Voice commands for AI video and voice lines — talk to the studio

The Ask bar at the top of the studio has a mic. Tap it, say what you want — a clip, an image, a voice line — and stop talking. The sentence sends itself, Ask sets up the right tool with what you said, and the page opens ready. The one thing it never does is press Generate for you.

This guide covers where the mic works, how to phrase a spoken ask so it lands the first time, follow-ups by voice, and the distinction that matters in a studio that also makes voices: a command versus a line to be spoken.

Where the mic is

• Ask is the bar at the top of the studio home (/studio, after sign-in) — the default tab of the home composer, with the placeholder "Ask anything, launch everything". /ask describes it in full.
• The mic sits in that bar and uses your browser's own speech recognition, which Chrome, Edge and Safari have. In a browser without it, such as Firefox, the mic is hidden.
• There is no send button to reach for. When you stop talking, the words send themselves.
• No mic? Type. Enter sends, Shift+Enter starts a new line. From any other studio page, ⌘K (Ctrl+K on Windows) opens "Search or ask anything…" — type your phrase and the last row, "Ask eroq: “…”", sends it to the home's chat.

What happens when you stop talking

Your first message turns the studio home into a chat: your words on the right, the reply on the left, written out word by word. While Ask works, a small orb turns and short hints show the step — "Reading that…", "Picking the right tool…", "Setting the parameters…".

Then comes the part worth reading. Under the reply, a grey line says what Ask is doing:

You say:    make a vertical clip of a woman walking through neon rain, 10 seconds
Grey line:  Setting up Video · 10 s · 9:16 with your parameters…

About a second later the Video page opens with the prompt written and those settings filled in. When Ask only proposes — you asked a question, the request was ambiguous, or you named an engine your plan does not include — you get a button instead of the jump, Open Video →, with a plan badge next to a locked engine.

Asking is free. Ask prepares and opens; the Generate button stays your click, with the usual credit check and plan unlock.

Say the medium, then the scene, then the settings

Spoken sentences run long and loose. One order keeps them easy to follow:

1. The medium — "a clip", "an image", "a voice line", "a short film". This picks the tool: Video, Image, Voice or Cinema.
2. The scene — subject, action, place, light. Ask writes it into the prompt.
3. The settings — the engine, the seconds, the format, the number of takes.

You say:    make an image of a lighthouse at dusk, square, four versions
Grey line:  Setting up Image · 1:1 · ×4 with your parameters…

Leave the medium out and Ask asks back: "make something cool" gets "An image, a clip, a voice line or some text?" with answer chips. Tap one, or say it.

For settings, use the plain words Ask already knows: "vertical", "Reels", "landscape", "square", "ultra-wide" for the format; "10 seconds" for the length; "with Kling" or "on Veo 3.1" for the engine. Anything you do not say is left as the tool already has it — Ask sets only what you asked for. How Ask picks the engine, length and format lists the words and what each one sets.

One habit pays off with dictation: read the grey line before the page opens. It names the engine, the length and the format that were set. If something you said is missing, add it in a follow-up — or type it.

Commands and voice lines

eroq makes voices as well as pictures, so a spoken ask can mean two different things.

A command asks for something to be made: "make a clip of…", "an image of…". Ask sets up the tool and writes the prompt. Video and image prompts are written in English for the engines, whatever language you spoke.

A voice line is text to be spoken. Say the voice first, then the words after a colon:

You say:    Aria, read this like a voicemail: Hey, it's me. I landed early, call me back when you're out.
Grey line:  Setting up Voice · Aria with your parameters…

The Voice tool opens with Aria selected and your line set as the text. A voice line keeps the language you said it in — it is spoken as written — so a French line stays French, and Ask replies in French:

You say:    Orion, lis ça doucement : la porte était déjà ouverte quand je suis arrivé.
Grey line:  Setting up Voice · Orion with your parameters…

The voice can be a roster voice (Aria or Orion), one of your cloned voices, or a character's voice, each found by name. The instruction goes before the colon and the line after it, so it is always clear which words are meant to be spoken. Prompts in any language covers the language side in more depth.

Because Generate stays your click, you see the line before it renders — worth a glance on a dictated script, since punctuation shapes the read (punctuation-driven voice direction). When you do generate, a line costs 3 credits per 100 characters on Voice One, 2 credits on Voice Turbo.

Follow up by voice

The thread keeps context, so the next sentence can be short. What you said earlier still counts:

You say:    make a clip of a woman walking through neon rain at night, 10 seconds
Grey line:  Setting up Video · 10 s with your parameters…

You say:    now make it vertical
Grey line:  Setting up Video · 10 s · 9:16 with your parameters…

You say:    make it with Kling instead
Grey line:  Setting up Video · Kling 3.0 · 10 s · 9:16 with your parameters…

Shorter phrasing works too: "the same but vertical", "rather with Kling", "make it 5 seconds", "now as an image". Go to Video and come back: the conversation is still there for the session. New chat starts over, and the thread is not synced across devices. Iterating by conversation takes this further.

Going places by voice

Ask does more than set up tools. Say "open my library", "billing", "clone my voice", "my API keys", "show me what the community made" or "new film" and it takes you there.

Say "open the dragon clip" and it searches your creations by the words of their prompts and titles, showing up to five matches as thumbnails; with no match, it opens your library filtered on those words. Characters, elements and cloned voices are found by name. It reads prompts and titles — it does not look inside the images or videos themselves.

Honest limits

• Ask does not render by itself. Nothing is spent until you click Generate.
• eroq has no upscaling, no lip-sync and no inpainting, and Ask says so instead of pretending.
• The thread lives for the session, on that device.
• Ask is part of the studio, not an API endpoint. Developers call the generation endpoints directly, or connect assistants through the MCP server at /mcp.

FAQ

Which browsers support voice commands on eroq?

Chrome, Edge and Safari — the mic uses the browser's own speech recognition. In a browser without it, such as Firefox, the mic is hidden and you type instead: Enter sends, Shift+Enter adds a new line.

Do I have to press send after speaking?

No. When you stop talking, the words send themselves. If a pause sent only half of what you meant, finish it with a follow-up — the thread keeps what you said before.

Can I give voice commands in another language?

Yes. Speak in your own language and Ask answers in it. Video and image prompts are written in English for the engines, while a voice line keeps the language you said it in.

Does a voice command spend credits?

No. Asking is free: Ask prepares the tool and opens it. The Generate button stays your click, with the usual credit check and plan unlock.

How do I make one of my characters say a line?

Name the character, then give the line after a colon — the same shape as "Aria, read this: …". Ask sets up the Voice tool with that character's voice and keeps the line in the language you said it in.

Open the studio, tap the mic and say your first clip — the page opens ready, and Generate is still yours.