Get FLUX fit - generative AI for retail
Image generation with AI seems to be maturing in ‘fits and starts’ and the release of the new FLUX range of models in August kick-started a fascinating new era.
The release of new models like FLUX and Leonardo.AI’s Phoenix unlock new potential for retailers and brands to use AI to visualise their products. Although many of these concepts and techniques are no longer novel, the quality of output is.
We have been doing some benchmarking of before / after to show the changes in the past 6 months, and wanted to share the latest fit-for-purpose techniques of generative AI for product image creation and manipulation. But before we get started some exciting news…
Time Under Tension are finalists in two categories in the upcoming 2024 B&T Awards. We’re proud to be shortlisted for Emerging Agency and Marketing Technology Company of the Year.
Why? We are helping some of Australia’s leading brands to understand and unlock the potential for generative AI, through a combination of inspirational content, education & training, and gen AI product development. We would love to help you too.
Now, back to some gen AI techniques you can use for retail;
Text-to-image
The major change in traditional text-to-image generation is the quality of the models. Midjourney 6.1, Leonardo.AI Phoenix, Google Imagen 3 and FLUX have all greatly improved the realism of AI generated images.
As an example, below on the left is a portrait ‘photo’ using Midjourney 5.2 (about a year old) compared to a recent image created by the incredible AI artist Roope Rainisto. A year ago the image on the left was impressive, now it pales in comparison to what is possible with the latest AI models in the hands of a talented artist;
And another AI image showcasing the realism possible with FLUX;
Fine-tuning
Fine-tuning is the method of training an image generation model with either a style or an object (such as a product), and this is where the greatest progress has been made for retailers and brands.
In particular, the new FLUX model performs incredibly well with fine-tuning. Here are some examples, the first from EverArt (a platform that allows you to create your own fine-tunes) showing how clothing items can be fine-tuned and then images generated of the items being worn by AI models.
This next example is from Swedish Art Director Boris Noll showing how accurate details can be with intensive fine-tuning on FLUX.
And some tests from me using FLUX fine-tuning on EverArt (more here):
And finally, Coca-Cola are using NVIDIA Omniverse and NVIDIA NIM microservices to create on-brand AI images for different global markets.
Reference images for product ideation
All the the major image generation apps have added or improved the ability to upload a ‘reference image’, which can be great for mixing and matching styles and generating new concept ideas.
For example, in Leonardo.AI you can create stunning images by transferring the shape and structure from one image to another using the Content Reference feature.
The new(ish) Midjourney Web interface now makes this easier than ever. This is possible in the Discord app, but not very user-friendly.
Background removal & generation
Slightly less exciting, but incredibly practical is the use of AI for post-production editing of images at scale.
We have just completed a project for the furniture brand KING Living, building a Web service that takes raw iPhone photos from the warehouse, and uses AI to turn the photo into the style and quality expected from a studio shoot. The automated process involves background removal, background generation with lighting and shadows, plus resizing.
What’s next?
There are interesting developments in virtual try-on, face-swapping for retail and 3D product imagery, most of which are currently in R&D stages.
Ready to get your product images AI fit?
If you are interested to learn more about how the latest gen AI tools could be used to visualise your products, please get in touch.
Further reading:
Hire Time Under Tension
We work with agencies, companies and brands to elevate your Customer & Employee experience with generative AI. Our advisory team help you to understand what is possible, and how it relates to your business. We provide training for you to get the most of generative AI apps such as ChatGPT and Midjourney. Our design & technology team build bespoke gen AI tools to meet your needs. You can reach us here: www.timeundertension.ai/contact
A handful of Gen AI news
Here are five interesting things we saw in the past week;
From PDF to Podcast, with the new Google NotebookLM
RunwayML released Gen-3 Alpha, a new video-to-video model;
Released on 5th September was the Aus Govt. Voluntary AI Safety Standard, a guide to safe and responsible use of artificial intelligence in Australia
OpenAI launched o1-preview, a new series of AI models designed for enhanced reasoning capabilities in complex tasks across science, coding, and math
Everies is a super cute AI+AR app that turns everyday's items into interactive characters, built with Google Gemini;
Thanks for reading, please share this with a friend if you liked it, and leave a comment with anything you wish to add.