How to use it

Recommended workflows

Choose the path that matches what you are making. Save text and generated audio to the Library when you want to keep them.

1. Clean up and publish text
   Clean Text → optional rewrite or summary → review the result → export or save.
2. Voice a long script
   Clean Text → Chunk Splitter → Neural TTS → download or save each audio part.
3. Russian narration with stress marks
   Clean Text → Stress Tool → check names and ambiguous words → Neural TTS.
4. Make a video script and visuals
   Script Writer → review the draft → Prompt Weaver → Neural TTS.
5. Turn recorded audio into a new voiceover
   Audio Processor: record or upload → transcribe → normalise → generate voice → save.

Use Send to between text tools to carry the current text without copying it by hand.

Clean Text

This is where most work starts. Text that came out of a word processor, a web page or a transcription service carries invisible formatting, stray double spaces, timecodes and line breaks in the wrong places. Clean Text takes all of that out and gives you a master version you can edit, save and pass to the other tools.

When to use it

Every time text comes from somewhere else. If you typed the script yourself in this app, you can usually skip it.

Step by step

1. Paste your text into the source panel, or upload a .txt, .srt or .vtt file.
2. Look at the options on the right. The defaults suit most scripts; turn on "Numbers to words" if the text is going to be read aloud.
3. Press Process. The cleaned text appears in the result panel.
4. Read it. Use find and replace, or the AI actions, to fix anything the rules could not.
5. Press "Send to" to hand it to the Stress Tool, the Splitter or Neural TTS, or save it to your library.

What the options do

• Remove formatting — takes out HTML tags and the markup a document paste leaves behind.
• Remove timecodes — deletes subtitle timings such as 00:01:24,500.
• Normalize spaces — collapses repeated spaces and tabs without destroying your paragraphs.
• Keep ё — on by default. Leave it on for narration: ё and е are different sounds.
• Numbers to words — turns 123 into сто двадцать три, so a voice engine reads it rather than spelling it.
• Add paragraphs — inserts breaks into a wall of text that arrived with none.

Every processing run is kept as a version. If a change made things worse, step back to the previous version rather than starting again.

A good habit: clean first, then change the meaning (rewrite, summarise, translate), and only then format and export. Doing it in the other order undoes your own work.

Chunk Splitter

One job: cut a very long text into parts small enough for a voice engine to accept, without breaking a sentence in half. A sentence cut down the middle is audible — the voice drops its intonation and the next part starts cold.

When to use it

Before generating a voiceover for anything longer than roughly nine thousand characters, or whenever you want to generate and review your audio section by section instead of all at once.

Step by step

1. Paste the long text, or send it here from Clean Text.
2. Press "Split text".
3. You get numbered parts, each under the limit and each ending on a sentence boundary.
4. Copy a part at a time, or copy all of them, or send them straight to Neural TTS.

The splitter knows about abbreviations, so "т. е." and "Dr." do not fool it into thinking a sentence ended.

Stress Tool

Russian stress is not decoration. Зáмок is a castle and замóк is a lock, and a voice engine has no way to tell them apart unless you mark them. This tool places the marks from a dictionary of about 140,000 words, and can use an AI model for words the dictionary has never seen.

When to use it

Any Russian text that is going to be spoken. Skip it for English.

Step by step

1. Paste the Russian text, or send it here from Clean Text.
2. Press Process to apply the dictionary. This is fast and predictable.
3. Press "AI stress" for the words the dictionary did not cover. Slower, and worth checking.
4. Read through the names and any word that can be stressed two ways — context decides those, and only you know the context.
5. Send the result to Neural TTS.

Teaching it

If your text already has correct stress marks, "Scan and learn" adds those words to the dictionary so you never have to fix them again. That dictionary is shared, so corrections compound over time.

Script Writer

A guided way to get from an idea to a finished video script. It works in steps rather than producing one long block, so you steer it at each stage instead of rewriting whatever it hands you.

When to use it

When you have a subject but no script. If you already have the script, use the editing mode to refine it instead.

Step by step

1. The idea. Describe the video, add source links if you have them, and set the audience, tone and length.
2. Angles. Ask for angles and pick the ones worth making. This is the step that decides whether the video is interesting.
3. Openings. Look at the sample hooks and choose one. The first fifteen seconds do most of the work.
4. The full draft. Generate it from the angle and opening you chose.
5. Refine. Select a paragraph and ask for it to be corrected, expanded, condensed or rewritten — only that paragraph changes.

The output is labelled [TITLE], [DESCRIPTION], [HASHTAGS], [TAGS] and [SCRIPT]. The other tools recognise those labels and will take just the script part when that is what they need.

Use the draft reader to hear the pacing before you spend a voice generation on it. It is the browser's own voice, not the final one, but it exposes sentences that are too long to say out loud.

Prompt Weaver

Turns a script into visuals. It splits the text into scenes and writes a detailed image or video prompt for each one, so a forty-scene video does not mean writing forty prompts by hand.

When to use it

Once the script is final. Prompts written against a draft have to be redone when the script changes.

Step by step

1. Paste the finished script.
2. Set the format, the visual style, the characters who recur, and anything you never want to see.
3. Press "Generate prompts". You get one card per scene, each showing the text it came from.
4. Copy the prompts you want, regenerate the ones that missed, or generate images directly.
5. Ask for the YouTube kit if you also need a title, description, hashtags, tags and a thumbnail prompt.

Describe your characters once, in detail, and keep that description identical across scenes. Consistency in the prompt is what produces consistency on screen.

Neural Voiceover

Generates the voiceover. Long scripts are split into parts automatically and each run becomes a job you can watch, so a twenty-minute narration does not depend on one request surviving.

When to use it

Last. Everything else prepares the text for this step.

Step by step

1. Paste the script, or send it here from Clean Text or the Stress Tool.
2. Choose a voice profile, and set speed and pitch. Small changes go a long way; leave them alone the first time.
3. Check the estimated duration — it is a good sanity check on the length before you spend the generation.
4. Press "Generate voiceover" and wait for the parts to finish.
5. Play each part, download it, or save individual parts to your library.

For Russian, run Clean Text and the Stress Tool first. It is the difference between a voice that reads your script and one that reads a transcription of your script.

Audio Processor

The whole audio workflow in one page, built around three steps: get a source, normalise it, generate speech from it. Use it when you are starting from sound rather than text.

Step 1 — the source

• Record straight from your microphone and have it transcribed.
• Upload an mp3 or wav to transcribe a file you already have.
• Upload a script if you only want the normalising and voicing half.

Step 2 — normalise

A transcript is not a script. Numbers, units, abbreviations and missing ё all read badly aloud. Turn on the rules you want and press Normalize. Every change is listed in the processing notes, so you can see exactly what was altered and why.

Step 3 — generate

Pick the voice settings and press "Generate voice". Listen in the preview panel, then save to your library.

Read the normalised version before generating. Transcription mishears names, and a name is the one thing listeners notice.

Library

Everything you save lands here: scripts, audio and images together. It is the part that matters six months later, when you need the file you made in March.

What it does for you

• Versions. Saving under the same name keeps the earlier copy rather than replacing it. Open Versions to go back.
• Notes and status. Mark what is a draft, what is approved, and what is published.
• Categories and search. Search narrows the list as you type.
• Archive instead of delete. Archived files leave the list but are recoverable. Deleting is permanent.
• Send to. Saved text can be opened in another tool without downloading it first.

Give files real names. "Ep 14 — intro, stressed" will mean something to you later; "text 3" will not.

Your profile

Your account page holds your personal settings and data controls.

• Update your name, email address and avatar.
• Review your plan and request a paid plan. An administrator approves paid requests.
• Change your password and review signed-in sessions.
• Choose which account emails you receive.
• Download a JSON copy of your account data.
• Delete your account and stored work when you no longer need them.

Download anything you want to keep before deleting your account.