Changelog

v1.3.1 — 2026-09-23

Fixes

  • Setup text no longer says "your Mac" on Linux. The models screen and the home-screen banner now say "your computer", which is true on both platforms.

v1.3.0 — 2026-09-22

Linux

  • Arqenne Studio comes to Linux. Ubuntu 24.04 and newer and Debian 13, with Arch to follow through the AUR. Two editions: Standard runs on any 64-bit PC; NVIDIA runs the on-device assistant and voice cloning on your graphics card, many times faster. Same projects, same files as on the Mac.
  • Install it the way Linux installs things. On Ubuntu and Debian, add Arqenne's repository once and the app arrives — and stays current — through sudo apt upgrade or Software Updater, alongside everything else on your system. Switching editions is one apt install; going back a version is one too.
  • Updates that respect your package manager. If apt installed Arqenne, Arqenne tells you a new version exists and names the apt command; it never reaches around apt to replace its own files. If you installed the .deb by hand, the app updates itself in place with an "Install and restart" button. On Arch, it points you at yay -Syu.
  • A smaller download. The Standard edition is about 40 MB, down from over 100 MB in testing, because Arqenne now bundles a video encoder built for exactly what it needs instead of a general-purpose one.
  • One way to start it. arqenne from a terminal, or the applications menu, on every edition.

Storyboard

  • Poster shape. The style bible's frame shapes gain Poster 2:3, the classic one-sheet portrait, and a cover set to it renders 1024×1536 on Auto — and a cover's Output menu now offers every frame shape, Poster included, with Auto following the bible.
  • See what made an image. Under the preview in the asset and shot editors, the prompt that rendered the image now sits with a copy icon in its header, beside its seed, size, engine and render time. Opening the prompt lays it over the picture instead of squeezing it, so the image never shrinks — press Escape or click away to put it down.
  • Copy the prompt behind any image. Right-click a reference sheet or a shot render and choose Copy prompt to put the exact text that rendered it on your clipboard.
  • Start with the cover. Every project's Library now opens with a Cover Art section. Its editor shows your style bible beside the cover, so you can fine-tune the look on one image before rendering a single shot; what you settle there carries into every sheet and shot. The cover's frame shape follows the bible unless you pick a size yourself.
  • Agents can build the cover first. An AI assistant working with Arqenne can now create a project's cover and render it straight from your style bible — and render any reference sheet or shot with the bible folded in automatically, so the look you settled on carries through without being restated every time. It tells you the exact description it sent, so a re-roll after a bible edit is one step.
  • Reuse a project image as a reference. In the asset editor, Choose from project lists every reference sheet and rendered shot in the project; pick one and it becomes the asset's main reference. So a shot that captured a location perfectly can define that location, without saving it out and importing it back.
  • Choose the size of a reference sheet. The asset editor's Output menu now offers the same size presets as the shot editor, and remembers your choice between assets. It applies to sheets rendered from scratch; a render that starts from the starred image keeps that image's size, and the menu says so.
  • Import any image as a reference. The asset editor's reference picker now offers an All files option and accepts JPG and WebP files however their extension is capitalised, so a camera's .JPG is no longer greyed out.
  • Render your storyboard on your own hardware. If you run ComfyUI, point Arqenne at it and shot images and asset reference sheets render there instead of in the cloud, with live progress on the tile as it works. Built-in workflows mean it renders the moment you switch it on, and Arqenne picks the best-matching one per shot unless you pin one yourself.
  • Build an image workflow around the models you actually have. A guided Create workflow step reads the models installed on your instance and assembles a working setup from them — so you are never stuck with a built-in that points at a model file you don't have. Workflows you create behave exactly like ones you import.
  • Push a workflow from Arqenne into ComfyUI. Any workflow can be loaded onto your instance as a real, laid-out graph you can open and edit — the quickest way to see exactly what Arqenne sends, or to start refining from a known-good setup. Each row tells you whether the copy there is current, out of date, or missing.
  • Rendering and chat can share one graphics card. If your image generation and a chat model you host yourself sit on the same card, Arqenne sees how much memory is free and moves one out of the way only when the numbers demand it. When a job genuinely won't fit, you're offered a choice — wait, use the cloud for that one message, or cancel — instead of a silent failure.
  • A simpler shot editor, with named shots and more than one subject. The composition panel is now a single Prompt field, with Framing and Style notes tucked behind optional reveals. Shots can carry their own name — click the title to edit it — and can cast several subjects at once.
  • Curated styles now shape the image, plus new size and quality controls. Picking a style used to be mostly a label; it now feeds into the description that gets generated, and each shot remembers its own. A new Cinema 2.4:1 preset joins the size options, and a Draft/Final control lets you rough shots out cheaply before committing to a final render.
  • Everything about an asset lives in one screen. Opening a character, location, or prop from your Library takes you straight to its editor instead of a read-only summary page. Its shot description, version history, and Delete now sit in the editor itself.
  • Chat while you work on a storyboard. The chat panel opens in the storyboard view from the same side rail as the text editor, so you can talk a shot through without leaving the board. It starts collapsed to keep the canvas wide.

Chat

  • Chat with a model server you run yourself. Point chat at any OpenAI-compatible server on your own hardware and replies stream in exactly as they do elsewhere. A new model selector switches between the on-device assistant, Arqenne Cloud, and your own endpoints, remembering your choice per project. Your own endpoint uses your compute, so it doesn't count against your plan's included usage, and any key you save is encrypted on your machine. A chat is never rerouted to another provider behind your back.
  • Cloud chat runs on a newer generation of assistant model. The assistant can hold far more of your project in view within a single conversation. Nothing changes in how you use it — your existing conversations and plan allowance work exactly as before.
  • A new Image prompt style. A fifth style turns whatever you describe into one polished text-to-image prompt — subject and action, setting, framing and camera, lighting, palette, medium, and mood. It applies when you chat with the on-device assistant, and each conversation remembers the style you picked.
  • Project sources are now a per-project switch. One control decides whether chat draws on that project's knowledge base and whether Arqenne keeps indexing your files in the background. It's on by default, and turning it back on quietly catches up on anything you edited while it was off.
  • Project knowledge works with every chat provider. Questions about your own manuscript, and the assistant's memory of earlier conversations, work whichever provider is answering — the on-device assistant, Arqenne Cloud, or a server you run yourself.

Voice

  • Voice cloning tells you when your computer can't run it. Voice cloning needs at least 12 GB of memory. On a computer with less, the Clone Voice option and the fine-tuned voices now appear greyed out with that reason, instead of silently missing, and Settings › Models no longer offers the Enhanced Text-to-Speech download there — if it was already downloaded, it points you to deleting it to get the space back.
  • Narration reverb has a slightly different character. The audio processing behind narration effects and audiobook export was rebuilt this release. Everything else sounds as it did, but if you had a mix carefully dialed in, give it a quick re-listen.

Improvements

  • Faster, lighter startup. Studio no longer loads its on-device AI models into memory when it launches — each one loads the first time you actually use it. If you write with your own chat endpoint, starting the app no longer costs you gigabytes of memory for a model you never call.
  • Projects reopen right where you left them. Whether you were last writing, working on audio, or laying out a storyboard, opening a project takes you back to that view instead of always starting in the text editor. Each project remembers its own place.
  • Reclaim the memory your project knowledge base was holding. The model behind project search used to keep roughly 2.5 GB reserved until you quit, and the unload button had no effect on it. You can now free it from the models screen; the next search loads it again by itself.
  • Connected AI agents can drive the whole creation loop. An AI agent working with Arqenne can now create and update projects, read and write your documents, narrate them paragraph by paragraph, and build out storyboard shots and library assets — including asking for images. Narration made this way lands in the Audio editor and in audiobook export like anything else.
  • Custom endpoint switches explain themselves. A saved endpoint's two switches read as one unlabelled pair, so it was easy to misread what the second one did. Each is now captioned, and hovering either explains what it controls and what turning it off actually does.

Fixes

  • Exported audiobooks now carry chapter markers. M4B and M4A exports were being written without any chapters, so players had no way to jump between them. Every export now includes per-chapter markers that Apple Books and other players recognize.
  • Audiobook speed no longer changes the pitch. Exporting faster or slower than 1× used to raise or lower the voice along with it, even though speed is meant to leave the voice untouched. The supported 0.5×–2× range now keeps the original pitch.
  • Quitting no longer freezes the app. Closing Studio while it was still downloading a model for the first time, or still working in the background, could lock the window until macOS offered to force quit. Quitting is now immediate, and you're asked whether to wait or quit right away when work is still in flight. Anything interrupted simply picks up next launch.
  • Prompt edits in the Shot Editor now stick. Wording you typed into a shot's prompt was thrown away — generating an image, choosing a frame, or re-anchoring a shot reset it to the automatically assembled text. What you see in the box is now exactly what gets sent.
  • The project knowledge base can no longer download itself twice. Starting the download while one was already running could fetch a second copy of the same multi-gigabyte file. Only one download can now be in flight at a time.

v1.2.0 — 2026-06-23

Storyboard

  • Turn your manuscript into a visual storyboard. A new Storyboard workspace lets you plan your book or script shot by shot. Build a library of your characters, locations, and props; anchor each shot to the exact passage it illustrates; and lay the whole thing out as a grid you can play through with your narration.
  • Generate consistent scene art. Create artwork for each shot that actually looks like the characters, places, and props you've defined — bind them to a shot and their look carries into every new image. Star a favorite version of any asset and the rest of your storyboard follows it. Lock an image's settings to keep the same look while you change the pose or angle, or reroll for fresh variations. Guided prompt forms assemble a clean description for you, with a live preview before you generate. Image generation uses the same plan allowance as your other AI features.
  • Export your storyboard as a video. Turn a finished storyboard into a shareable video. Each shot's image is held for the length of its narration and stitched together in playback order, with your generated voiceover as the soundtrack. Preview it, choose a resolution, pick where to save, and Studio renders the finished file.

Improvements

  • Faster startup. Studio now loads significantly less code before the first screen appears, split into pieces that update and cache independently — launching the app and installing updates both feel snappier.
  • Fully self-contained launch. Interface fonts now ship inside the app instead of loading from the internet, so the app looks right immediately even offline — and makes zero third-party requests at launch.
  • No more dark flash at launch. If your Mac is in light appearance, the window no longer flashes black while the app starts.
  • Smoother AI chat. Long conversations no longer stutter or lag while the assistant is streaming a reply.
  • Keyboard-friendly dialogs. Tab now stays inside open dialogs and cycles through their controls, and closing a dialog returns focus to where you were — no more focus escaping into the background.
  • Snappier typing in long documents. The editor's per-keystroke bookkeeping now skips text that didn't change.

Fixes

  • Screenplays now display in their proper typeface. The screenplay editor's Courier Prime font was silently missing from the shipped app, so scripts rendered in a generic monospace font. It now loads correctly.
  • An error in one screen no longer takes down the whole app. If something goes wrong, you'll see a recovery card with "Try again" and "Reload app" options instead of a blank window — and your work elsewhere stays intact.
  • Last-second edits are no longer lost when switching files. Changes made in the moments before clicking to another file (or closing it) are now saved instead of silently dropped.
  • Project settings no longer reset each other. Saving project settings — or simply changing the editor page width — could erase your saved image-generation defaults and re-enable audio enhancement after you'd turned it off. Every settings save now preserves everything you've configured.
  • A damaged document no longer crashes the app. If a document's primary data is ever corrupted, Studio now opens it from its Markdown copy instead of failing every time you open that file.
  • Voice model downloads now report failures honestly. A failed model download or delete previously showed "ready" anyway, leaving voice features broken with no explanation. Failures now show a clear error state.
  • Undo can no longer erase a freshly opened document. Pressing Undo immediately after opening a file could revert it to an empty page — and auto-save that. Opening a file no longer counts as an undoable step.
  • Chat replies stay in their conversation. Switching conversations while the assistant was mid-reply could deliver the rest of the response into the wrong chat. Replies now stay bound to the conversation that asked.
  • Audio players no longer fight each other. Starting playback in the audiobook export preview while Play-All was running could make both skip ahead and interleave; pausing a preview also restarted it from the beginning. One player now owns playback — starting a new one cleanly stops the other, and pause picks up exactly where you left off.
  • Screenplay imports keep everything you wrote. Importing a .fountain script used to silently drop notes, page breaks, lyrics, centered text, dual dialogue, and several title-page fields. All of it now survives import and export, exactly as written.
  • Nested checklists no longer corrupt saved files. A checklist inside a list item (or a list inside a checklist item) could be written incorrectly to the saved document. Saving now preserves the structure exactly.
  • Quitting right after typing no longer loses your last edits. Changes made moments before closing the window are now written to disk as part of shutdown.
  • Login errors now tell you what's wrong. Wrong password, no connection, and server trouble each show their own message instead of everything claiming "invalid credentials."
  • Sidebar drag-and-drop is disabled for now. Dragging files appeared to reorganize them, but the change was never saved — it silently reverted on reload. The drag affordance is off until real file-move support lands.

v1.1.1 — 2026-05-03

Fixes

  • Voice Settings panel now scrolls. The right-side Voice Settings panel could clip the speaker form and error messages off the bottom of the screen with no way to reach them. The panel now scrolls within its bounds, with the heading staying pinned at the top.

v1.1.0 — 2026-04-29

  • Arqenne MCP. Your AI assistant can now generate speech with your Studio voices — locally, no API keys, no per-call charges. Works with Claude Desktop, Claude Code, Cursor, and opencode. Available on every plan; Free includes one voice, paid tiers unlock all 28.

Fixes

  • New Project wizard now opens reliably from a scrolled home screen. Previously, if you scrolled down on the home screen before clicking New Project, the wizard could open above the visible area and you'd have to scroll back up to see it. It now stays anchored in view no matter where you've scrolled.

v1.0.0 — First public release

The first public release of Arqenne Studio for Apple Silicon Macs.

Writing

  • Rich-text editor with slash commands, and a floating formatting toolbar (bold, italic, underline, links, color, highlight, "✦ Suggest").
  • Screenplay mode. Write scripts in standard Fountain format (.fountain) with proper scene headings, character cues, and dialogue formatting — set per file at creation, alongside regular Markdown manuscripts in the same project.
  • Resizable inline images with alignment controls.
  • Dual-format persistence. Every document is saved as both structured JSON (perfect fidelity) and clean Markdown (portable, diffable). Auto-saves on a 2-second debounce.
  • Local AI Assistant. Runs entirely on your Mac's Apple Silicon to power suggestions in the editor's selection toolbar. The app auto-selects a smaller or larger model based on your available RAM.
  • Project knowledge base. Ask questions about your own manuscript — Studio indexes your project locally so the assistant can pull from your worldbuilding, character notes, and prior chapters. Your text never leaves the device.
  • Arq design system. Warm parchment in light mode, deep umber in dark — auto-switches with your system theme.

Voice

  • Two on-device voice engines. Your audio never leaves your Mac.
    • Standard Text-to-Speech — fast, lightweight, instant in-app preview.
    • Enhanced Text-to-Speech — studio-quality narration, generated entirely on your Mac's own hardware.
  • Voice cloning. Clone any voice from a short reference clip with the Enhanced engine — no upload, no third-party account.
  • Paragraph-level audio editor. Generate, replace, and play back narration paragraph by paragraph. Stale-audio detection flags clips that have drifted from the edited text.
  • A growing voice library. 28 voices on Standard TTS, plus voice cloning on Enhanced TTS — more on the way.

Export

  • PDF export on every plan. Ship your manuscript anywhere.
  • Audiobook export on paid plans. Turn your narration into a finished file:
    • Starter — MP3
    • Creator — MP3 + WAV
    • Studio — MP3 + WAV + full M4B audiobooks with chapter markers

Offline-first

  • First-run model downloads with a progress UI; cached locally and skipped on relaunch.
  • Works offline after first launch. Writing AI, voice synthesis, and projects all work with the network off. Cloud chat features need a connection.

Plans

  • Free, Starter, Creator, Studio. A single shared budget per user across writing and voice.
  • No surprise bills. 80% usage alert, hard stop at 100%.
  • Gift codes. Single-use codes unlock an upgraded tier for a set number of days.

Security

  • Encrypted at rest. Your authentication tokens are encrypted on your machine.
  • Atomic writes. A crash mid-write can't corrupt your session.
  • Signed auto-updates. Every update is cryptographically verified before it installs.
  • Privacy-first crash reporting. We log a user ID only — never your email or document content.