AIRabbitAI image generators such as FLUX and Stable Diffusion (I refer to them as models) can generate stunning images that can bring a lot of attention to a person or brand. One of the images I remember going crazy viral was a picture of the Pope wearing a puffer jacket. It was hilarious and got a lot of attention.
This image was generated with a single prompt, and it was possible to generate it because the AI (Midjourney as far as I know) already knew all the subjects in the picture (Pope Francis and the puffer jacket).
The applications are endless because the generators were pretrained on many other things that we could combine to generate crazy stuff.
The problem
But what if the AI model does not know the subjects that should be part of the picture, for example a brand logo or icon? This could be the case if the brand is not as popular as a big brand with hundreds of thousands of images and media on the internet used to train these models.
Here is an example
In the course of this tutorial we will be using the logo of one of my favourite Open Source UI frameworks called Svelte. This is what the logo usually looks like when you Google it.
If we ask the Foundation Flux model (not fine-tuned) to generate a scene with this look, it will fail because it does not know what it looks like.
Here is what I get on Segmind:
And on Replicate with a different set of parameters, it does not look much different. The icon is not the Svelte icon.
I tried it on different platforms and it does not look much different.
Flux does not really know how the Svelte icon looks like.
AI Rabbit News & Tutorials\
*AI Innovations & Productivity Hacks: Trends, Tools, Tutorials*airabbit.blog
AI Rabbit News & Tutorials
Solving Real-World Problems with AI: Harnessing ChatGPT, Claude & More. Explore Innovations, Productivity Hacks, Trends, Tools, and Tutorials.
airabbit.blog
The solution
Instead of trying to describe the shape of your brand or logo, which can be difficult in most cases, you could fine-tune the model with the subject you want to have in the generated images. These themes can be anything:
- shapes
- Logos
- People
- Literally anything
These objects don't exist in a vacuum, of course, but the real power and creativity comes, as with pre-trained objects, from combining them with other custom or generic objects to create stunning images that can be used for marketing.
An example of this is Merchandising or technically Mockup Generation with a given image or logo of your brand (could even be a person). By fine-tuning the model with this object, you can combine them to create scenes that you can use to advertise your brand or even generate ideas for new products that you can actually sell, for example for Print on Demand.
Here is a concrete example:.
Imagine you have a POD business and you have a creative logo, and you want to see this logo on all sorts of objects, like mugs, T-shirts, etc., in order to attract customers, but also to make new products.
You might say: "But there are already tools that can show me my brand on mugs, etc.".
Well, yes and no. They usually only show you the projection of your logo for predefined shapes and scenes. But with Fine tuning for YOUR own brand you can create an infinite number of scenes and combinations of other objects and the sky is the limit! It's a completely different game.
