> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Design Wizard
> Describe how the voice should sound. We generate it from scratch.
**Design a voice** is the two-step wizard that builds a brand-new voice from attributes (language, dialect, gender, age, pitch) without a sample recording. You reach it from the **Create A Voice** button in the [Voice Picker](/en/Systems/Text-to-Speech/Voice-Picker), the **+ Create Voice** button in the topbar, the empty-state button on the [Voice Library](/en/Systems/Text-to-Speech/Voice-Library), or the picker's **Design a voice** card.
## The wizard shell
The Design Wizard opens as a wide modal that takes most of the screen. The header shows the two steps as a brand-gradient progress bar, identical in shape to the [Clone Wizard](/en/Systems/Text-to-Speech/Clone-Wizard).
| Region | What you see |
| ------------ | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| Step 1 chip | The label **Step (1) Voice Details** on a brand-gradient pill on the start edge of the header. The number turns into a check mark when you advance. |
| Step 2 chip | The label **Step (2) Generate Custom Template** on a muted pill on the end edge. Activates when you reach Step 2. |
| Close button | A small **X** icon button. Closes the wizard immediately, your in-progress fields are dropped. |
## Step 1: Voice Details
The body is one column with a header (cover image and name), a grid of attribute dropdowns, an optional **Generation Settings** panel, and a footer.
### Cover image (optional)
| Element | Notes |
| --------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
| Header | The label **Cover image**. |
| Drop area | A 110 × 110 px tile with a dashed border. Click to open a file picker. The hint inside reads **JPG, JPEG, PNG** with **Max 5MB** below it. |
| Preview | When an image is uploaded, the tile fills with a cropped preview. |
| Errors | *Image must be JPG, JPEG, or PNG* (wrong type) or *Image must be 5MB or less* (too large). The error appears in red below the tile. |
### Name (required)
| Element | Notes |
| ----------------- | ------------------------------------------------------------------------------------------- |
| Field | A standard text input. Placeholder *Enter voice name*. Maximum 64 characters. |
| Empty state error | *Name is required*. The field outline turns red when you click **Next** with an empty name. |
### Language (optional)
A label pill **Language (optional)** with the help text *Keep as Auto to auto-detect the language*. Click the field to open a popover with the same shape as the Studio's language picker.
| Inside the popover | Notes |
| ------------------ | ----------------------------------------------------------------------------------------------------------------------- |
| Search | Filters the list by English name, Arabic name, or language code. Up to 200 rows render at once. |
| **Auto** row | Default. The 🌐 globe glyph and the label **Auto**. |
| Per-language rows | Each shows a country flag, the language name, and a small **+** badge on the end edge for languages with dialect lists. |
### Dialect
The dialect field appears only when the chosen language has a dialect list. The label pill reads **Dialect**. Click to open a popover.
| Inside the popover | Notes |
| ------------------ | -------------------------------------------------- |
| **Auto** row | Default. Uses the parent language's flag. |
| Per-dialect rows | Each shows the regional flag and the dialect name. |
There is no separate **English Accent** or **Arabic Dialect** field on this wizard. The single **Language** plus **Dialect** picker pair covers both. The same picker is shared with the Studio, the agent builder, and the [Clone Wizard](/en/Systems/Text-to-Speech/Clone-Wizard) so the catalog stays consistent everywhere.
### Attribute grid
Below the language section, three dropdowns sit in a responsive grid (one column on phones, two columns from small screens up). Each label sits inside a brand-gradient pill. The default for every dropdown is **Auto**, which lets the engine pick.
| Dropdown | Options |
| ---------- | ------------------------------------------------------------------------------------------------------------------ |
| **Gender** | **Auto**, **Male**, **Female**. |
| **Age** | **Auto**, **Child**, **Teenager**, **Young Adult**, **Middle-aged**, **Elderly**. |
| **Pitch** | **Auto**, **Whisper**, **Very Low Pitch**, **Low Pitch**, **Moderate Pitch**, **High Pitch**, **Very High Pitch**. |
The active option carries a small green check mark on the start edge inside the popover.
Leaving every attribute on **Auto** is fine. The engine balances the sound automatically. Pin specific attributes only when you want to enforce them, the more you pin, the less freedom the engine has to find a natural-sounding match.
### Generation Settings
A collapsible panel below the attribute grid. Header reads **Generation Settings (optional)** with a chevron. Default state is collapsed. The same six controls as the Studio appear here:
| Control | Default | Notes |
| ---------------------- | -------- | ---------------------------------------------------- |
| **Speed** | Normal | Three-tier slider, **Slow** to **Fast**. |
| **Quality** | Balanced | Three-tier slider, **Faster** to **Higher quality**. |
| **Similarity** | Balanced | Three-tier slider, **Stable** to **Similar**. |
| **Denoise** | On | Checkbox. |
| **Preprocess Prompt** | On | Checkbox. |
| **Postprocess Output** | On | Checkbox. |
### Step 1 footer
A single brand-gradient **Next** button on the end edge. The only validation that runs at this step is the name check, every other field is optional. Errors light up the offending fields in red.
## Step 2: Generate Custom Template
Same shape as the Clone Wizard's Step 2.
### Header row
| Element | What it does |
| ------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Title | **Generate Custom Template**. |
| **Example** button | Restores the textarea to the default sample text. |
| **Generate** / **Regenerate** button | Brand-gradient pill with a lightning icon. Reads **Generate** before the first preview, **Regenerate** after a preview has rendered. Disabled when the word count is below the minimum. |
### Sample text
| Element | Notes |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Label | **Text**. |
| Default text (English) | *Welcome to our voice cloning service! This is an example of how your custom voice will sound. You can use this for various applications like audiobooks, podcasts, or personalized voice assistants. The quality and naturalness of the cloned voice depends on the audio samples you provide.* |
| Default text (Arabic) | The Arabic version of the same paragraph. |
| Maximum length | 1000 characters. |
| Word count footer | Reads **Word count: `{used}/{min}`** with a chip on the end. **Valid** in green when the count is high enough, **Too short** in muted text otherwise. |
| Minimum | 5 words. |
### TTS Voice Preview
| State | What you see |
| --------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Idle (no preview yet) | Title **Preview Ready** and body *Click the 'Generate' button above to create an audio sample of your voice. This will help you hear how your custom voice will sound before proceeding.* |
| Loading | Spinner with title **Generating preview\...** and body *This can take a few seconds.* |
| Ready | A brand-gradient round play button on the start edge plus a built-in audio player. Play and pause are aria-labelled **Play preview** and **Pause preview**. |
| Failed | Title **Preview failed** with body *Something went wrong. Please try again.* and an alert icon. Click **Regenerate** to try again. |
### Step 2 footer
| Button | Behaviour |
| ------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Previous** | Returns to Step 1 with all fields preserved. |
| **Create** | Brand-gradient submit. Disabled until a preview has rendered. The label flips to **Creating...** with a spinner during the save. On success a toast confirms with the localized **Create** word, and the wizard closes. The new voice lands in your [Voice Library](/en/Systems/Text-to-Speech/Voice-Library) immediately. |
## Provider routing and tenant rules
Designed voices share the same provider routing as cloned voices. The card never surfaces the underlying provider, the designed voice surfaces as a Wittify voice everywhere.
Voices are Wittify voices. Provider names must never appear on the wizard, the Voice Library card, or the Voice Picker row.
Designed voices are scoped to your account. Other users in the same workspace cannot see them, and the backend rejects cross-account access.
## Common questions
You will get a generic neutral voice in the language the engine detects from your sample text. Useful as a quick first pass, less useful when you need a specific tone.
You are not aligning the engine to a real audio sample. The only hard requirement is a name so the voice has an identity in your library. Every other choice is a hint.
Designed voices are best-effort. The engine tries to match every pinned attribute, but extreme combinations (Whisper plus Excited plus Elderly, for example) can produce unnatural results. Loosen one or two attributes back to **Auto** and regenerate.
Not from this wizard. Use cases are only collected during the Clone Wizard's Tags step. Designed voices show up in the Voice Library with the **Designed** badge and no use-case chips.
Yes. Once it reaches **Ready** in the Voice Library, it appears in every voice slot across the app, including the agent builder's voice step.
The wizard is intentionally stateless, closing it discards your choices. Plan to finish in one go. The wizard is short, picking five dropdowns plus a name should take under a minute.
## Where to go next
Find the designed voice and manage it.
Use the designed voice to generate.
Pick the voice from My Voices.
Make a voice from a sample instead.