How to create your next product photos using AI

New AI driven tools can create stunning product photography, and it only takes a few seconds.

Read on to find out how to do it yourself.

New Developments in AI

In August 2022 stability.ai released Stable Diffusion, a new open-source AI model that can generate beautiful images from simple text prompts. This is the latest development in the field known as "Generative AI" - using AI to generate new creative content like text, image and audio from scratch.

These new AI tools can be used to generate product photography without ever picking up a camera, and the quality is impressive.

How it works

First, take an existing image of your product from an old photoshoot. It's okay if it has a background already, we can remove it.

Second, pick a "style" image that will tell our AI what kind of style you're looking for. It's okay if the style image contains a different product.

Here's an example (this took about 12 seconds to generate):

Finally, you sort through the generated images to find one that you like.

Traditional photoshoots suck

The main benefit of these new tools is that you don't have to organize a traditional photoshoot to get the new creative you need (e.g. for your social media posts, newsletter, or that upcoming holiday sale). You don't need to talk to the photographer, send product samples, pick locations, pay high costs, and you don't need to wait days for the results.

If you run a small e-commerce business that's a lot of time, money, and effort you can spend on more important things.

Split testing and rapid iteration

One of the "killer features" of this new AI tech is it makes split testing photos much easier.

With traditional methods it's hard to get a lot of diverse images of your product to test. Your photographer can slightly change the angles and lighting, but what if you want to photograph your product at a Parisian Café? You better send your photographer to Paris. Or have them spend a few hours in photoshop.

With generative AI you can quickly generate and test dozens of unique photos. Put them out there and measure the results.

Once you've found the images that sell your product you can ask your photographer to re-create them if you feel that will be beneficial (at the moment, photography still produce slightly higher quality images than AI can do).

The AI doesn't always understand what it is supposed to do with what you've given it. So, the AI creates a lot of images that seem reasonable to it, but are not usable as product photos.

If you want to work with generative AI you need to sort through a lot of these bad images to find the few good ones.

We think the tools that eventually succeed in this space will be the ones that give the AI a better understanding of what you want, which will in turn improve the ratio of good images to bad.

Some common problems we see in the "bad" images are:

• Photos that don't depict your product accurately
• Photos that are look "off", for example sometimes the AI will generate hands with the wrong number of fingers
• Photos that don't match your brand styling

Here are some examples of bad images

In our testing we found that the "bad" images are usually very obvious and can be skipped quickly, but you must be aware of them.

What about text?

There are two different methods for generating product photographs. Every tool we've seen uses one or the other.

The first method is called fine tuning (or "DreamBooth"). You give the AI model about 5 images of your product and let it train for up to 30 minutes. The model can then generate completely new views of the product in new contexts. For example, it could generate a picture of sunglasses on a person's head. The biggest drawback of this technique is that current AI models can't create readable text. If your product has any writing or labels on it the text will become unreadable.

The second method is called inpainting. You supply a single image of your product, and the AI model replaces the background and blends the product into the scene naturally (including creating realistic shadows). Your labels and logos are perfectly preserved.

Both methods have a place but we prefer the inpainting method for most applications.

Want to try it out today?

We created photoshoot.ai for exactly that purpose. You can try it for free with your own products in just a few seconds and find the photos that will sell your products. 

If we can't solve your problem, here are a few of the other services we're aware of that might be able to help:

• Vue.ai - Focused on clothing
• spyne.ai - Focused on automotive
• DTC Forever
• osmosis.studio
• PhotoRoom.com

(We are not affiliated with any of these other services)

This is only the beginning

This technology is useful today, and will be ubiquitous in the future. We think the early adopters stand to profit greatly from this trend because they can produce product photography that their competitors can't match using traditional techniques.