Your characters can now speak out loud with full text-to-speech, including custom voice design. Each character now keeps memories. You can stop generation at any moment and respond. And custom instructions let you tell the AI how you want it to write, either per story or across every story you ever create.

Important Note: The default model for "auto" has been changed, if you don't like how the auto model speaks, try another one in the settings! The previous auto router often used Gemini 2.5 Flash Lite from Google if you would prefer that model again.

Voice — Hear Your Story Out Loud

This is the headline feature of this update. Every line in your story can now be spoken aloud, with a distinct voice per character and the narrator.

What's included:

• A voice for every character. Assign a voice to each character and the narrator from the voice slideover. Lines are spoken in the right voice as the scene plays out.
• Designed voices. Describe the voice you want in plain language — "warm, gravelly, late 40s, slight British accent" — and the AI generates it. Preview, tweak, and save voices you like.
• Live volume and pitch. Drag the volume or pitch sliders for any voice and the reference loops while you adjust, so you can hear changes in real time before saving.
• Per-character tabs and settings. The voice slideover gives each character their own tab with per-slot settings, so a long cast stays organized.
• Stop and volume controls in-story. A Stop button and live volume slider stay visible while the story is generating and playing — no digging through menus.
• Voice on Depict. Optionally have the AI voice your Depict frame descriptions so even your generated scenes are narrated.
• Setup that gets out of the way. Voice is off by default with a one-time setup offer when you start a story.

Voices are also bundled into your .drsf story exports, so when you share a story your designed voices come with it.

Character Memory — Remembered From Their Own Point of View

We've reworked how the AI summarizes your story so it stays coherent over long adventures and feels more personal.

• Concise bullet summaries instead of narrative prose, so the AI can pull facts faster and waste less context on flowery recaps.
• Character-perspective memories. Each character now keeps a bulleted list of what they remember and care about — written from their point of view. The grumpy mentor and the wide-eyed apprentice will remember the same scene very differently.
• Memories build on each other. Each character's description, personality, and existing memories are now fed into the summary prompt, so new memories stay consistent with who that character has always been.
• Regenerate on demand. A "Regenerate latest" button in Story Settings lets you redo the latest summary in place if it missed something important, with a clear credit cost chip up front.

The result: longer stories that stay coherent, and a cast that feels like they each have their own head full of memories rather than sharing one global narrator brain.

Stop Anytime and Respond

You no longer have to wait for the AI to finish a long generation before jumping in.

• A Stop button is always visible while a generation is streaming or playing.
• Press Stop and the story truncates at the last line you actually saw.
• Your input frees immediately. You can type your response the moment you hit Stop.
• TTS gets cancelled cleanly. Any voice lines that hadn't started playing yet are cancelled end-to-end, so you're never charged for lines you didn't hear.

Custom Instructions — Tell the AI How You Want It to Write

A new way to shape every generation the AI makes for you.

• Per-story custom instructions. Tell the AI things specific to one story — pacing preferences, narrative voice, things to avoid, recurring themes. (These are also included when uploaded!)
• Account-level custom instructions. A "global" set of instructions that apply to every story you create. Great for things like "write in present tense" or "skip morality lectures."
• Both at once. Story and account instructions stack, so global rules apply everywhere and per-story instructions layer on top.
• 5,000 characters of room for each, so you can be as detailed as you want.
• Applied to impersonations too. When the AI speaks as a character on your behalf, your custom instructions guide it there as well.

Quality of Life

A lot of smaller improvements landed alongside the headline features:

• Edit character responses by clicking the story text while in edit mode. No more diving into menus to fix a line.
• Edit story name and setting directly from Story Settings.
• DeepSeek V4 Pro added as a selectable text model.
• Reworked outfit and appearance designer with a clear new / edit / upload flow.
• Story thumbnails auto-seed from your most recent scene when uploading.
• Smoother scrolling on the story grid while images are still loading.

Stability and Behind-the-Scenes

We've put a lot of work this cycle into making the app feel more reliable:

• Better error attribution so when something fails we can actually find and fix it
• Smarter retry logic for transient AI failures, including a clearer "switch models" prompt when a model keeps failing
• Cleaner cancellation paths for in-flight image and voice jobs

As always, thank you to everyone in our Discord who tested early builds of Voice and helped us shake out the issues. Voice in particular is a big feature and your feedback shaped where it landed. Keep it coming — we've got even more on the way.