Slatesslates

Changelog

Every update ships automatically. Open Slates and it installs in the background.

v1.5.6September 8, 2026

Your own voice, speaking your script

  • NewVoice generation. Switch the prompt box to Audio, pick Inworld TTS-2, type the words you want spoken, and press Generate — what you type is what gets said, and the character count is the price. The Voice control gives you four ways to decide who says it: ready-made presets you can play before choosing, any audio clip already in the project, a character's voice, or a description you write in plain words ("a warm, unhurried man in his fifties") that gets built into a voice on the spot.
  • NewUse your own recording. Voice → Clips → Import voice clip takes an audio file straight from your computer into the project, where you can play it back and then use it as the voice for a take. Importing costs nothing; only generating spends credits. Use a clean recording of one speaker you have permission to use.
  • NewCharacters can own a voice. Add voice on a character card in the gallery attaches a recording or a generated take to that character, and it stays attached — so that character turns up under Voice → Clips → Characters every time you write a line for them.
  • NewLTX-2.5 joins the video line-up, in both of its seats.
  • NewShots. Everything you set up for one generation — the line, the references, the model, the settings and every take you have made of it — is now a single thing you can save, come back to, and fire again. The storyboard is that list: Board is for arranging it, Document is for reading it, and there is no separate shot list to keep in step.
  • ImprovedDocument reads like a script again. Script now shows the scene headings, the speakers and the words and nothing else — no delivery notes, no model or price detail, no gaps where a warning used to sit — so you can genuinely read it aloud. Shots still shows everything, and Custom lets you turn each layer on or off yourself, including a new Warnings layer for missing models, unresolved mentions and lines too long for their cut. Hiding a layer only changes what you see; your text and your generation checks are untouched.
  • ImprovedAsk the app how the app works. The in-app assistant, the CLI and the Claude/MCP connection now answer "how do I do X" from the actual product manual, naming the real buttons and tabs, instead of reasoning from memory about a screen it cannot see.
  • ImprovedDelivery is optional again. It is a performance note for you and your collaborators, not a label every line needs, and it is never quietly bolted onto what a voice model reads aloud. If you want a spoken cue, write it into the words themselves the way that model's prompting guide describes.
  • FixedVoices made from a recording or a description were being turned away before they ever reached the provider — only the ready-made presets worked. All three ways of choosing a voice now generate.
  • FixedGenerated audio would not play back. Clips loaded with no length, so the player showed an endless track, the counter never moved and seeking hung. Audio now streams the same way video does, and a clip that genuinely cannot be played says so instead of failing silently.
  • FixedText-to-speech was showing Loop and Wording, which belong to Sound Effects. Voice is the only control it needs.
  • FixedThe Studio Agent no longer thinks forever if a reply stalls on the way back. It gives up after 90 quiet seconds and offers to retry, instead of leaving a spinner you have to guess about.
  • FixedImported audio keeps its filename instead of being listed as "Imported file", and search boxes stop shifting under the pointer when you hover or click them.
v1.5.5August 27, 2026

MiniMax H3 — sound you direct, and a seat that renders in seconds

  • NewMiniMax H3, the first model in Slates where the sound is something you write rather than a switch you flip. Dialogue and action sounds go on the beat they land, ambience runs underneath, and a score only the audience hears sits in its own section — all three generated with the picture in a single pass, in eleven languages. It's also the only model where you say how much of a reference should survive: keep this person exactly, take only the brick wall from that café, or move the worn leather of one jacket onto a different character. 5 to 15 seconds, up to 4K, with nine reference images plus reference video and audio.
  • NewMiniMax H3 Max, the same model tuned for speed. On the same five-second 768p prompt it came back in about 5 seconds where standard H3 took about 57 — fast enough to iterate in front of a client instead of waiting on a render. Start a shot from a still and it animates it just like H3 does, first frame and last frame both. It's the premium seat, not the budget one: it stops at 768p and costs more per second than H3 at the resolution they share. The one thing it won't do is references — where H3 reads your characters, styles and environments alongside the prompt, Max works from the prompt and your frames. A Realtime badge marks it in the model picker.
  • ImprovedSlates turns off MiniMax's prompt rewriter on every generation. The model's API will quietly rewrite your prompt before generating; we send yours instead, because what you see in the composer should be what the model reads. The trade is that nothing will pad a thin prompt — the prompting guide has the shape of a full one.
  • ImprovedH3 reads five reference images for free and charges for each one after that, so the price you see updates as you attach them — the Generate button, the length menu and the model picker all count the same way. More references usually make a shot worse, not better; attach what it needs, not the maximum.
v1.5.4August 16, 2026

Sharper Seedance 2.5 guidance

  • ImprovedBetter advice when an AI agent is driving Seedance 2.5. Two corrections: the resolution ceiling is now described as what Slates actually offers (480p and 720p), and the old "draft at 480p first" tip is gone. A 480p pass doesn't preview a 720p one — every generation is a fresh roll of the dice, so the 720p version is a different take, not the same shot at higher quality. Explore with shorter clips instead. Length is what moves the price.
v1.5.3August 16, 2026

The manual copy button works again

  • Fixed"Copy the manual for your AI" in the help menu was failing silently on every click. It now copies properly, and it copies more: the full manual plus every documentation page, so pasting it into ChatGPT or Claude teaches it the prompting guide too, not just the feature list.
  • ImprovedSeedance edit requests are recognised more reliably. Prompts using "insert", "delete" or "modify" could be treated as a fresh generation instead of an edit; all three are now handled.
v1.5.2August 10, 2026

Audio, Seedance 2.5, and references beyond images

  • NewAudio is now a full media type alongside images and video. Generate music, voiceover, and sound effects right in the composer, then drop them onto the timeline. Audio gets its own gallery tab with waveform cards and inline playback, and you can drag your own MP3s and WAVs in from your desktop.
  • NewSeedance 2.5, a second Seedance seat built for long takes. Thirty-second single shots instead of fifteen, up to 50 references in one generation, and audio references the model actually listens to. It runs at 480p and 720p, so Seedance 2.0 stays the default when you want 1080p or 4K.
  • NewSeedance 2.5 Edit. Point it at any clip in your project, describe the change, and get the edited clip back at the source length. It joins Kling O3 Edit and Omni Flash Edit in the edit lane.
  • NewReferences beyond images. You can attach reference video and reference audio now, not just stills, up to 10 of each on Seedance 2.5. Drive motion from one clip and a voice from another, and the model reads them in the order you arranged them.
  • NewMove and copy assets between projects. Send a clip, an image, or a whole character to another project from the selection bar, the right-click menu, or by dragging onto a project chip. Characters and environments can bring their images across in one step.
  • NewA help menu in the title bar. One question mark with everything support-shaped behind it: the quick start, the full reference, the tutorial replay, generation logs, and Discord. It also copies the entire Slates manual to your clipboard, so you can paste it into ChatGPT or Claude and ask it anything about the app.
  • NewDrag to reorder the reference rail. The number on each tile is literally the number the model is told, so you decide exactly what image 2 means.
  • ImprovedThe whole prompt surface lives on one bar. Every generation setting renders inline and overflows into a More menu that measures the space actually available instead of guessing. The old settings panel is gone.
  • ImprovedA rebuilt model picker with search, vendor tiles, and capability chips read straight from each model's real limits, so what a model offers and what the chip says can never disagree.
  • ImprovedSee exactly what gets sent. The prompt preview renders the request builder's own output, so the text and images you see are the ones that go to the model. Any mention or tag that matched nothing is named instead of silently disappearing.
  • ImprovedDuration is a slider now instead of a thirteen-row menu, with the price for each length shown as you drag.
  • ImprovedA chrome and elevation pass across the app: quieter utility controls, a scarcer accent color so it means something when it appears, and popovers that land where you clicked.
  • ImprovedOut-of-credits messages point at Settings, Account instead of API key setup, and the credit badge stays visible at a zero balance so the buy path is there when it matters.
  • ImprovedMotion transfer, relocate, and lip-sync-on-video are meaningfully cheaper on the AI-face route. Same generations, same quality, lower credit cost.
  • FixedMP3s with embedded cover art no longer land in the video tab. The artwork read as a video stream, so the song filed itself under video, resized the timeline canvas to the album art, and took its frame rate from it.
  • FixedPasting an image into the prompt box now creates a real asset with a code, a thumbnail, and a folder, instead of a loose file nothing could list or delete.
  • FixedThe gallery no longer kicks you out of the folder you were working in when you paste, drop, delete, or move something.
  • FixedDatabase upgrades are crash-safe. Table rebuilds run in one transaction and derive their columns from the live database, so a failed upgrade rolls back completely instead of leaving things in a state that quietly broke unrelated features on every later launch.
  • FixedThe spend total shown for a project counts each generation once, even after you copy an asset into another project. Deleting an asset also cleans up its waveform and thumbnail files now, instead of leaving them behind.
  • FixedAudio clips on the timeline no longer come back silent and zero-length when the source has no frame rate of its own.
v1.5.1July 23, 2026

One identity image, cleaner frames, and a design pass

  • ImprovedCharacters and environments now use a single reference image each. One identity image per character, one reference image per environment, so every mention adds exactly one picture. Faces stay consistent and you stop burning reference slots on every shot.
  • ImprovedSet a character or environment straight from the gallery. Right-click or drag any image to bind it instantly, with a separate Generate path when you want to create a new one.
  • ImprovedFirst and last frames are now explicit choices on the prompt bar. Pick which image starts a shot and which ends it, with last-frame support wired up for Kling, Seedance, and Veo. No more guessing from the order you added them in.
  • ImprovedA surface, light, and motion pass across the app: softer shadows, film grain, a gentler accent glow, and smoother transitions when you click and drag.
  • FixedDragging clips onto the timeline works reliably now. A drag that started inside the app sometimes delivered no drop, so clips silently refused to land. That is fixed.
  • FixedThe gallery no longer flashes black while you scroll, select clips, or open one fullscreen.
  • FixedThe editor preview no longer shows the next clip bleeding through the black letterbox bars.
  • FixedThe API Keys section in Settings now shows up for accounts re-enabled to use their own keys, so those users can see and manage the keys their generations already run on.
v1.5.0July 18, 2026

Sound on the timeline

  • NewAudio tracks in the editor. Drop in music, voiceover, or AI audio and it plays back in preview and mixes into the exported MP4. Add as many audio tracks as you need, so a music bed, a voiceover, and effects each get their own lane.
  • NewA volume fader on every track, plus a master output fader for the final mix. Both run from silent to +12 dB of boost, and both apply to preview and export, so what you hear is what you render.
  • NewA guided tour on first launch: create a project, pick your models, start a storyboard, export an MP4. Plus tips the first time you open Storyboards or the Timeline. Replay it any time from Settings, Tutorial.
  • NewDrag clips straight out to your desktop. Drag any gallery card into Explorer, Finder, Resolve, Premiere, or anything else that takes a file. No export step, no save dialog.
  • NewTrim and Split workbench. The trim tool now splits too: scrub a single bar, drag split points, or auto-split with overlap, and every chunk becomes its own clip linked back to the original.
  • ImprovedSeedance now works as a full edit engine. 'Edit with AI' is one umbrella across Seedance, Kling, and Omni Flash, and clips outside the 3-15 second window automatically pick the engine that can handle them.
  • ImprovedThe project view switcher moved into the title bar, and gallery card sizing is now continuous. Hold Ctrl or Cmd and scroll over the grid to zoom, and the grid no longer jumps when you select items.
  • ImprovedA styling pass across the app: consistent controls, a cleaner marker color palette, and prompting tips that always match the model you picked.
  • FixedThe timeline handles multiple tracks properly now. Every project starts with video and audio tracks, but the track headers didn't stay locked to their lanes while scrolling, so headers drifted out of line and the playhead and markers fell out of alignment. The two audio lanes also couldn't accept anything until now. If you opened the editor early on and it felt broken, it's worth another look.
v1.4.0July 10, 2026

Gemini Omni Flash: fast, cheap video with sound built in

  • NewGemini Omni Flash, a fast, low-cost video engine with native synced audio included in the price. Text-to-video, image-to-video, and reference-to-video with up to 7 reference images, at 720p and 3-10 seconds. The cheapest way to draft shots, test ideas, or get sound and picture in one generation.
  • NewOmni Flash Edit, prompt-only video editing. Change a prop, effect, environment, or lighting in an existing clip with a single instruction, footage-synced. The lowest-cost edit option, alongside Kling O3 Edit for reference-image swaps.
  • NewTrim clips to fit. Trim any video to an exact window from the clip menu (or via the agent). The trimmed result saves as a new clip linked to the original, so a long clip fits an edit model in seconds.
  • NewImport your own footage through the agent. Reference uploads now accept video files, probed on import so duration and dimensions are available immediately.
  • ImprovedClearer errors when an edit prompt can't be satisfied, real cost quotes on agent-driven video edits, and automatic duration backfill on older imported clips.
v1.3.0July 7, 2026

Clearer credits and pricing

  • ImprovedStreamlined the billing display across the app for a clearer, more consistent view of your balance and per-generation costs.
v1.2.2July 5, 2026

New image models, video editing, and exact pricing

  • NewThree new image models. Nano Banana Pro for hero frames and crisp typography, Nano Banana 2 Lite for fast, cheap 1K drafts, and GPT Image 2 for razor-sharp text, character sheets, and grids (with a medium or high quality toggle). All three work for both generating and editing, and the Studio Agent can drive them too.
  • NewEdit any video clip with a single instruction. New Kling O3 video-to-video editing lets you swap a character, change the environment, or restyle a shot in one pass, with no masking. The original motion, camera, and audio are preserved, and only what you describe changes. Chain edits as many times as you like.
  • NewSeedance 2.0 as a premium engine for lip-sync and motion transfer. A single-pass path where the speech is generated inside the video itself (a video source even keeps its own voice) and motion is driven natively from a reference clip, with natural audio included.
  • ImprovedPricing is now exact to the option you pick. The price you see always equals what you're charged for that exact combination. Kling audio is now priced on its own, and video-reference work and video edits are billed precisely on the seconds they actually use.
  • ImprovedRefreshed model picker. Models and tools are grouped into clean tabs so the one you want is easier to find.
  • FixedTool prices shown on the model panel now match the Generate button. Lip Sync and Motion Transfer previously showed a per-block rate on the model chip while the button showed the total.
v1.2.1July 3, 2026

The Studio Agent

  • NewMeet the Studio Agent, a conversational agent built right into Slates. Tell it what you want to make and it plans and runs the whole production for you: script, characters, images, video, and quality checks. You approve the credit cost once, then it works. Open it from the top bar or press Ctrl and period.
  • NewConnect Slates to Claude and other AI agents. Slates can now be driven from the outside through MCP or the command line. Connect once in Settings with a magic link, and tools like Claude can create projects and generate images and video straight into your gallery, exactly as if you clicked the buttons yourself.
  • NewMove your whole library to another computer. Settings now has Backup and Migrate. Export everything (projects, media, and references) to a single file, then import it on another machine and it all lands in the right place. For security your API keys and login stay on your original machine, so you re-enter those once on the new one.
  • NewKling O3 Pro is here, with native audio and 4K. Kling can now generate in native 4K on every tier.
  • NewSeedance 2.0 can now generate with real people. When it detects a real face it asks for your consent and shows you the price before it runs.
  • NewEvery image, video, and clip now shows a short code like IMG-A12 on its card, so you and the Studio Agent can point to the exact asset without digging through the gallery.
  • ImprovedSeedance 2.0 now runs on a single model with a clean resolution ladder. Just pick 480p, 720p, 1080p, or 4K. This replaces the old Fast/Standard and Economy/Priority tiers.
  • ImprovedBig cleanup to character sheets and references. Character sheets are now one simple mode with an optional style tweak, environments are a single establishing image instead of a grid, and the number on each reference is exactly what the model sees. More consistent results with less fuss.
  • ImprovedKling V3 Standard is now the default video model.
  • ImprovedNew generations now land in whatever gallery folder you have open, instead of the project root. Edits, extracted frames, lip-sync, and motion-transfer follow their source.
  • ImprovedYour credit balance now lives in the top bar, with a one-click Enable auto-topup button when you're running low.
  • FixedBase accounts are no longer locked out of Seedance and other credit-based models. A layout bug hid the control that let you switch, and that's fixed.
  • FixedRecovered images now show up correctly in the image gallery with proper thumbnails, instead of appearing broken in the video gallery.
v1.2.0July 3, 2026

The Studio Agent, and Slates opens up to outside agents

  • NewStudio Agent: describe the production you want and an in-app agent plans and runs the whole pipeline, from script to characters to images to video, after one itemized credit approval. Toggle it from the top nav or with Ctrl+. Its generations run through the same pipeline as a button click, so progress cards, gallery landing and per-model behaviour are identical.
  • NewConnect Slates to external AI agents. A loopback control server (127.0.0.1 only, bearer-auth, single instance) lets the Slates CLI, the MCP server and clients like Claude drive the desktop app: create projects, generate images and video, run lip-sync and motion transfer, all landing in your gallery like a manual run. Connect from Settings → Agent Control with a magic link.
  • NewMove your library between computers. Settings → Backup & Migrate exports projects, media, thumbnails and references into one .slatespkg file and imports it on another machine, with every stored path rewritten and nothing colliding.
  • NewKling O3 Pro joins the model picker, native 4K arrives on every Kling tier, and every asset now carries a short readable code (IMG-A12, VID-V3) so you and the agent can point at the same thing.
  • ImprovedSeedance 2.0 runs on a single 480p / 720p / 1080p / 4K resolution ladder, the default video model is Kling V3 Standard, generations land in your active gallery folder, and the credit balance moved to a top-nav badge with an auto-topup link.
  • FixedBase-tier accounts are no longer locked out of Seedance and credits-only models, recovered images land in the image gallery with correct thumbnails, and launching Slates twice focuses the open window instead of racing the database.
v1.1.4April 28, 2026

Intel Mac fix, drag-to-reorder folders and scenes

  • FixedIntel Macs can launch Slates again. The 1.1.3 build was missing the Intel version of one of the app's native components, so the app hung on startup on Intel hardware. Apple Silicon Macs were not affected.
  • NewDrag to reorder folders in your project Library. The new order persists.
  • NewDrag to reorder scenes inside a storyboard.
  • ImprovedFloating prompt box: the settings panel now has its own scroll. References, source images, lip-sync controls, and the prompt textarea stay pinned while you scroll through settings.
  • ImprovedLip-sync avatar model selector moved into the left column of the prompt box settings.
v1.1.3April 14, 2026

Base and Pro tiers

  • ImprovedSettings panel now reflects your account tier. Pro accounts see the full API Keys section exactly as before. Base accounts see an upgrade prompt instead, since base plans run on credits only and hiding the key inputs avoids confusion.
  • ImprovedLicense validation now tracks tier and vault ownership so the app stays in sync with the new funnel. Existing customers are automatically on the Pro tier. Nothing changes for you.
v1.1.2April 10, 2026

Friendlier failures

  • NewGeneration Logs in Settings. Every failed generation now shows up in Settings → Generation Logs with a one-click Copy button that produces a paste-ready support report (error, prompt, model, settings, job ID, app version). Drop it in an email and support can diagnose without 10 rounds of back-and-forth.
  • NewFriendlier failure messages. Errors from PiAPI, fal.ai, and Google are now translated into plain-English explanations (face filter, content policy, timeout, rate limit) with a suggested next step and an explicit note that your credits were refunded. The raw provider error is still included on a Debug line for support.
  • ImprovedFailure toasts no longer falsely claim 'credits refunded' for BYOK users (where no credits were charged to begin with). Toasts also include an 'Open Logs' action that jumps straight to the failure report.
  • ImprovedDelete Project confirmation now spells out exactly what will be deleted: characters, generations, storyboards, timelines, and references. No more surprise wipes.
  • FixedVeo first/last frame generations now send the correct frame fields when both frames are provided. The old request used the wrong field names and could fail or silently drop one of the frames.
v1.1.1April 10, 2026

Seedance Priority tier

  • NewSeedance 2.0 Priority tier. When Economy is backed up, switch to Priority and skip the queue entirely. Routes through fal.ai with no wait times and no watermark. Costs more per second but ships instantly.
  • NewPrompt character count indicator with per-model limits. See how close you are to the max as you type so you don't hit a wall mid-prompt.
  • ImprovedLogin screen now reminds you to sign in with the email you bought your license with. Cuts the most common support question in half.
  • ImprovedSeedance Economy jobs no longer get prematurely refunded when PiAPI queues back up. Reaper threshold extended from 25 minutes to 4 hours so legitimate long-running jobs complete instead of failing.
  • ImprovedSeedance Economy output now has the watermark automatically removed as a second phase. If removal fails for any reason, you still get your original video so you never lose credits.
  • FixedCross-project generation routing is fixed. Generations now correctly stay with the project they were started in.
  • FixedPiAPI race condition where new tasks briefly returned as 'not found' before indexing is now handled gracefully instead of hard-failing the job.
v1.1.0April 7, 2026

Seedance 2.0

  • NewSeedance 2.0 by ByteDance. Native 2K video, audio co-generation with 8+ language lip-sync, multi-modal references. Two tiers: Fast ($0.12/s) and Standard ($0.15/s).
  • NewText-to-video, first/last frame, and ingredient reference modes for Seedance. Same storyboard workflow you already know.
  • ImprovedModel selector now uses collapsible provider groups. Kling, Veo, and Seedance each collapse to save space.
  • ImprovedImage references are now automatically linked to generated videos regardless of how you add them.
  • ImprovedSettings panel scrolls properly on small screens.
v1.0.2April 4, 2026

Polish and fixes

  • FixedVarious bug fixes and stability improvements.
v1.0.1March 31, 2026

Launch patch

  • FixedInitial post-launch fixes and refinements.
v1.0.0March 2026

Launch

  • NewSlates launches with Kling 3.0, Veo 3.1, Nano Banana 2, FLUX.2 Max, Seedream 5 Lite, lip sync, motion transfer, and the storyboard workflow.

One-time purchase · 30-day money-back guarantee