> ## Documentation Index
> Fetch the complete documentation index at: https://guide.ourdream.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Step 2: Build the appearance
> Create a stable visual identity by separating permanent traits from temporary scenes.
This is the visual pass. Your goal is a recognizable base character—not a finished gallery image.
Before starting, complete [Step 1: Plan the concept](/characters/concept-and-brief). The concept should guide the appearance without turning the visual prompt into a biography.
## 1. Choose the broad appearance
Start at [ourdream.ai/create](https://ourdream.ai/create). Work through the style, gender, general appearance, face, and body steps using presets first.
Treat the body controls as a combined system. Body type, breast size, and butt size can influence each other, so the result may not match each label in isolation. Start with the closest preset combination and fine-tune later.
Avoid age-ambiguous wording such as *little*, *tiny*, or *childlike*. All
characters must be unambiguously 18 or older. See the [underage-content
policy](https://safety.ourdream.ai/policies/prohibited-content#underage-content).
## 2. Keep the first generation simple
Fill in the required basics, but keep optional visual details light. Generate a baseline before adding a long custom prompt.
Ask:
* Is the overall body shape close?
* Is the face moving in the right direction?
* Are the hair and skin tone stable enough to continue?
* Did an occupation, hobby, outfit, or other selection introduce an unwanted visual element?
If a selected detail keeps appearing, remove it temporarily while tuning the base.
## 3. Tune one trait at a time
Use **Custom Physical Details** and, where available, **Custom Face Details** for stable appearance traits that presets cannot express.
For realistic characters, begin with the same focused appearance prompt in both fields. Some face-related terms influence one field more than the other, and duplication makes the base easier to evaluate.
Use this loop:
1. Add one trait.
2. Generate a new image.
3. Compare it with the previous result.
4. Keep, reword, or remove the trait.
5. Only then add the next trait.
If a term has no visible effect, try a common visual synonym. If it is too weak or strong, adjust its weight in small steps. Use the [Image prompting reference](/characters/image-prompting) when you need syntax or model-specific help.
## 4. Separate permanent and temporary details
Anything in the character's appearance fields may affect every future image.
| Keep permanent | Keep temporary |
| ----------------------------------- | -------------------------------- |
| Face shape and stable facial traits | A pose for one scene |
| Hair color and core hairstyle | A special outfit |
| Body shape and stable features | A location or background |
| Defining markings | A temporary expression or action |
If shoes, a uniform, or a location is placed in a permanent field, the model may try to show it in every generation.
## 5. Choose a useful base
The best base image is easy to reuse:
* The face is clear and unobstructed.
* The pose is neutral.
* Lighting is simple.
* The background is uncomplicated.
* Temporary clothing and props do not dominate the character.
Choose the result with the strongest identity, even if the scene itself is plain. You will create polished scenes in [Step 5: Create the gallery](/characters/photo-shoot-and-gallery).
## Appearance checkpoint
Continue when:
* The character is recognizable across several generations.
* Permanent prompts contain identity traits only.
* You know which visual details still need image editing.
* One clean result can serve as the base for future scenes.
Add the Public Description, Greeting, Scenario, personality, and supporting
details.