Should I use voice input or typing when directing an AI video production workflow?

Use voice for the directorial passes — scene direction, mood notes, on-the-fly revisions, and jamming with the invideo agent the way you'd talk to a DOP on set. Switch to typing for the precision inputs — exact prompt blocks, character sheet names, parameter values, and the locked style language you don't want a transcriber to mangle.

The invideo agent is an agentic video creation tool that takes both voice and typed input across the whole crew of sub-agents you spin up (creative producer, storyboard, DOP, costume, production design), so the question isn't which to commit to — it's which mode fits which task.

Use voice when you're directing, not specifying. Speech-to-text is explicitly recommended as a faster alternative to typing for agent jamming — describing a scene, giving mood notes to a costume sub-agent, talking through coverage with a DOP sub-agent, asking for status. Hridaye, invideo's creative director, describes the loop as "pretty much exactly like how I would talk to my DOP on set or how I would talk to my DA on set" — natural on-set language consistently outperforms formal prompt-engineering syntax. Voice also keeps you in the film: "I wanna talk about my shot like this because then I can keep thinking about my entire film in my head without breaking it." That flow state matters when you're running 6–8 sub-agents in parallel and context-switching is the real tax.

Type when accuracy is load-bearing. Anything where a transcription slip changes the output — the locked style block applied to every prompt, exact hex values in a tonal mode, character sheet panel references, negative-prompt lines ("not live-action, not photorealistic"), aspect ratio and lens specs, the four pre-production answers (character, entity, prop, deliverable format) — gets typed and reviewed before you hit send. The same goes for the initial treatment / context upload: that document governs every downstream frame, so it lives as a written file, never dictated.

The hybrid pass is the actual workflow. Dictate the direction ("hold on him right up till he lunges, no back-and-forth cutting"), let the invideo agent draft the structured prompt with your style block attached, then read it before approving — using shot-by-shot approval mode means you catch any transcription drift before credits are spent. For granular fixes (close-up crops, swapping one panel in a character sheet), take manual control, type the change directly, and log the result back so the sub-agent's memory stays accurate.

Practical defaults across the production:

• Pre-production briefing, world-building chat, costume mood direction → voice.
• Treatment doc, style block, character sheet prompts, negative prompts, parameter values → typed.
• Rough-cut critique, status summaries, "what's working, what's not" passes → voice.
• Surgical fixes and manual overrides → typed, then re-logged to the sub-agent.

These are starting points — what works depends on how you think and how clean your transcription is. Worth noting: voice-native direction of visual agents is still emerging across the category, so the hybrid pattern is the safe bet today regardless of which models you're routing to.

Watch some of these to see what works for you:

I wanna talk about my shot like this because then I can keep thinking about my entire film in my head without breaking it.

— Hridaye, invideo's creative director