

Image 1 - Philips Air Fryer

Philips Air Fryer 5000

For this image, I wanted to animate the plants in the background as if they were swaying in the wind. This is the prompt I used:

“Plants swaying in the wind in the background”

Let’s see how these tools perform.

Hailuo (Minimax)

Hailuo followed the prompt really well, animating the plants in the background and also the vegetable in the foreground. The oil bottle seems to be feeling some of the wind too!

Luma Dream Machine

Luma seems to have ignored the prompt entirely and opted for a simple camera pan. Even after reworking the prompt, it still produced the same result. But I would say this is still a decent output as it displays the product while adding some liveliness to it.

Kling AI

Kling also opted for the camera pan instead of actually following the prompt. You may notice that the numbers on the machine actually look a bit haywire, while Luma was able to maintain it quite well.

Kling also has a feature called “motion brushes”, where you can hand-select a part of the image and instruct the model to make it stationary or move in a certain path. While this feature has worked for me in the past, for this image the result was not ideal.

Image 2 - Breville AquaStation

Breville AquaStation Hot

Again, for this image, I wanted to create a dynamic scene by animating the plant, and maybe panning over the product.

Let’s see how the different tools performed.

Hailuo (Minimax)

For this run, I actually didn’t give Hailuo a prompt - I wanted to see what it would do on its own - and it seems to have animated a quick coffee brew!

Luma Dream Machine

Luma utilized a camera pan to the left, further revealing the plant and a bowl of limes. I tried getting the model to pan right, but it still insisted on going left.

Kling AI

I would say Kling performed the best on this image. The slightly shaky camera movement gives a lively feel to the video, and the plant swaying in the wind is a nice touch. However, you can see that the machine itself, specifically the buttons, are a little blurry.

Comparison

These experiments give a taste of the functionality of some of the latest video models. My own experimentation has led to draw these conclusions about them:

  • Luma Dream Machine offers the best balance of speed and quality

  • Hailuo (Minimax) excels at simple tasks and provides excellent quality, but takes longer than the others

  • Kling is great for life-like animations with a lot of movement, and it also has the “motion brushes” feature which makes prompt engineering even easier

For the experiments in this post, I used relatively simple prompts. You can expect to get better and more refined results if you engineer your prompts effectively.

Conclusion

Trying out these image-to-video models has been a great experience. Each tool has its own strengths and weaknesses, making them useful for different types of projects. Luma Dream Machine is a solid choice if you want quick and good-quality results. Hailuo (Minimax) is great for simple tasks with amazing quality, but it takes a bit more time. Kling AI is perfect for adding realistic animations and has helpful features like motion brushes, though it can be a bit tricky at times.

In the end, the best model depends on what you need for your project, whether you're making product videos for an e-commerce site, creating animated scenes, or just exploring what these tools can do. This experience has shown me how exciting video models can be, and I’m looking forward to seeing how they improve in the future!

Previous
Previous

The 3 biggest underestimated factors of generative AI

Next
Next

Product imagery API for King Living