AI - Editing Images (Generate Variation)

[info-creaticode]

Introduction

 

You are not only generate images using a text prompt, but also modify an existing image with a text prompt. This tool can be used in many creative ways. Below are some examples.

 
 

How to generate a variation

 

You can generate a new image based on an existing image, which can be either a backdrop (a rectangular image that covers the entire stage) or a costume (with a transparent background). Here are the steps:

 

Step 1 - Search for any image in the library

 

For example, you can open the AI image tool for backdrops, and search for an image of a school like this:

 
 

Step 2 - Select an image as “reference”

 

You can select any image in the grid by checking the “use as reference” checkbox below that image, and a new button for “Generate Variation” will appear.

 
 

Step 3 - Specify how you want the image to be changed

 

Next, in the input box, remove the existing text and write down how you want the selected image to be changed. Note that this is different from the prompt you use to generate a new image. It should clearly tell the AI tool how the image should be changed, such as “make it …”, “change it so that …”, “add … to it”, “remove … from it”, etc. It is a common mistake to repeat the same prompt that was used to create the reference image.

For example, we can tell the AI tool to remove all the kids in the image and only show the background:

 
 

Step 4 - Click the “Generate Variation” button

 

Lastly, click the “Generate Variation” button and wait for a few seconds. During that time, the AI tool will automatically review if your request is appropriate, and if so, it will try to generate a new image based on the reference image and your prompt.

Note that sometimes the request may be rejected in the review process. You can try again when that happens, and the AI reviewer might let it pass sometimes. Of course, you can also change the reference image or your request.

 
 

Step 5 - Decide what to do with the new image

 

After the new image is generated, you can click on it to use it in your project. If you are not happy with it, you can either delete it or regenerate it.

 
 
 
 
    

Interesting Use Cases

 

Now let’s look at some interesting ways to use this new tool to help you generate the perfect images for your projects.

 
 

Use Case 1 - Remove People/Objects

 

You have already seen an example above, where we removed the kids from the image. Here is another example. Suppose we really like a table, but we want it to be completely empty, we can change it:
 
Request: remove objects on the table
Reference and Result Images:

 
 
Note that the current AI tool often fails if we ask it to remove some objects and keep other things. For example, if we say “remove the knife on the table but keep the plate of fruit”, it would still remove everything on the table.

 
Request: remove the ground so only show the wolf
Reference and Result Images:

 
Request: remove all the pills outside the bottle
Reference and Result Images:

 
 

Use Case 2 - Add People/Objects

 

Here are some examples of adding new people/objects.

Request: put on a green t-shirt
Reference and Result Images:

  
 
Request: add a car in the parking lot
Reference and Result Images:

Note that the newly added car is drawn with the same style as the reference image.

 
 

Use Case 3 - Change properties of an object or scene

 

You can also keep the same object or scene, but change some properties like color, shape, material, etc.

  
 
Request: change it to rainy and cloudy with no sunshine
Reference and Result Images:

  
 
Request: make the hand 5 fingers
Reference and Result Images:

  
 
Request: change the robot’s eyes to blue
Reference and Result Images:

  
 
Request: turn the car to face straight towards the viewer
Reference and Result Images:

  
 
Request: change this table to steal
Reference and Result Images:

 
 

Use Case 4 - Change the pose of a character

 

You can also change the pose of a character so you can make animations. However, this feature is still very limited and often fails. Usually it works better with non-human characters. You might need to retry a few times and adjust your prompt to find out what works.

Here are some examples:

  
 
Request: make the dog body wave one hand
Reference and Result Images:

 

  
 
Request: frustrated with both hands covering face
Reference and Result Images:

  
 
Request: Rest his chin on hand, tilt his head and think
Reference and Result Images:

  
 
Request: surprised with both hands on face
Reference and Result Images:

  
 
Request: frustrated with head facing ground
Reference and Result Images:

  
 
Request: happy and jumping up
Reference and Result Images:

  
 
Request: raise both arms to show big thumb up to the viewer
Reference and Result Images:

  
 
Request: running and facing the side and looking straight forward
Reference and Result Images:

  
 
Request: sitting and playing piano
Reference and Result Images:

  
 
Request: eating a big burger with both hands holding it
Reference and Result Images: