Seamlessly generate unlimited versions of a character across different eras or historical contexts.
STEP 1 - START FROM ONE IMAGE.
Our starting point is an image of a young woman with striking, curvy red hair, generated using a simple prompt on a custom model.
You have over 170 models to choose from on Scenario, each offering a variety of distinct styles!
OPTIONAL
For those seeking advanced style customization, it's possible to train your own style or to create a LoRA composition (from over 12 million unique combinations.)
Below are some of the training images utilized for this character-centric model, guiding the AI to "learn from" this CG/illustrative style:
STEP 2: UPLOAD TO "IP ADAPTER“
Import (or drag/drop) your original image into the “Reference Image“ section to the left, and select "IP Adapter" in the drop-down menu.
We recommend testing some simple prompts and experiment with various influences to find the ideal settings. In this case, the sweet spot was 20.
Additionally, we also tested the model's influence (top-left corner) and settled on 0.7.
For some basic tips on IP Adapter, you can also check out this guide by clicking here.
YOUR MODEL IS IMPORTANT
To maintain the unique style of your images, consistently use the same model.
Setting the model influence to 0 effectively means utilizing SDXL, which might affect the style (as shown below).
SDXL can still be used in certain cases however, especially for generating realistic images.
STEP 3. START PROMPTING
After preparing your image and parameters, you're ready to prompt some variants. The prompts in this example all followed a consistent format:
Half-body shot, (character description), (outfit description), (background description), short curly red hair."
Keep the descriptions short! For instance:
Classical Greece woman, toga, temple, marble columns
You might also consider using negative prompting to slightly enhance the quality of the output ("blurry, ugly, draft, lowres, artifacts" etc)
As shown below, with only 18 inferences (producing 74 images), we selected images across epochs (seen at the top of the page).
Alternatively, you can make just 10 inferences, each generating 4 images, which would have been more than enough. After all, over 80% of the images generated were satisfactory already.
All of this is also available through the Scenario #API, ready for integration into any UI, app, or game!
