Rise of the machines – AI content creation

Rise of the machines - AI content creation

Trender recently deployed an internal Stable Diffusion instance to streamline our content creation projects and so far the results have been impressive.

Running on modest consumer level GPU compute hardware, it’s quickly becoming apparent we’re on the cusp of a revolution with regard to how we create content for both online applications and marketing material.

For those not familiar with the current state of play in the AI space, or more precisely the image generation (or image synthesis) space, Stable Diffusion is a tool that allows you to create completely original image content from a text prompt. 

With a varied selection of pre-trained models available, Stable Diffusion can create anything from highly detailed real world imagery to fantasy, cartoon, vector or icon based images and a huge variety of art styles in between. 

The image content created is based on a random initial seed number, so the same text prompt will result in a unique image each time it is run.

The examples below were generated in approx. 30 seconds and were the result of the text prompts displayed above them:

Dinner table with christmas tree in background presents under tree:

Boat in lake with a forest in the background blue sky and mountains oil painting style:

A diffusion network consists of forward and reverse modes of operation. The network gradually adds noise to an image in a forward direction and then removes it gradually until the image is restored based on a number of predictions.

Client Benefits

The use of this technology has an immense impact on the time allocated to image content creation. 

Project budgets can benefit from image synthesis as well by devoting some of the allotted monetary resources to other areas in preference to digital rights holders.   

Brochure sites, (websites that are heavily image based) can enjoy very rapid deployment to production with image content being created in a matter of minutes in preference to days or weeks.

Moving forward, several other image diffusion models will be appearing over the coming months and Trender will be investing heavily in local versions of this technology to drive our content creation processes.