Until now, we worked with existing personas and assumed that we already had at least two images:
one clothed image,
one nude image.
Based on those, we built safe-for-work and not-safe-for-work photoshoots.
Now we’ll break down how to create those key base images from zero, which everything else depends on:
the character’s face
the character’s body
combining face and body
generating the first clothed image
This lesson is the foundation. If you do it correctly, all further tools will work consistently, quickly, and predictably.
Core Logic: Face and Body Are Created Separately
The first important principle:
An AI model always consists of two independent components:
The face — how the character looks
The body — how the character’s physique looks
We do not try to generate everything at once. Instead, we deliberately split the process, because different tools handle different tasks more effectively.
Step 1. Creating the Character’s Face
Face Generator as a Fast Starting Point
To create the face, we use the Face Generator tool. Its purpose is to quickly produce a high-quality portrait with minimal effort.
In Face Generator, we:
select gender,
set age,
choose appearance type,
define hair color and other basic traits,
specify the number of images.
In the examples, a portrait of a 26-year-old European woman is generated. The result is produced quickly and delivered in 4K quality, making it suitable for further work.
Face Generator is not magic — it’s a convenient interface. Internally, it:
collects parameters,
converts them into a well-structured prompt,
sends that prompt to the generation model.
When Face Generator Is Not Enough
In some cases, Face Generator’s default parameters are too limited. For example, when you want:
a specific emotion,
an unusual gaze,
a dimple,
a very particular facial expression.
In these cases, we switch to Text-to-Image.
Here, we describe the face in detail:
age,
skin type,
makeup,
gaze direction,
head angle,
overall mood.
This approach allows you to generate a face that matches your concept more precisely than Face Generator alone.
Choosing the Right Face for Further Work
After generation, it’s important to select the correct face to serve as the main reference for the model.
Recommendations:
the face should look straight at the camera or be turned no more than 20–30 degrees,
no strong head tilt,
no glasses,
no facial occlusions.
Faces with strong angles, partial coverage, or complex lighting are not suitable as a base reference.
The selected face should be:
saved,
used consistently in future generations,
added to the persona to avoid losing it.
Improving Facial Realism (Upscaling)
If the face looks too “AI-like,” use the Upscaler:
mode: Natural Clarity,
Creativity value: approximately 0.5.
Important:
the face itself does not change,
only fine details improve: skin texture, pores, eyebrows, eyelashes, micro-details.
This makes the portrait feel more realistic and natural.
Step 2. Creating the Character’s Body
Next, we need one key nude body image, which will be reused across all future photoshoots.
Why SDXL Is Used
To generate the nude body, we use Text-to-Image with the SDXL model.
The reason is simple:
SDXL renders anatomy more accurately,
it works better without body references,
it handles breasts and genitals more correctly.
The General Safe / Not Safe for Work model:
works well when body references are provided,
but performs poorly when generating a nude body from scratch.
The examples show that General often produces flat or under-detailed results that cannot be used as a base.
How to Write a Body Prompt
In Text-to-Image, we describe:
body type (fit, slim, curvy, etc.),
breast size using words, not numbers,
overall proportions,
age,
camera angle (preferably full-body or near full-body).
It’s important to understand:
neural networks do not interpret exact numbers well.
For example, “170 cm” is unreliable — descriptive terms work better.
Using LoRA for Body Control
If SDXL alone is not enough, LoRA can be added:
to increase breast size,
adjust proportions,
enhance muscle definition, and more.
Recommendations:
avoid strength values above 2,
multiple LoRAs can be combined if stronger effects are needed.
Examples show that excessive strength leads to extreme shapes, which should be used intentionally.
Selecting and Upscaling the Body
After generation:
choose one image where the body is clearly visible,
full height is not mandatory,
breasts, hips, and proportions must be visible.
Since SDXL generates images at around 1K resolution, it’s essential to:
send the image to Premium Upscaler,
upscale to 4–16 megapixels,
save it as the main body reference.
Step 3. Combining Face and Body
At this point, we have:
a separate face,
a separate body.
To combine them, we use Head & Face Swap.
Which Mode to Use
Use Full Head, not Only Face, because:
hair color may differ,
hairstyle may not match.
Recommended resolution is 2K:
more stable,
fewer artifacts.
4K can be used, but results are less predictable.
Result of the Combination
The output gives us:
the same body, unchanged,
the face correctly integrated,
the first complete image of the nude model.
If the result looks good:
save it,
optionally apply Natural or Ultra Clarity upscaling.
Step 4. Generating the First Clothed Image
Now we need a clothed version of the model in order to:
create safe-for-work content,
use it as input for future generations.
For this, we use Editor & Combiner.
We:
upload the nude model image,
write a simple prompt (for example, “dress her in a red dress”),
disable Sequential Generation to get variation,
run the generation.
Result:
body proportions remain consistent,
breast size is preserved,
clothing appears correctly.
Final Model Creation Pipeline
The full process can be summarized as follows:
Create the face
Face Generator or Text-to-Image
Create the body
Text-to-Image + SDXL
Upscale the body
Premium Upscaler
Combine face and body
Head & Face Swap (Full Head)
Generate the first clothed image
Editor & Combiner
After this, you have:
a base face,
a base body,
a nude model,
a clothed model.
With this foundation, you can:
create safe-for-work photoshoots,
create NSFW photoshoots,
use carousels, photoshoots, image-to-image, editor tools, and more.
Final Note
If questions or doubts arise during the process, you can always contact support — it’s available on every interface page and responds fairly quickly.
A properly built model at the start accounts for 90% of future success.