Historically, marketers have found more success with AI image editing than with AI image generation. The reason: it’s been easier for them to start with an on-brand image, which can then be modified (changing dimensions, removing backgrounds, replacing text, etc.) versus trying to generate on-brand images at scale. Even with robust prompting, AI Image generation in its current state tends to produce a variety of image style outputs — not an ideal scenario if you’re a marketer trying to work within strict brand guidelines. 

There’s been a lot of research on AI image generation consistency, but most of that research doesn't consider the nuance of enterprise brands.

The Jasper Research team has been hard at work researching ways for marketers to jump this hurdle; to be able to generate on-brand AI images at scale. Specifically, they’ve focused on improving the current models and evaluating both the available and custom methods on enterprise-style images—and have seen promising results.

These efforts aim to help marketers spend less time on design review and editing AI-generated images, and more time building and launching personalized and impactful campaigns. How? By allowing them to input reference images and have AI generate new images for enterprise marketing campaigns that consistently maintain brand style. 

In the example below, you can see how the style is maintained from the original top image via the prompt to the generated image on the bottom:  

And here’s where it gets interesting for marketers — below you can see the result of Jasper’s research, applied to more traditional “enterprise” brand guidelines (with the original reference image on the left and generated images on the right): 

We’re closer than ever to a world where marketers can input on-brand reference images into Jasper and generate infinite variations across campaigns, formats, and channels to drive business impact at scale, all while staying true to their brands and driving increased relevance for their target audiences. 

Keep your eyes out for more exciting Jasper Research announcements in this space soon. 

Open Source Insights From The Jasper Research Team

In Jasper’s continued commitment to advancing research and knowledge with the Open Source community, we’ve pulled together our initial findings below, including:

• An evaluation dataset of an aggregation of ~100 images collected from Open Source styling research papers

• An early-stage Hugging-face demo showcasing the most promising Inversion-InstantStyle

‍