- AI-Enhanced Creator
- Posts
- Easily Swap Colors with Embeddings
Easily Swap Colors with Embeddings
Next level text prompting
Hello Friend
Today, let’s dive into colors.
We’ve all been fascinated by them in one way or another. Soft pastels, Bright neons, Complementary, monochromatic - Colors have the power to evoke emotions, convey messages. A powerful creative language.
A language AI can interpret. In different ways.
Using Colors with AI
So, how do we use colors with AI? One way is by using text prompts.
You can describe a color scheme you want to achieve (e.g., "a warm and cozy autumn palette") and an AI model will try to interpret that information in the images it generates.
It’s a great way to get started. We will often have good results and we will expand our creative language vocabulary as we look for words that describe the colors we have in mind.
Today, I want to take you a step further from text prompts. With Embeddings.
What are Embeddings?
At the base level you can see embeddings as prompt packages. Packages of information. In our example, this information are colors. But instead of words, like with a text prompt, the information are mathematical, numerical values.
Feels a bit abstract, right?
Think of a blue dress. when you see it, the color is clearly blue (this would be a text prompt for AI). There are a lot of shades of blue, a lot of ways to achieve the color blue. When you go into the details and zoom into the blue dress. You will find the fabric. And beyond the fabric the pigment from a specific flower used to create the blue dress (think of this information as an embedding).
Can you see the power of embeddings?
It’s a communication closer to the language of AI.
Here is an example with color palette embeddings:
I used this embedding to generate this image:
![]() picture with a flat color background of a woman with dreadlocks, wearing japanese techwear, <Red Black White color palette> | ![]() Color palette in the embedding |
I used the same prompt with this embedding for this image:
![]() picture with a flat color background of a woman with dreadlocks, wearing japanese techwear, <Cyberpunk color palette> | ![]() Color palette in the embedding |
Refining the text prompt will further help us control the colors and the elements in the images.
You can try out embeddings with Stable Diffusion based interfaces.
Personally, I use Invoke.
I can upload the embedding into their cloud interface or on my local installation.
In the main generation interface I can add a prompt trigger, which can be selected next to the text prompt.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/d5c2050a-a387-4ec8-886c-91f7fde0dd9d/image.png?t=1716204223)
Top right button </>
It’s another great way to explore AI image generation. Both on creating consistency in images and discovering something new.
Keep creating and always remember to have fun my friend. ☀️
Join our Group: Group for professionals to connect, collaborate and share insights.
AI-Enhanced Creator: Power of a Production House: Naturally integrate Language models and Image generators into your work.
Coaching and Workshops: Looking for a hands-on transformation? Book a call and we’ll get to know each other. 🤗