Following below steps, you would expect to get a custom model that helps you forge:
100% the same character.
Characters with new expressions.
Characters with new poses.
Characters with new clothes.
Images on a neutral background easy to remove.
Essence
To achieve these 5 goals, please remember the following:
Select 1024x1024 as default for characters.
Choose the high-resolution option which provides ultimate consistency and the quality you're looking for.
It has to be 100% the same character.
Obviously, only upload training data containing the same character.Characters with new expressions.
Training data has to have pairs of image and caption that cover many expressions.Characters with new poses
Training data has to have pairs of image and caption that cover many poses.Characters with new clothes
Training data has to have pairs of image and caption that cover many clothes.Images on a neutral background easy to remove.
Best to have your training images with this kind of backgrounds or even no background i.e., transparent background pixels in your RGBA images.Ideally upload closer to 25 images. AI learning is more effective if you input more variations, and this requires a richer training set. Upload images respecting upper limits.
All these may already sound familiar to you because they have been mentioned everywhere in our documents e.g., How do I write good captions? The rule of thumb is to be as descriptive as possible. AI can help generate novel attributes based on the combination of the provided training examples, but you need to first call them out in your data!
Once your training is complete you should be able to achieve Luna in different poses, themes, emotional states, and any context you like by prompting then refine your creations on canvas.
Example Prompt 1: Luna, carrying an Easter basket full of eggs
Example Prompt 2: Luna, in a red dress, celebrating Lunar new year, she's on stage with the half moon at the back
Note that all assets shown below except the first one was refined on Layer's canvas.
