Every Character Gets Their Own Voice

Here's what's new in Voice Creator Pro v2.2.7. Last release wired language models into VCP for Viral Clips. This one points them at your books. Everything below is live on both the desktop app and in VCP Cloud.

Casting

Assigning a voice to every character in a novel used to mean reading the whole thing first and clicking through it line by line. Now a language model does it, through the same AI Models connection Viral Clips uses.

• It finds the speakers. The manuscript is read in parallel, nicknames and aliases collapse into one character, first-person narration is linked to the narrator, and a repair pass picks up quotes the first pass missed.
• It designs a voice for each one. The traits it picks during casting go straight into a voice design, one job per character. Every prompt is editable and you can re-roll any voice you don't like.
• It marks the mood. Emotion is tagged per span so the narration is read with feeling, on a full cast or on a single narrator.
• Pauses between speakers land automatically when the voice changes mid-segment.

You see a cost estimate before a run starts and the tokens it actually used when it ends. On desktop you bring your own key (Gemini, OpenAI, OpenRouter, DeepSeek, or any OpenAI-compatible endpoint). In the cloud it's included.

Emotion Steering

There's a new emotion engine for cloned voices. Instead of acting the line in a different voice and converting it back to yours, it steers your voice while it generates. It's faster, and it stays much closer to your recording.

Subtitles Improvements

• Save your own caption styles as presets, on desktop and in the cloud.
• Redesigned settings with a large live preview, a preset gallery, and Text, Layout, Box, and Motion tabs.
• Real box controls. Ink-tight backgrounds, separate side and top-bottom padding, corner radius, and per-caption outline and opacity overrides.
• After Effects export now writes one layer per phrase instead of one per word, and carries the glow across.
• Transparent overlays export as ProRes 4444 XQ or QuickTime Animation with premultiplied alpha, at the footage's own frame rate and at 1x, 2x, or 4x its size, for dropping onto a bigger timeline.
• Edit a segment's start and end times straight from the list.

Also

• Voice Changer takes an hour of audio now, up from five minutes, and a conversion keeps running if you leave the screen
• Audible-ready export for audiobooks, meeting ACX loudness rules
• Imports skip the filler. Copyright pages, tables of contents, and Project Gutenberg licence text are left out, and listed so you can add any back
• Numbers and dates read correctly in 11 more languages, including Hindi, Chinese, Greek, Malay, and Filipino
• A warning before you clone when your text is in a different script or a language the model doesn't support
• Copy pronunciations from one book into another
• Flash Attention as an opt-in toggle for AMD RDNA 3 and RDNA 4 cards
• Half-downloaded models repair themselves instead of failing forever
• Many smaller fixes across mobile, recording, and the editor