One of the most popular use cases amongst our gaming audience is characters. Our top users are teaching Layer their unique character style and use Layer to generate new character concepts and with some refinement production assets.
Step 1: Gathering Assets
The first step is to upload assets to Layer. For effective character set training, we need to collect a diverse range of character assets, each featuring the different characters in different ages, physical features, close up and full body shots.
Upload character assets ideally in the same format: 1024 x1024. Learn more about the specs and why they are important here.
Aim for around 25 assets to ensure a rich and varied training input for training a set of characters. if not lowest you can get away with is around 12.
Vary your images: different age, gender, emotional state, poses, shots (medium, full body close up), outfits are all important variations.
Make sure to have enough full body shots in the training to give more information about the characters.
Maintain style consistency across all assets. Learn more about what images to upload here.
If you have fewer characters let's, say 4 make sure to have at least 6 variable images for these 4 characters to a total of 24 images. In this case naming characters could also be a good idea as long as each character has enough input images in the training
Here are some example assets and captions which can be used when training a set of characters:
Step 2: Input Captioning
As you upload images to Layer your captions will be auto-populated thanks to auto-captioning feature. Yet since captioning character assets in detail is essential to guide Layer's AI model accurately, we recommend reviewing these captions to make sure they include all the important details. Here is a list:
If you wish any physical feature of the characters to change, try to describe all these details to Layer.
If you want certain details to be unchanging don't caption them.
For great character captions, include character's positions.
Include camera angles and shots where relevant (e.g.,
close up shot,
full body, l
ow angle shot,
above the waist shot) so that you can achieve them in these angles and shots more accurately.
If all of your assets are from same shot, no need to caption that information.
Step 3: Naming The Style and Example Prompts
Once the captioning is complete, we proceed (or go to the "Settings" tab) and name our art style. Additionally, provide evaluation prompts for the variations that Layer will generate. These prompts will guide the AI model in producing artwork that aligns with our creative vision.
Example style name: Game X - Characters. As you'll need to be training a few styles per game this type of formatting helps to organize all your styles per game which you will be needing assets for.
Step 4: Generating The Style
Now that we are ready, we will start training our style. To do this, we click on "Generate Your Style" to initiate the training process. Layer will create 8 variants per generation, showcasing different attempts at generating the style. This allows us to choose the variant that best matches our desired output.
Step 5: Evaluating and Regenerating Variations
As the style is trained, we should evaluate the generated variations to find the best and most accurate attempt. Don't worry; if we are unsatisfied with the results, we can modify your captions or images and re-generate the style. This will add extra variations to consider for selection.
Step 6: Forging with the Style
After selecting the variation that aligns with our artistic vision, we can start creating artwork with our trained style. When forging you can use simple prompts like in the example 1 or refer to the prompt enhancer - magic wand button - on your prompting box.
a girl in a school uniform wearing glasses and a backpack.
Before prompt enhancer
a guy in Michael Jackson outfit.
After prompt enhancer
full body shot of a charismatic guy dressed in an iconic Michael Jackson outfit, adorned with a sparkling sequin glove, his curly black hair falling around his face, perfectly executed dance moves under the colorful stage lights, capturing the essence of Michael Jackson's unparalleled stage presence and showmanship.
As you forge, using prompt enhancer along with the combination of prompt strength bar and reference images or poses can lead to better results. Every user has their own artistic vision, and these controls allow you to experiment with to reach the outputs you're looking for. Here is a quick snapshot of the prompting box features. Read more about them here
Whether you're aiming for a medieval-fantasy world, a retro-inspired adventure, or a futuristic sci-fi experience, Layer provides you with the tools to transform your concepts into mesmerizing artwork that captivates players and immerses them in your game's universe.
Start by creating your own style!