2 ‐ Image to Image
You've learned how to create images from text prompts. But what if you want to create a new image inspired by an existing one? Eikon Diffusion's Image-to-Image feature lets you do just that!
How it Works
Image-to-Image is like text-to-image, but with an extra ingredient: your starting image.
- Encoding: Your image is converted into a special code called "latent space".
- Adding Noise: Random noise is added to the latent code.
- Diffusion Magic: The model uses your text prompt and the noisy code to predict and remove the noise.
- Decoding: The cleaned-up latent code is transformed back into a brand new image!
Key Parameters
Just like with text-to-image, you can fine-tune the Image-to-Image process with these parameters:
- Strength: This controls how much the generated image resembles your starting image.
- High Strength (closer to 1.0): The model gets more creative, producing an image that's quite different from the original.
- Low Strength (closer to 0.0): The generated image stays closer to the original.
- Guidance Scale: This determines how closely the generated image follows your text prompt.
- High Guidance Scale: The image adheres strictly to the prompt.
- Low Guidance Scale: The image has more freedom to deviate from the prompt.
- Negative Prompt: This tells the model what not to include in the image. Use it to improve quality (e.g., "blurry," "poor details") or modify the image by excluding specific elements.
Combining Parameters
Experiment with different combinations of strength, orientation scale, and negative prompts to achieve your desired results. For example:
- High Strength + High Guidance Scale: Unleash maximum creativity!
- Low Strength + Low Guidance Scale: Create subtle variations of your starting image.
Let's take a closer look at how this works:
1 - An astronaut
Let's start with a cool sample image. You'll find it in the assets folder.
- Find the "Use image as input" option: Head over to the Process tab and look for this checkbox. Make sure it's selected.
- On the left click on the Drop File Here option and select the image in the folder: assets/astronaut.png
- Denoising strength ratio: Keep default value 0.7
- Prompt: In the "Prompt" field, type the following: "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
- Prompt Method: For now, leave it set to "Normal".
You can put the value "1077105641" in the Seed field and uncheck the "Random seed" option to get the same result or leave it checked for random seed.
Click in "Generate" button to generate the image!
The "Denoising Strength Ratio" slider lets you fine-tune the balance between your starting image and the AI's creativity.
- Lower the value (decrease by 0.05): Your generated image will stay closer to the original astronaut photo.
- Raise the value (increase by 0.05): Unleash the AI's imagination! The image will become more unique and diverge further from the original.
| Original | Strength ratio=0.65 | Strength ratio=0.7 | Strength ratio=0.75 | Strength ratio=0.8 |
|---|---|---|---|---|
 |
 |
 |
 |
 |
You can also generate images in batches, just select the number of images you want as input. You must type a prompt for each image in the same order they are displayed in the interface, from left to right and always starting at the next row in the grid.
Let's see an example:
Do the same steps as you did previously, however, just select one more image from the same directory.
- Select "assets/astronaut.png" and "assets/astronaut horse.png" image.
- Denoising strength ratio: Keep default value 0.7
- Provide Prompts, type one prompt per line:
an astronaut riding a horse in mars, detailed, 8k
Astronaut in a jungle, cold color palette, muted colors, detailed, 8k
- Prompt Method: For now, leave it set to "Normal".
Generate: Click "Generate" and Eikon Diffusion will create all images based on your prompts!
| Original | Original | Astronaut on Mars | Astronaut in the Jungle |
|---|---|---|---|
 |
|
 |
 |