Advanced Character Design

Using Image to Image, the right reference image and the right Prompts help you achieve the right result

GC1_-_oXcAAaA87-1

Step 1: Go to Models on the Left Hand Navigation Panel of the Webapp and Choose a Model that suits your style.

For this example, we used the Platform Model: Stylized Fantasy + Storybook + 3D

This LoRA Composition allows styles such as 3D, 3D cartoon, flat anime, oil painting, game art, etc.  Click the "Generate with this Model" button.

Screenshot 2024-01-11 at 12.38.54 PM   Screenshot 2024-01-11 at 12.43.59 PM

Step 2: Use the following basic image reference for the character in three angles: front view, right view, back view. 

You can download it here: https://drive.google.com/file/d/1OUvLRimSA0lKJXRen0yw0SlGIKBZuUBh/view?usp=sharing

Paste it into the "Image to Image" field.  

Set Influence 0. If the results don't work for you adjust the influence to between 0 and 25

4MODEL3.    Screenshot 2024-01-11 at 1.45.48 PM

Step 3: In the prompt field, write a description of your character

Describe what he/she looks like, and what he/she is wearing. You can use a simple description such as: pirate

Add the Prompt: "front view and right view and back view," You can also add an any art style you want to influence the output. When your prompt is ready, click 'Generate'.

Screenshot 2024-01-11 at 2.53.23 PM

To get more accurate results faster, use the sketch tool to change colors on the base image or add details.

Screenshot 2024-01-11 at 1.58.20 PM

Screenshot 2024-01-11 at 2.48.52 PM

Step 4: Click on the Image you like, then select Upscale

Upscale the image by 4x, then download the result.

Screenshot 2024-01-11 at 2.58.09 PM

Step 5: Train Your Character Model

Press the "+" button on the Models tab of the Navigation Bar to create a new model and then select the "Start Training" button.

Screenshot 2023-12-26 at 11.23.15 AMScreenshot 2023-12-24 at 1.32.47 PM

To train the model, select the 'Subject' LoRA Preset and add images of your character. (Beforehand, in Scenario's Canvas or any other image editor, make each of the character's angles and portraits a separate image)

Screenshot 2024-01-11 at 3.10.36 PM

GC2BQkWXcAAqBh9

Now, under each image, modify the provided caption to make it easier for the model to understand what is shown here. It is also better to add a unique name for your character (e.g. TreeMan) - for more accurate character generation. For more information on Captioning your images Click Here.

Screenshot 2024-01-11 at 3.20.47 PM

Click on the Start Training button.

Step 6: After training, use LoRA Compositions to create a style for your character.

Press the "+" button to create a new Composition and then click 'Start Composing'

Screenshot 2023-12-26 at 11.23.15 AM.  Screenshot 2023-12-23 at 12.15.45 PM

Add Your Model and choose some other LoRAs to mix it with. For this example we will use the following: - Stylized Fantasy Iconic Imagery - Vibrant Storybook Illustration - Classic 3D Realism

Screenshot 2024-01-11 at 3.23.24 PM.  Screenshot 2024-01-11 at 3.25.49 PM

Once you have your composition ready select 'Save Blend'

Review some Tips:

1. Where the Reference Image for Image to image comes from:

  • Prompting
  • Free drawing in Scenario's Sketch feature
  • Any image or smartphone photo

2. Sketch details to the reference image to generate desired colors

3. Take an output and pass it through again as a Reference Image

GC2Bs6NWIAAMKKu

Step 7: Create a scene with Sketch

GC2B1HFWkAAp9Kt

In the Prompt field write a Prompt that describes your sketch:

TreeMan stands on a rock in the forest

Continue to pass your generations through as a Reference Image and add more details with sketch:

GC2CUqBWgAAZTw0

Additional Examples:

GC2Caa8XgAAaGng