Skip to main content
All CollectionsCopyright and AI model
Photoroom AI model and data diversity
Photoroom AI model and data diversity

Learn about the Photoroom AI, including its capabilities, training and data diversity.

C
Written by Craig Wright
Updated over 5 months ago

What's different about the Photoroom AI model?

Photoroom's proprietary AI model is tailor-made for photography, especially product images. It is more powerful and versatile than our previous AI model and offers:

  • New and improved AI features: We have improved AI backgrounds and have a range of new AI features, including the ability to generate entire AI images and to use AI to enhance a specific area of a design. With this level of control, you can quickly turn your creative ideas into impressive and appealing designs.

  • Higher quality: The AI model can produce more realistic, detailed, and higher resolution product photos for your web site, product listings, and social media.

  • More flexibility: You can expand your image beyond the borders and adapt an image size to your needs. Ideal for promoting your products on different social media platforms.

  • More data diversity: Photoroom AI is trained using a diverse range of images so that it reflects our global audience (see Tackling bias in the AI). It also prevents users from creating explicit or violent images (see No explicit or violent images).


How do we train the Photoroom AI?

We use the following to train Photoroom's AI:

  • Open license images and images

  • Images purchased from our partners for the explicit intent of AI training.

We do not use your data or any other customer data to train the AI.


Tackling Bias in the AI

Photoroom has a diverse range of users in over 180 countries and we recognize the importance of tackling bias. We want to make sure that everyone feels included and represented.

Team diversity

We believe in having a diverse and inclusive team that can identify and challenge any bias in our dataset.

Bias in the dataset

Most AI models use datasets that are biased from the start. Because the data is created by humans and so reflects the bias in our society.

To challenge this bias, we work with a diverse range of image providers and train our AI model using images from around the world. This means our AI model offers a better reflection of our global audience.

Better feedback loops

Any dataset will have some degree of bias, which is why it's important that we have strong feedback loops. These allow us to react quickly to feedback and make meaningful changes to our AI model.


No explicit or violent images

We have designed the Photoroom AI so that it cannot generate violent or sexually explicit imagery. The prompts will prevent you from requesting those types of images. Also, the dataset does not contain any images of this type, so Photoroom cannot generate them.


Did this answer your question?