Innovation Hub
An initiative by Al Bayat Mitwahid in UAE
Course 2
Using and Training AI
Unit 2 – Generating images
Lesson 3: Getting better images
You may have found in lesson 1 that some of your generated images weren’t exactly what you were looking for. Similar to chatbot prompts, we can get some more success with image prompts if we practise some refinement techniques.
​
INTERACTIVE
You can usually see the prompts used for generated images in the user gallery sections of image generator websites. Have a look through these to get ideas for your own prompts.
​
Stable Diffusion has a service where you can search their prompt database to find out what wording was used by other people. You can use this website - https://stablediffusionweb.com/prompts - to search. Be aware that although this website won’t generate inappropriate images, you may find that images from other users contain grotesque or medical style content that you might not like.
​​
To improve the contents of your image and make them as interesting as possible, think about the subject of the image and what is happening in the scene.
​
Subject: Describe in detail the important parts of the image. If you are including people or animals, describe their physical appearance in terms of position, personality, clothes, hair, eyes etc. If there are objects or landscapes, describe their colours, shape and positions.
Background: Add details about anything you want to show in the background of the image, such as mountains, trees, buildings, people etc.
Action: Include words to describe any actions that you want to happen in the image. This could be actions like running, jumping, dancing etc. It could also be how something is moving, for example a dog running energetically, or a tree drooping sadly.
Mood: You can describe the overall mood of the image by using adjectives. For example, a happy bird singing with delight.
Light: You can describe details about how you want the scene to be lit. If it is an indoor scene you can add details for the colour, quality and source of light. If the image is outside, it could be direct sunlight, shaded, moonlight, glare, lens flare etc.
EXAMPLE
Prompt: "A photorealistic happy squirrel who is waving one hand at the camera enthusiastically. They are sitting on a sleek desk chair next to a wooden ornate desk. Most of the room is in darkness, but there is light coming from a desk lamp. The light has a soft orange glow."
There are newer AI techniques that create video from static images. Here is an example from Runway with Image Motion https://app.runwayml.com where the AI program has created motion from a single image. The user can select which areas of the image to add motion in certain directions, and there is an option to add a zoom effect. Notice that the original image doesn’t include more than one yellow flower.
​
You can try this yourself using an image you already have, or from a text prompt. When you have an image, you can use the Motion Brush button to select the area where you want motion generated. Then use the options on the toolbar to control the direction of movement and any camera movement.
​
EXAMPLE
This is the image from the beginning of course 1, with added motion around one subject, and added camera movement.
