What's new

Synthia tasks: saved prompts that run on a schedule

• New Synthia has a Tasks tab. Save a prompt, choose every day or every week and a time, and Synthia runs it for you. The answer waits in the Results inbox, and the Synthia menu item shows how many you haven’t read.
• New A task only writes text. It never publishes, schedules a post or messages anyone. You can have up to 5 active tasks, and pause, resume or delete them at any time.
• Improved Each answer uses credits like any other answer from Synthia. If there aren’t enough credits, that run is skipped, says why, and nothing is charged. A run that gets no answer is not charged either.

Deep analysis: a research report on your own data

• New Synthia has a Deep analysis tab. Ask a question about your posts, comments or schedule and she plans the research, reads the parts of your workspace that answer it, and writes a report with a summary, findings and next actions.
• New Every finding cites the posts behind it. Click a citation to open that post, or copy the whole report. A report can only cite posts from your own workspace that the analysis actually read.
• New A deep analysis costs two AI credits, the same as two drafts, and is checked against your credits before each step.

Synthia can think harder before she answers

• New Turn on “Think harder” in the Synthia chat or the chat bubble and she takes longer to reason through your question before replying. Her reply comes first; how she reasoned sits in a panel under it that you can open. A message costs the same with it on or off.

Customize Synthia: a persona and your own instructions

• New In Synthia, Customize Synthia lets you pick her role (Strategist, Copywriter, Community manager or Analyst) and write up to 1,500 characters of standing instructions, such as “always answer in bullet points”. They apply to every chat in your workspace.
• New The chosen persona shows as a badge in the chat header. Settings are saved per workspace; owners and editors can change them, viewers can see them.

Synthia remembers what your team tells it

• New A Memory card under Ask Synthia lists what Synthia remembers for your workspace, like a launch date or the tone you prefer. Synthia uses it in every new chat and analysis, so you don’t have to repeat yourself.
• New When Synthia thinks something is worth keeping, it asks “Remember this?” under its reply. Nothing is saved until you press Remember, and you can forget any item at any time.
• Improved Memory belongs to your workspace only. Everyone on the team sees the same list; viewers can read it, and owners and editors can change it. A workspace keeps up to 50 items of up to 300 characters each.

Your conversations with Synthia are saved

• New Every conversation in Ask Synthia is now kept. Open Chats to see your earlier conversations, pick one up where you left off, rename it or delete it. New chat starts a fresh one.
• New Your chats are yours alone: teammates in the same workspace cannot see them. Deleting a chat deletes all of its messages. Photos and files you shared are not stored, only their names.

Synthia reads your PDFs and data files

• New Attach a PDF to Ask Synthia or the Synthia bubble and she reads the whole document, text, tables and charts included, then answers about what it actually says. Up to 2 PDFs and 6 MB per message, up to 100 pages; the chip shows each PDF's page count and size.
• New CSV, JSON, Markdown and text files are read in full up to 20,000 characters, and a spreadsheet's column names are pointed out to Synthia so she reads the rows correctly.
• Fixed The studio checks every attachment itself, by its real contents rather than the label it arrives with, and a file it cannot open is still described honestly. A chat with a document costs the same as any other Synthia message.

Make and edit images right in the Synthia chat

• New Turn on Image beside Synthia’s message box, or start a message with /image, and Synthia makes the picture you describe in square, 4:5, 9:16 or 16:9. It works in Ask Synthia and in the floating chat bubble.
• New Attach a photo with the request and Synthia edits that photo instead of starting from scratch.
• New Each image shows in the conversation with Use in post, which puts it in the post composer, and Try again. Images use your credits the same way the Images tab does.

Paid features are checked by the studio itself

• Fixed Design studio, Video & AI Studio, Link in Bio and your saved influencer list are now checked against your plan by the studio itself, the same way API keys and recycling already were, including scheduled and automatic video work. If something isn’t included in your plan, the studio says so and opens Billing.
• Fixed Everything you already made stays yours on any plan: you can still view your videos, frames and generations, delete saved influencers and auto-translate rules, and your public Link in Bio page stays online.

The studio opens without loading Eagle AI Search’s artwork

• Fixed Eagle AI Search’s logo used to download every time the studio opened, whichever page you were on. It now loads the first time you open Eagle AI Search.
• Fixed Opening a link straight to Eagle AI Search could show it without its logo or channel list until you clicked away and back. It now opens complete.

Pages download only your language’s translations

• Fixed Every page used to download the translations for all 14 languages to show you one. Now it downloads only yours (in English, none at all), so pages start faster, especially on a phone. Switching language downloads just the one you pick.

The home page loads only your language’s dashboard preview

• Fixed The “A dashboard that speaks your language” preview used to download the screenshots for all 14 languages to show you one. It now downloads only the one in your language, and another only if you tap it.

Faster return visits

• Fixed Your browser now keeps the studio’s largest scripts, the translations and the logo set, between visits instead of checking them again every time a page opens. When we release an update, those files get a new address, so you still get the new version on your next visit. Pages themselves are always checked for updates.

The home page loads its videos as you scroll to them

• Fixed The home page used to start downloading and playing every card video the moment it opened, including the ones far down the page. Each card now shows its still image first, loads its clip as it comes into view, and pauses it once you scroll past, so the page opens much faster, especially on a phone.
• Fixed With reduced motion turned on, the cards show their still images and no clips are downloaded.

Plan features are checked by the studio, not only on screen

• Fixed Creating an API key and recycling evergreen posts are now checked against your plan by the studio itself, the same way influencers and approvals already were. API keys come with the Business plan; recycling comes with Starter and up. Keys you already have keep working, and can still be listed and revoked on any plan.
• Fixed When something isn’t included in your plan, the studio now says so in plain words and opens Billing, which shows the plan that includes it.

Posts are only ever aimed at networks the studio can publish to

• Fixed Importing, bulk-pasting, duplicating or recycling posts, or editing one, could still save Google Business Profile as a destination even though nothing can be published there yet. The studio now keeps only networks it can actually post to, everywhere a post is saved. A post aimed only at Google Business is refused with the same plain reason the Channels page gives.
• Fixed A scheduled post left with no network it can be published to is now kept as a draft, and if one ever reaches the scheduler it is marked as not sent with the reason, never as published.

Create post opens cleanly without an image

• Fixed Opening Create post before attaching an image hit an error in the background, so the post preview did not appear until you typed or attached something. It now opens with the preview, the checks and the image tips in place.

Google Business stays off the composer even when the app is slow to load

• Fixed If the studio was slow to learn which networks can be posted to, the post composer could still briefly offer Google Business Profile and include it when you published. It now keeps it greyed out from the start.

Google Business is no longer offered as a place to post

• Fixed The post composer let you pick Google Business Profile as a destination, and the post then could not go out. Its button is now greyed out and labelled Won’t publish yet, and it is never included when you publish.
• Fixed The AI support assistant listed Google Business among the networks it can publish to. It now knows exactly which networks can be published to, and that Google Business Profile and WhatsApp can’t yet.

Google Business: no hint that signing in will make it post

• Fixed On Channels, the home page and the how-to guides, Google Business Profile was described as “sign-in works” or “can sign in”, which read as if signing in would let the studio post there. It now says plainly that it won’t publish yet, even once its sign-in is set up — in every language.

Links to a how-to open it

• Fixed A link to a specific how-to (for example /how-to.html#q-connect-x) only showed the topic list. It now opens that walkthrough and plays it. A link to a how-to that doesn’t exist simply shows the topic list.

How-to animations show the right address

• Fixed The browser bar drawn in the how-to animations showed the address of a different AI SNS product. It now shows studio.ai-sns.io/app, where the studio actually lives.

“We sent a link” reads naturally in every language

• Fixed In the confirm-your-email banner, “we sent a link to” your address was stitched together in English order and always ended with an English full stop. It is now one sentence per language, so Hindi reads “हमने … पर लिंक भेजा है।” and Japanese and Chinese end with “。”.
• Fixed Translated text could garble an email address containing a “$” sign. Addresses now appear exactly as entered, in every language.

Your plan’s channel limit applies to one-click connect too

• Fixed Connecting an account with one-click sign-in did not check how many channels your plan includes, so it could go past the limit. It now checks, the same way pasting a token always did, and says plainly: “You’ve reached your plan’s channel account limit.”
• Improved Reconnecting an account you already have — for example after its sign-in expires — never counts as a new channel.
• Fixed After signing in with a network, you now come back to Channels and see whether it worked, instead of landing on the home page with no message.

How-to guides tell you what you can connect, not what you might

• Fixed The how-to guides for Instagram, X and TikTok said "click Connect and approve" as if every network could always be connected. They now explain what each card on Channels shows — Connect, Not available or Won’t publish yet — and what to do in each case.
• Fixed "Which platforms can I connect?" in the how-to guides now lists the platforms you can connect when they are set up, says Google Business can sign in but won’t publish yet and WhatsApp is not a place to publish, and points to Channels for what works right now.

The confirm-your-email banner speaks your language

• Fixed The banner asking you to confirm your email address was in English in every language. It is now translated in all 14 — including the line that says plainly when this site cannot send email yet.

Pop-up windows sit above the assistant button

• Fixed On a phone, the round assistant button floated on top of pop-up windows such as Connect, covering part of their text — most of all in Arabic. Pop-ups now cover it like the rest of the page, and it is back as soon as you close them.

The Connect window speaks your language

• Fixed The title of the Connect window on Channels began with the English word "Connect" in every language. It is now translated in all 14, for example "ربط WhatsApp" in Arabic.
• Fixed The one-click Connect X button was in English in every language. It is now translated too.

Home page FAQ: which platforms you can connect, plainly

• Improved The FAQ Which platforms can I connect? now answers for every platform the page names: which ones you can connect right now, which can sign in but won’t publish yet, and which are not available yet. It uses the same words as the Channels screen and updates itself as platforms are switched on.

The home page no longer counts platforms it cannot connect

• Fixed The home page said 8 platforms and 8 networks, and listed Google Business, which the studio cannot post to yet. It now says Platforms you can connect when ready, names the seven the studio can post to, and points to the FAQ for which ones are switched on today.

Channels: clearer labels when a platform is off

• Improved On Channels, a platform that can't be connected yet now says Not available, and one that can sign in but can't post yet says Won’t publish yet. The reason is still shown underneath, in plain words.
• Fixed The Connect window for WhatsApp now names it properly.

Arabic on a phone: the menu and composer stay out of the way

• Fixed In Arabic on a phone, the side menu and the post composer stayed half on screen even when closed, covering most of the page. Both now slide fully off the right side when closed, and open from it when you tap them.
• Fixed A closed menu or composer can no longer be tapped or reached with the keyboard by accident.
• Fixed In Arabic, pop-up notices are centred on the screen again.

The website only promises connections that work

• Fixed The home page, Apps and every model page said you could connect Meta, Google and TikTok ad accounts. Ad accounts can't be connected here yet, and the pages now say so. They name the ad platforms again as each one is switched on.
• Fixed The home page promised one-click connect for eight networks, including Google Business Profile, which the studio cannot post to. It now lists only the networks that can be connected and posted to right now, and says plainly when there are none.
• Fixed Google Business Profile shares its sign-in with YouTube, and WhatsApp had a Facebook-style login. Once those sign-in keys were set, either route could save a channel as "Connected" that would never post. Both now refuse with the same reason the Channels page shows.

Channels say plainly what can be connected

• Fixed The One-click connect buttons on Channels were always clickable, even when a platform could not be connected yet. Each one now says whether it is available, and why not when it isn't.
• Fixed Pasting an access token for a platform the studio cannot post to yet (WhatsApp, Google Business Profile) used to save it as "Connected" and use up a channel slot on your plan. It is now refused, with the reason.
• Fixed Opening Connect for WhatsApp did nothing. It now explains what WhatsApp needs.
• Fixed Per-second prices on the model pages read "24credits/sec". They now read "24 credits/sec".

No blank tiles on the home page, and empty samples say so

• Fixed Twelve cards on the home page still showed two letters instead of footage. They sat next to a card that already had a clip, and the page mistook the neighbour's clip for their own. Every card now shows footage from the studio's own reel until something is uploaded for it.
• Improved On the model catalogue, a model with no sample yet now says No sample yet, in every language, instead of showing a letter tile that looked like a missing image. Nothing stands in for a model's output: the tile fills when the studio has a real sample from that model.

One set of prices on every page

• Fixed Pricing, Enterprise and Team plan showed older plan prices and credit allowances than the home page and checkout. They now show the same figures: Free ¥0 with 120 credits, Starter ¥11,920 with 1,500, Pro ¥31,920 with 4,500, and Business ¥79,200 with 12,000 credits a month.

No placeholder requests on the Enterprise and Company pages

• Fixed Enterprise and Company showed seven bracketed notes asking for details we have not written yet — support terms, a security statement, the team. Until those details are written, the pages leave them out instead of showing the request. Nothing was added in their place.

The dashboard finishes loading

• Fixed Opening the dashboard hit an error in the Channels grid on every load. Everything after it — which channels can connect, the inbox, the growth chart, the default schedule time — never started, so the page opened half-working. The WhatsApp card now draws properly, and each dashboard panel now loads on its own, so one fault cannot stop the rest.
• Fixed When a sign-in expired, signing back in dropped you on the default page. It now returns you to the screen you were on.

Settings in one place, and studio names on the studio's shelves

• New Settings now has a section nav — Account, Workspace, Billing, Vendors, Channels, Notifications, Team and Brand. Billing and Channels still live on their own pages; Settings shows where you stand and takes you there.
• New Vendors and Notifications say plainly what is switched on — writing, generation, audio, security email — and what is not available yet. No key is ever shown.
• Improved Three catalogue models now carry studio names: Character Still 2.0, Cinema stills and Director camera moves. Each model's own page still says which API provider runs it.
• Fixed Translate said 15 languages. It now offers exactly the 14 languages the studio itself speaks, and says 14.
• Fixed Settings had switches that only your browser remembered and that changed nothing, and an "API access — coming soon" message beside API keys that already work. Both are gone.

Only our own footage, and a picture before it plays

• Improved Home page. Seventeen cards still showed files that came with the first storefront build and that nothing records as ours. They now show the studio's own reel, and all 62 clips carry a still frame, so no card sits black while its video loads.
• Improved Effects. All fifteen effects now have a clip — five had none — each badged Sample and each with a still. The wording no longer says the clip comes from that model's reel: it is a mood board from our own footage, not the output of the effect or its model.
• Fixed Three clips named as our own turned out to be the same first-build files, re-encoded under a new name. They are no longer used anywhere, and a test now fails if a page serves one.
• Improved Model passes in Billing show the same footage as the storefront, with stills.
• Fixed Images, documents and generated audio or video saved to a workspace were answered as saved, but they live on the server's disk, which the backup does not copy. Each save now says it is not durable and why, and Admin → Launch readiness counts the files at risk.

Prove the backup actually restores

• New Verify restore. Admin → Launch readiness now has a button that downloads the newest snapshot, opens it read-only, runs SQLite's own integrity check and counts the rows inside it against your live data. It never touches the live database.
• New The checklist asked you to confirm a restore had been exercised, and gave you no way to do it. It now tracks when one last ran and whether it passed — and remembers that across a rebuild, which is exactly when you would want to know.
• Improved A snapshot that uploads cleanly every few hours and would restore an empty product looks healthy from the outside. Row counts are compared against live, so that case is caught.

Every channel says what it is waiting on

• New Publishing channels on the Launch readiness screen lists all nine platforms and, for each one that can publish, the exact sign-in variables it still needs. Nine platforms share five credential pairs, so one pair can light up three at once.
• Improved A platform with no publish adapter says so instead of showing keys that would let a client connect an account that can never post.

Decide credit enforcement from the console

• New Refuse generation when a workspace is out of credits is now a switch in Admin → Billing & Plans. It used to be a server variable, which put a commercial decision behind a rebuild. Off, usage is still metered and still charged — the balance just goes negative.
• Improved A server variable still wins either way, so a staging copy pinned off cannot start refusing customers because someone clicked a toggle.
• Fixed The AI Studio provider key was a plain text field whose own help text said it was never sent to a browser. It was. It is now stored server-side and the console shows only whether a key is set.
• Improved Launch readiness rows now link straight to the panel that fixes them instead of only naming a variable.

One screen that says what is switched off

• New Launch readiness. Admin console → Launch readiness checks seventeen things about this running deployment — payments, the Stripe webhook, transactional email, off-site backup, credit enforcement, session and token encryption keys — and lists what is blocking launch, with the exact setting that fixes each one.
• Improved The checklist already existed but was reachable only with a server secret, so the person who had to act on it could not see it. It now sits behind the ordinary admin sign-in.
• Improved Before credit enforcement is switched on, the page shows how many workspaces are already overdrawn and by how much — switching it on refuses further generation to every one of them.

Send mail from your own domain

• New Sending domains, in the console. A mail provider key on its own can only send from the provider's own address. Admin console → Email & SMTP → Sending domains now lets you add your domain, shows the exact DNS records to publish at your registrar, and checks them when you press Verify.
• Improved The records shown are exactly what your provider returned — nothing is guessed — and a domain that is not verified says so rather than implying mail will arrive.

Set up email from the admin console

• New Email settings now include the key and the sender address. Admin console → Email & SMTP had a provider list and a rate limit but nowhere to put a key, so email could not actually be switched on from the console. Both fields are there now, and Send a test email proves it works before a client ever needs it.
• Improved The key is stored on the server and never sent back to the browser — the page shows only whether one is set. Saving the page again keeps the key you already had rather than overwriting it.

The sign-in page stays put

• Fixed Signing in could get stuck in a loop. If your previous session had ended, the sign-in page sent you to the studio anyway — because it only checked that a sign-in existed, not that it still worked — and the studio sent you back. The page appeared for a moment and then kept reloading. It now checks with the server first, and if the old session is finished it clears it and lets you sign in, which is what you came for.
• Improved If your connection drops while it checks, the form simply stays on screen rather than sending you somewhere that cannot load.

Sign up now arrives as sign up

• Fixed The create-account page arrived as the sign-in page. Every sign-in link shares one page, and which form you get was decided after the page loaded. If anything stopped that step — an extension, a blocked file, an error — you were left on the wrong form, and the tabs that switch between them need the same step to work. Each link now arrives as the right form on its own.

An ended session sends you back to sign in

• Fixed You could get stuck signed out without being able to reach the sign-in page. If your session had ended, the app still opened — it only checked that a sign-in existed, not that it was still valid — so every action quietly failed, and the front page offered "Dashboard" instead of "Log in". The app now notices the moment the server stops accepting a session, signs you out properly and takes you to the sign-in page, which tells you what happened.
• Improved A wrong password still just says the password is wrong — it does not end a session you already have.

The logo is visible again

• Fixed The mark was white on white in light mode. There are two versions of it — one for light backgrounds, one for dark — but only the front page ever switched between them. Every other page, including sign-in, used the white one, so in light mode it simply disappeared. Every page now picks the right one.
• Improved It follows your device's light or dark setting, your own choice if you have made one, and changes immediately if you switch — no reload.

Two more ways to sign in

• New Continue with Google. Sign in with a Google account — it asks only to identify you, never for permission to post anywhere. If you already have an account with that address, it signs you into it rather than starting a second one.
• New Sign in with a code sent to your phone. Enter your number, receive a code, and you are in.
• Improved Both appear only where they are set up. On a deployment without them you see the ordinary sign-in and no button that cannot work.

The front page plays your own footage

• Improved The cards that showed a two-letter badge now play video. They fall back to the studio's own committed clips, so the page is full instead of showing sixty placeholders. Anything you upload to a card still wins — the fallback only fills what is genuinely empty.
• Fixed Posters are carried with the clips, so a card shows a still frame rather than a black box while it loads.
• Improved The storefront hero, unlimited rail and offer grid are filled from the Founder's own studio-media clips via bulk ingest, so those slots are no longer empty on a fresh Space.

A front page you can swipe, and a dashboard that greets you

• Improved The three feature cards now scroll properly on a phone. They used to wrap awkwardly; they now swipe with snap points on a small screen and sit as a clean row of three on a large one.
• Improved The list of what you can make now leads with a larger tile, so your eye has somewhere to land instead of meeting fourteen identical boxes.
• New Your dashboard says hello. It opens with the time of day and your name — in all fourteen languages — above the things you are most likely to do next.
• New Build on the studio. The front page now points at the three ways to use it from your own code — the REST API, the MCP server for Claude, Cursor or a terminal, and the browser extension. All three already existed; none of them was on the front page.

The newest panels look like the rest of the studio

• Improved More depth on the front page, the dashboard and the model list. The panels added over the last day were flat while the rest of the site has lit edges and shadow. They now sit on the same depth as everything else, so the set reads as one product rather than a stack of boxes.
• Improved Keyboard users get the same highlight the mouse does, and all of the movement is dropped for anyone who has asked their device to reduce motion — the depth carries the meaning, so nothing is lost.

Clearer answer when a payment cannot be taken

• Improved Checkout says what happened in plain words. Where card or PayPal payment is not enabled on a deployment, it already refused rather than pretending — it now also says so readably, and states that nothing has been charged.
• New The launch checklist reports whether card and PayPal payment can be taken at all, as separate lines from the receipt webhook — the two fail independently, and a healthy webhook never meant a customer could actually buy anything.

Narrow the model list, and find your place in the reference

• New Filter the models by what they make. All, Video, Image or Audio — each with the real number of models behind it, so the count on a button is always what you get when you press it.
• Improved The API reference shows where you are. It is a long page, and the list on the left was static — you had to count headings to place yourself. The section you are reading is now marked as you scroll.

Both front doors say what you can do

• Improved The front page now shows everything you can make, near the top. You used to scroll past several feature panels before finding out what the studio actually does. There is now one compact grid of every part of it — each one a real page, one click away — before the pitch begins.
• Improved Your dashboard opens on what to do next, not on four numbers. Create a post, open AI Studio, connect a channel, see the calendar — these come first now, with the statistics underneath. Numbers tell you where you have been; they do not tell you what to do.

The gallery loads sixty at a time, and tells you how many there are

• Improved Your library no longer loads everything at once. It shows sixty, says how many you have in total, and offers Load more when there are more — which adds to what is on screen rather than replacing it, so you keep your place.
• Fixed The count was wrong while searching. Searching for a word with eleven matches could still say "60 of 400", because the total counted everything you had ever made rather than the results you were looking at. It now counts what is actually on screen.

When a post cannot go out, it tells you why

• Fixed A post to a disconnected account failed without explaining itself. If the account was never connected, or the connection had expired, the attempt went ahead anyway and you were left with whatever the network said about a missing sign-in. It now stops before it starts and says which account it is and that you need to reconnect it under Channels.
• Fixed A post can no longer be marked Published on a shrug. If a connector throws an error, or answers with nothing at all, that is now a failure with a reason on the post — not a success. Only a run where every network actually accepted the post marks it Published.
• Improved Channels that cannot post yet say so by name when you schedule to them, instead of failing later as if the sign-in were the problem.

One header on the front page, not two

• Fixed The home page had two stacked bars at the top. The site map sat on its own row underneath the main bar, with a dividing line that read as a second header and took up space on every screen. There is now a single bar — logo, the main sections, theme and language, and sign in.
• Improved Everything that was on the second row is still one click away under More, which opens over the page instead of pushing it down. The section links on the rest of the site are unchanged, so the whole site now reads the same way.

An upload is not "saved" until it is actually safe

• Fixed Uploaded video could be lost when the site was redeployed. Uploads are kept in the database, and only the database is copied off the server — so anything uploaded since the last copy was lost when the site next rebuilt. Three videos were lost this way. An upload now waits for that copy to complete before it reports success.
• Fixed The copy could be postponed indefinitely. Each save pushed the backup back by twenty seconds, so on a busy afternoon it never ran at all. A pending backup now happens within a minute of the first change that made it pending, however much else is going on.
• Improved You are told the truth. If the backup does not complete, the upload says so and names the reason, instead of reporting "Saved" for something that might not survive the hour. A bulk run backs up once at the end rather than after every file.
• Improved When the backup last succeeded is now remembered across restarts — it used to reset to zero on every boot, which hid exactly this problem — and the launch checklist treats a backup that has never completed as a blocker.

Schedule a generation for later

• New Ask for a generation at a time of your choosing. Set when it should run and close the tab — the studio starts it for you when the moment comes, and it is waiting for you when you come back. Nothing needs to stay open.
• Improved Nothing is charged until it actually runs. Scheduling costs nothing; the credits come off when the generation starts, at exactly the price the same generation costs if you run it yourself. A schedule you cancel before it fires costs nothing at all.
• Improved A mistake is caught while you are still looking at the form — an unknown model, a time in the past, or not enough credits — rather than silently failing overnight. If a run does fail, the schedule keeps the reason instead of disappearing.

Uploaded video is no longer drawn over

• Fixed Three home-page cards drew a graphic on top of the video you uploaded. Those cards paint their own artwork while they are empty, and only part of it was being cleared when a video took its place — so the leftover sat over your footage. The whole placeholder is now removed, and a check runs on every build so no card can draw over its own media again.

A generation finishes even if you close the tab

• Fixed A render only settled while you were watching it. If you closed the tab or lost your connection part-way through, the job stayed open — neither delivered nor refunded — until you came back and opened it again. The studio now settles finished jobs on its own, through exactly the same path your browser used, so the credit lands the same way whether you are watching or not.
• Improved While the generation platform is not connected, the sweep does nothing and says so, rather than retrying against a service that is not there.

Channels tell you what they can actually do

• Fixed Two channels could be connected and would never post. Google Business Profile and WhatsApp could complete sign-in and show as a connected account, but neither has publishing wired — anything scheduled to them would simply never go out. They now say so plainly instead of offering a Connect button that leads nowhere.
• Improved WhatsApp appears in the channel list at all. It was missing from the grid entirely, so its state could not be seen one way or the other.
• Improved Each channel card now says why it is unavailable — sign-in keys not set on this deployment, or posting not wired yet — rather than showing the same marketing line whatever the state.
• New The launch checklist reports which platforms are connectable, and warns when sign-in keys are set for a channel that cannot publish — the case that would otherwise quietly mislead someone into scheduling to it.

Public site IA aligned with Create → Cinema → Company

• Improved Navigation and footer now carry Create, Apps, Models, Effects, Community, Studios, Soul, Cinema, Platform, Pricing, Enterprise, Team and Company — the same public map the studio is building toward.
• New /cinema — a thin, honest section: Cinema Studio is marked not offered yet; Soul Cinema links to the live model. No competitor media and no invented community content.
• Improved Resources joins the public top nav next to Platform, matching the section map.
• Fixed Soul section copy describes Soul models as they run in this studio — it no longer reads as if the public site were Higgsfield.

Transactional email hardened; billing mail stubs refuse to pretend

• Improved The five real service emails (confirm, reset, password-changed, invite, invite-accepted) now expose pure builders for preview/tests, and still refuse to claim a send when Resend is not configured.
• New Sixteen billing email stubs (receipt, invoice, failed payment, trials, seats, …) always refuse to send until Stripe/PayPal are connected — and stay unimplemented even then, so the product never invents a receipt email.

Every card on the home page can now hold your own video

• New 65 more places to put a video. The home page draws 82 cards and only 17 of them ever showed a file — the rest drew a two-letter badge. Not because they were unfinished, but because there was no way to put anything on them. Now there is: Admin console → Website → Home cards lists every card, and dropping a clip or a still on one replaces the badge.
• Improved Bulk upload reaches them too. Choosing "Home page" used to offer 12 places. It now offers every empty card, so a folder of clips fills the page in one drop.
• Improved Nothing is overwritten. Your upload is layered over the page rather than written into it, so removing one puts the card back to exactly what it shipped with. Each card is keyed to the model it links to, so the page can be reordered without anything landing in the wrong place.
• Improved A card that has always held a still will take a video and become one, at 1080p, watermarked and checked on the way in like every other upload. Where the model behind a card makes images rather than video, the console says so, so the choice is yours to make deliberately.
• Fixed Cards that draw their own animation are left alone. The Video Translate card animates four languages being dubbed; it has no picture to swap, so a clip dropped on it would have sat on top of the animation instead of replacing it. Cards like that are no longer offered as upload slots, and the animation stays exactly as it was.
• Fixed Generate no longer pretends a run can start when there is no platform key. With Higgsfield unset, the composer disables Generate, refuses ⌘+Enter locally, and the API answers 503 with setup:true — credits are never taken. The same honesty applies to audio when ElevenLabs is unset. The catalogue, quote and effects presets remain usable for wiring.
• New A regression test proves the no-key path: catalog still lists models for wiring, quote still answers, generate/audio return 503, and the credit balance does not move.

Honesty pass: payments, models, email, and public prices that match the composer

• Fixed The pricing and enterprise pages no longer claim Stripe and PayPal are live when they are not. "How can I pay?" answered "By card through Stripe, or with PayPal" unconditionally. It now reads from the same check the checkout route uses, and says plainly when payments are not enabled yet.
• Fixed The support assistant could tell you to pay by card when it was not possible. What it claims about paying now follows what is actually connected.
• Fixed "We sent a link to your inbox" — when nothing had been sent. The confirm-your-email banner now says plainly when email is not configured on this deployment.
• Fixed The AI Studio storefront named two cards after models this studio does not run. Image cards are renamed to what they actually are; tiers with no honest on-sale equivalent are removed rather than relabelled, along with empty Google and OpenAI filter tabs.
• Fixed An unconfigured Stripe webhook now has a test proving it fails closed. Unsigned and forged-signature requests are refused while STRIPE_WEBHOOK_SECRET is unset.
• Fixed The public Effects gallery and model pages were quoting the flat rate, not the model's own rate. Both pages now price through the same model-aware call the composer and the generate route use, so a card never promises a number the studio would not charge. Corrected numbers only go down.
• Fixed Arabic mirroring only ever covered the main app's own stylesheet. The admin console, the landing page and several other pages still had spacing pinned to the physical left or right — a badge pushed the wrong way in a toolbar, a list indented from the wrong edge, a label aligned to the wrong side. All of them now use the same direction-aware CSS as the main app, so nothing on the site fights the reading direction for an Arabic visitor.
• Fixed Three admin settings pages were unreachable from their own sidebar link. Users & Roles, Workspaces & Teams, and Email & SMTP shared an internal id with a different panel, so clicking any of them always opened that other panel instead — silently, with no error. Each now opens correctly.
• New scripts/ingest-media-db.js — an operator CLI that re-ingests a folder of owner-supplied video/images straight into the same database-backed storage the admin console's Bulk upload uses (site_uploads / site_slots / site_samples), so footage lost when the old storage wrote to the container filesystem can be restored without dragging files into a browser one at a time.
• Security The script takes no default folder — it refuses to run without an explicit --from=<dir> from the operator, and refuses outright to read from the competitor crawl at ~/Downloads/videos or anywhere under it.
• Improved Storage reporting. --storage and --list print the same total-uploaded-MB figure the admin panel shows, against the 250 MB warn threshold — this media rides the app.db backup archive, so its size is backup time, not disk space.

Bulk upload asks which page, and everything lands in 1080p

• Fixed Bulk upload could not fill the home page. Dropping videos with the home page in mind changed nothing there, and it never could: the only places a bulk run knew about were the AI Studio storefront and the model cards. The home page's video is set separately, and it was not on the list.
• New Choose the page you are filling. Admin console → Website → Bulk upload now starts with a choice — Home page, AI Studio storefront, Model cards, Effect cards, or everywhere — and shows how many empty places each one has before you drop anything. Files go only to the page you picked.
• Improved Uploads are now encoded at 1080p, up from 720p for page media and 480p for the model cards, so a clip still looks sharp on a large screen. A video still only ever lands on a video position and an image on an image one.
• Improved The panel shows how much uploaded media you are storing, and says so when it grows large enough to be worth clearing out.

AI Studio: every model priced as itself, and a Recreate button on every result

• New Recreate — every finished image and video now carries Recreate, which loads its prompt, its model and its exact settings back into the composer so you can change one word and run it again. Beside it: Again (run it exactly as it was), Copy prompt, and Use as input, which sends the result into the next generation as its start frame or source clip.
• Fixed Every model was priced the same. There was one price for all image models and one for all video models, so Seedance 2.0 Mini cost exactly what the flagship cost. Each model now declares its own rate. No price went up — the cheaper tiers came down: Seedance Mini 60 and Fast 80 credits for 5s (was 120), LTX Fast 70, Wan 2.6 80, Kling Standard 100, Z-Image Turbo 6 credits an image (was 12), Qwen Image 3 10.
• Fixed A batch was quoted as one image. Asking Soul for four images showed 12 credits and charged 48. The quote now shows what you will actually be charged, before you press Generate.
• New Your generations are kept. The model platform's links are temporary, so older results quietly stopped working. Every finished generation is now copied into your own media library the moment it lands, and a result that pre-dates this says so instead of showing you a dead link.
• New Empty slots are caught before you spend. Models that need a source clip or a frame — Seedance Edit and Extend, Kling O3 and O1, Motion Control, Kling 2.5 — now tell you which one is missing instead of failing at the model.
• New Post-format buttons — Reels / TikTok, Feed, Portrait, YouTube, Cinematic. Pick where it is going and the aspect ratio follows, landing on the nearest one the model actually offers.
• Improved Search finds everything you have ever made, not just the most recent page of it, and the filters now include audio. Each result shows the settings it was made with and what it cost.
• Improved Every model card wears its lab's real logo — Higgsfield, Bytedance, Kling, Black Forest Labs, xAI, Ideogram, Recraft, Alibaba, Qwen, Wan, MiniMax, Lightricks, PixVerse and ElevenLabs.
• Improved A finished generation no longer interrupts the clip you are watching, a tab left in the background stops polling every three seconds, and a long video render can notify you when it is done. The model picker is now a proper dialog: Escape closes it and the keyboard stays inside it.
• New Drop a hundred videos and they place themselves. Admin console → Website → Bulk upload. Every file fills the next empty spot, re-encoded, watermarked and checked on the way in.
• Improved Everything about the website is now in one place. Admin console → Website holds the visual editor, the page text, the storefront media slots and a sample per model — each with its own link, so you can send someone straight to the tab you mean. The separate editor page is gone.
• New Drop a whole folder into the community gallery. Several videos at once, each given a title from what is in it, or from its filename.
• New Post your own work to the community gallery. Admin console → Community gallery now takes a video or image straight from your machine and puts it live, credited to AI SNS Studio. Client posts still go through review; yours do not, because you are the reviewer.
• Fixed Arabic now mirrors properly. The site read right-to-left in Arabic but the layout did not follow — sidebars, badges and icons stayed pinned to the left. The whole stylesheet is direction-neutral now, so Arabic lays out as a right-to-left reader expects. Nothing changes in the other twelve languages.
• New Visual website editor. Admin console → Website. Pick any of the 79 pages, click anything on it, and change its text, its size, its colour, or the video/image it shows — then Save and it is live. Uploaded video is re-encoded and checked before it can reach a page, and everything you upload is stored so it survives a rebuild.
• New Site editor. The website's copy and media are editable from the admin console under Website — no redeploy. Text left empty keeps showing its placeholder on the live page rather than inventing a fact, and uploaded video is trimmed, re-encoded and checked before it can reach a card.
• New The AI Studio storefront now runs on our own footage. The hero backdrop, all five unlimited cards and the all-access banner play video from AI SNS, each one re-encoded for the size it is actually shown at and given a still frame so nothing is a grey box while it loads.
• Fixed Password reset no longer says "check your inbox" when it could not send. If email is not configured on the site, the reset page now tells you so and points you at support, instead of sending you to wait for a message that was never going to arrive.
• Fixed Two internal maintenance endpoints were reachable without a key when no key had been configured. They now refuse outright in production. Scheduled posting is unaffected — it runs on its own timer.
• Fixed Model-card samples had nowhere to land. The tool that generates a sample for each model wrote its results to a file nothing ever read, so running it would have changed nothing on screen. Cards now show the sample the studio generated with that model, labelled Our output — and the model list says how many cards still have none, instead of leaving you to scroll and find out.

Work from Claude, Cursor or the terminal

• New API keys — under Settings → API keys you can create a key for your workspace, name it, see when it was last used and revoke it. The key is shown once, when you create it; we keep only a scrambled fingerprint, so nobody — including us — can read it back later. A key can generate and read like an editor, but it cannot change your billing, your team or your keys.
• New MCP server — connect Claude, Cursor or any MCP client to the studio and ask it to list the models, price a generation, make an image or a video, check on it and read your balance. Everything goes through the same checks and the same prices as the studio: a generation costs exactly what it costs in the browser, a pass that covers it makes it free, and a failed one is refunded. The how-to lives on the Platform page.
• New The studio command line — studio models, studio effects, studio balance, studio generate and studio jobs. One file, nothing to install besides Node. It shows the cost and asks before spending anything, and can save the finished file straight to disk.
• Improved The Platform page's MCP/CLI card now leads to real instructions: the endpoint, how to mint a key, example client configs and the install line. The API reference documents the new routes and the protocol.

A new front page

• New The front page has been rebuilt around what you can make. It opens on three feature cards — AI Studio, Video Translate and Visual Effects — then a panel of the studio's surfaces (Edit Studio, Static → Motion Ad, Audio, Effects, Ads Manager and the free Social Analyzer), each one click away.
• New Model showcases — Seedance, Kling 3.0, Higgsfield Soul and the image models each get a gallery of portrait cards with the sample clip or still that model's page shows, and a "View all" button to that model. Models without a sample on disk wear a clean typographic card rather than borrowed footage.
• New Community on the front page — up to four of the latest approved pieces from the Community gallery, with the maker's workspace name and the model. Until someone publishes, the strip says so plainly instead of showing work nobody made.
• New Explore more — a cloud of links to every section page, every model and the effects gallery, and a full-width footer whose columns are read from the site map itself: anything the studio does not offer yet is shown as plain text, never as a link.
• Improved Pricing, the video-translation simulator, the dashboard demo, the localisation preview, the full model catalogue, FAQ and every section other pages link to are all still there, in the same places, with the same links.
• Improved Everything new is translated into all 14 languages, works in light and dark mode and on a 390px phone, plays video only while it is on screen, and stays still when your device asks for reduced motion.

• New Effects, at /effects — every Visual Effects preset in the studio now has a public page. The gallery shows all fifteen as portrait cards; those with a sample clip play it as you scroll, the rest wear a clean typographic card. Filter by type (video or image) and by model family — the chips are read off the presets themselves, so a new preset adds its own.
• New One page per effect — what it does, the model behind it with a link to that model's page, the exact settings the composer is set to, and the real price of one run in credits (video by the second, images per generation), taken from the same figures the studio charges.
• New Use this effect — the button on each page opens AI Studio with the effect already loaded: model, settings and prompt filled in, one click from Generate. If you are signed out, you sign in and land on the same effect.
• Improved Effects is in the header and footer of every public page, in the Create section, and in the sitemap. Sample clips are labelled as samples of the model, never as the effect's own output.
• Improved Everything on the new pages is translated into all 14 languages, works in light and dark mode and on a phone, and stays still when your device asks for reduced motion.
• New Community — a public gallery at /community of images and video made in the studio. Every piece on it is real client work: it appears only because the person who made it pressed Publish to community on their finished generation, and only after we have reviewed it. Nothing is published by default, there is no sample content, and the page says plainly when it is empty rather than padding itself with work nobody made.
• New Publish from AI Studio — a finished image or video in your gallery now has a Publish to community button. Give it a title if you like, choose whether to share your prompt (it stays private unless you tick the box), and it goes into review. Your generation history shows whether each piece is under review, live or hidden, and Unpublish takes it down at once, whatever its state.
• Improved Your work is credited to your workspace's name, never to an email address. The public page and its feed carry the title, the model, the media and that name — no workspace identifiers, no account details, and no prompt you did not choose to share.
• New Every card names the model that made it and links to that model's page, and Make one like this opens AI Studio with that model selected — and the prompt filled in, when its maker shared it. Filter the gallery by image or video, and by any model that has published work.
• Improved Community is in the site header and footer, in the sitemap, translated into all 14 languages, and checked in light and dark mode, with reduced motion, and on a 390px phone.

The front page shows what you can make

• New Every model, on the landing page — all 41 models in the catalogue (8 image, 30 video, 3 audio) as cards that say what each one is good for, with its top resolution, clip length and headline capability at a glance. Filter by image, video or audio. Cards with sample footage play it as you scroll; the rest carry a clean typographic card. Every card opens AI Studio.
• New A media-first hero and a "What you can make" rail — the page now opens on real motion from the studio, then a scrollable row of outcomes: cinematic video up to 30 seconds, native 4K, photoreal stills, voiceover and music, camera moves.
• New Motion tools — camera moves, motion transfer, first-and-last frame, extend, edit by prompt, native audio, multi-shot and reference-to-video, each labelled with the model that does it.
• Improved The whole landing page now shares the app's glass-and-depth design: cards tilt under the pointer, videos play only while they are on screen, and anyone who asks their device for reduced motion gets a still, complete page. Checked in light and dark mode and on a 390px phone.
• Improved Everything new on the page is translated into all 14 languages.

• Improved Model and effect cards are much larger. They were half the height they should be, which made a page of generated work read as a sheet of thumbnails. Every media card is now a full portrait frame — the shape the models actually output — so the work carries the page. Two per row on a phone, so a catalogue of 41 is still quick to get through.
• Fixed Every model pass now names the model it actually covers. Five image passes were sold under names for models this studio does not run, and all five delivered the same thing — so two passes at different prices bought exactly the same access, and the cheaper one came with the higher daily allowance. The same was true of a video pass sold at a quality tier its model does not offer. Passes are now named for what they cover, the duplicates have been taken off sale, and the price ladder rises with what you get. If you already hold one of the retired passes nothing changes: it keeps working to its last day, with the allowance it was sold with.
• Fixed A sample clip on the AI Studio storefront was encoded in a format Chrome and Firefox cannot play, so that card showed nothing at all. It has been re-encoded and now plays everywhere, and loads faster.
• Improved Four Visual Effects were renamed — Walking flame, Skyline tightrope, Infinite selfie and City giant. The effects themselves are unchanged.
• Fixed Monthly plans now deliver every month. A subscription granted its credits when you first subscribed and then never again, even though the card kept being charged. Renewals now add that plan's credits as each payment succeeds, and the renewal date on your billing page is kept up to date. A failed payment marks the account past due rather than cutting you off while the card is retried, and a cancelled plan keeps every credit you had already bought.
• Fixed A generation that ended in an unexpected state on the provider's side used to sit in your gallery as though it were still running, and its credits were never returned. Such a generation is now marked failed and refunded like any other.
• New A new look across the studio. Deeper contrast, glass panels with a lit edge, larger rounded cards and a gentle 3D tilt that follows your cursor. Model cards now play their example clip on their own as they come into view, and stop when they leave, so you can see what a model produces without clicking anything. If your device is set to reduce motion, everything holds still and the site stays fully usable.
• Fixed Every model pass is on sale, and each one says what it covers. The AI Studio model cards offered only one pass per model, so six of the thirteen passes could not be bought from there, and a pass's card never said which model or tier it pays for. Every card now lists every pass for that model (cheapest first), and each pass card in Billing states the models it covers and the highest resolution it pays for.
• Fixed A 720p pass is a 720p pass. Passes for the same model at different resolutions were treated as identical, so a lower-tier pass also covered the model's highest tier. New passes now cover their stated tier and below; the composer says so when a chosen resolution is above your pass. Passes bought before today keep exactly the coverage they had.
• Fixed The composer no longer says "included" once today's allowance is spent. After a pass's daily fair-use allowance ran out, the model card and composer kept showing "Included in your pass" while generations were charged at the normal rate. They now switch to the priced view, show how much of today's allowance is left, refresh after every generation, and stop advertising a pass the moment it lapses.
• Fixed Generations are priced from the settings actually used. A render asking for a duration below a model's minimum was billed for the duration it asked for, while the model still produced its full-length clip — so a four-second video could be charged as one credit. Price now follows the settings the model is really given, and your job history records those same values.
• Fixed Refunds happen exactly once. A failed or cancelled generation refunded once per open tab, so watching the same job in two windows returned the credits twice. The refund is now tied to the moment the job settles.
• Fixed One credit balance per workspace. Credits bought through PayPal, and credits granted by an administrator, could land on a second balance the app never read — leaving the purchase invisible. All credit movements now reach the one balance your workspace spends from. Existing balances are unaffected.
• Fixed A credit pack bought through PayPal now always reaches your balance. Some purchases were recorded against a second, unread balance, so the credits looked as though they had never arrived. Every affected purchase has been folded back into the balance your workspace spends from, with the movement written to your credit history — no balance was reduced.
• Fixed Batch renders are priced per image. Asking a model for four images was charged as one. It is now charged as four, which is what it produces.
• Fixed Motion Ad is priced by length, like the rest of the studio. A thirty-second video made there was charged at a short clip's rate; it now costs the same as the identical render made in AI Studio, and tells you the price before it starts.
• Fixed Voiceovers are checked against their real per-second cost before they start, instead of being approved at a short clip's price and charged for the full length.
• Fixed Nobody can lock you out of resetting your password. "Forgot password" was limited to five requests an hour per email address, so anyone who knew your address could use up your five and leave you unable to request a reset. The limit is now counted per requester: a stranger's requests count against the stranger, and your own still go through. The page still gives the same answer whether or not an address is registered.
• Fixed Removing a generation from the AI Studio gallery answered "done" even when nothing was removed — for example a generation that belongs to another workspace, or one already gone. It now says "not found" in those cases, the same as opening it would.
• New Visual Effects — big-budget effects from explosions to surreal transformations, as fifteen one-click presets in AI Studio: Incline, Act natural, Skyline tightrope, Walking flame, Melting, World morphing, High flip, City giant, Infinite selfie, Cutout, Floating fall, Eyes in, Wild ride, Smash and grab and Studio slide. Click a card and the composer is filled with a tested prompt, the model it was tuned on and that model's settings — add your own photo, or just press Generate.
• New Presets built on the newest models carry a New model marker, and View all presets opens the full set.
• Improved Every preset card shows what one run will cost in credits before you click, using the same per-second video and per-image figures that are actually charged. A model covered by one of your passes reads as included, exactly as it does in the composer.
• New The public site has section pages. Create, Apps, Video Models, Image Models, Studios, Soul, Platform, Resources, Company and Community each have their own page, reachable from the header and footer of every public page. Each card links to the real tool behind it, and every model card carries the same price the studio charges. Where something in the site's plan is not part of the product yet, its card says Not available yet and is not a link, so nothing on the site promises a feature that does not exist. A single /sitemap.xml now lists every public page, including every model; the old models-only sitemap points there. The pages follow your light or dark choice, work at phone width and are translated into all 14 languages.

Model passes — unlimited use of a model for a month

• New Model passes in Billing — 13 passes covering the frontier catalogue, from ¥800 for Nano Banana Pro 2K up to ¥36,000 for unlimited access to every model. Buy one and that model generates free for 30 days: no credits are deducted while the pass is live.
• New The AI Studio storefront — a hero, a rail of unlimited bundles, filterable offer cards and an all-access card, each showing what the model actually produces. Every card's "Try it" opens that model in the composer, already selected.
• Improved Passes are a real entitlement, not a large credit grant dressed up as one, so "unlimited" means it. Buying the same pass again extends it rather than starting a second one, and a generation covered by a pass is still recorded in your usage even though it is not billed.
• Fixed A generation that failed under a pass would have refunded credits that were never charged — minting balance out of a failed job. Refunds now only return what was actually taken.
• Fixed Team invitations to an address that already has an account now always ask for that account's password. Previously the invitation link alone could open the existing account, including its other workspaces. Choosing a password for a brand-new address is unchanged.
• Fixed Suspending an account now signs it out everywhere at once. A suspended or deleted account could keep using the studio, the Ads Manager and AI generation on a session it had opened earlier, until that session expired on its own. Every request now checks the account's current status; reactivating an account restores its sessions without a new sign-in, and nobody else is signed out.

AI Studio — 38 frontier models behind one prompt

• New AI Studio rebuilt — one composer now drives every model. Pick a model, describe the shot, attach references, and the studio renders exactly the controls that model accepts: aspect ratios, resolution, duration, batch, sound, prompt adherence, multi-shot and output format all come from the model's own declaration, so no control is offered that the model would refuse. ⌘/Ctrl + Enter generates.
• New 38 frontier models — 8 image and 30 video: Higgsfield Soul 2.0 and Soul Cinema, Seedance 2.5 (plus Edit and Extend) and Seedance 2.0 (Fast / Mini), Kling 3.0 (Turbo / Standard / Pro / 4K / Motion Control), Kling 2.6 / 2.5 / O3 / O1, FLUX 2 and FLUX 3 Video, Grok Imagine 2.0 and Grok Imagine Video 1.5, Wan 3.0 / 3.0 Prime / 2.7 / 2.6, MiniMax H3 and Hailuo 2.3, LTX 2.5 Pro / Fast, Happy Horse 1.1 / 1.0, PixVerse 6, Ideogram 4.0, Recraft 4.1, Qwen Image 3, Z-Image Turbo and Higgsfield DoP. Searchable picker, grouped by image and video.
• New Media inputs by role — start frame, end frame, references, video and audio, each capped at what the chosen model actually takes (Seedance 2.5 accepts 30 references; Kling Motion takes a still plus a driving clip).
• New Generations persist — results survive a reload, a redeploy or a different device, and anything still rendering when you close the tab carries on. "Use in post" copies the file into your own media library, because the platform's asset links are signed and expire.
• New Admin → AI Studio and Model provider keys — connect generation with one Higgsfield platform key, and store per-vendor keys (OpenAI, Google, xAI, Black Forest Labs, Kling, ByteDance, MiniMax, Alibaba, Ideogram, Recraft, PixVerse, Lightricks, Fal, Replicate, ElevenLabs) with a direct link to each provider's console.
• Improved A failed or filtered generation refunds its credits, and the charge is only taken once the job is safely recorded — so a generation you cannot see is never one you paid for.
• Improved Eagle AI is now Eagle AI Search, renamed across the app and the site.
• Improved The whole studio is translated into all 14 languages, including the new per-model controls.
• New Ads Manager — connect Meta, TikTok and Google Ads, and AI plans, generates, launches and continuously optimizes the campaigns behind deterministic budget guardrails. Ad creatives are generated on the same frontier models as AI Studio.
• Improved Every image and video in the product now comes from a frontier model. Create → Images, the Autopilot swarm, the Ideas grid, Synthia's Content Crew, the Frame generator, reference images and Static → Motion Ad all generate on the real catalogue and keep the file in your media library, instead of returning a link to a free image service.
• Fixed The Frame generator charged 12 credits and handed back a free image it never stored. Same for Synthia's Content Crew, which charged and then had the picture generated a second time somewhere else.
• Fixed Static → Motion Ad was not a video model at all — it was a zoom over a still, and it was free. It now generates real video and charges for it.
• Fixed Reference-image generation refused to work at all unless a Google key was set; a Higgsfield key now satisfies it too.
• Fixed Frontier generations were invisible to your plan's monthly AI allowance, so they never counted toward it.
• Fixed The image proxy behind several of these screens was reachable without signing in, and could reach the paid engines. Paid engines now require a signed, expiring link.
• Improved The sidebar's 27 entries are grouped into Create · Grow · Publish · AI tools · Account.

Self-serve onboarding: verify, reset, invite

• New Password reset by email — a "Forgot password?" link on the sign-in page sends a single-use link that expires in an hour. Until now only an administrator could reset a password, so a locked-out account needed a support ticket.
• New Email verification — new accounts get a confirmation link, with a reminder in the app and a one-click resend. Confirming proves we can reach you for approvals, receipts and resets.
• New Team invitations — invite people by email as Owner, Editor or Viewer, change roles, cancel a pending invite and remove members, all from Settings → Team. Seats are counted against your plan (a pending invite holds one), and the last owner can't be removed or demoted.
• New Getting-started checklist on the dashboard — confirm your email, describe your brand, connect a channel, write and schedule your first post. Each step is measured from your real data, so doing it any other way still ticks the box.
• New Your account panel — change your own name, phone number and password (the current password is required) without contacting support.
• Fixed A new workspace showed the sample brand's content as its own — demo posts, channels and inbox items. The first edit saved them into the workspace as real data. A signed-in workspace with no content now starts genuinely empty.
• Fixed The Business profile shipped pre-filled with another company's details, so one click on "Save profile" made that the brand every AI draft was written from. It now starts blank, with guidance in each field.
• Fixed Plan & usage always read "310 / 500 posts" no matter the real figure — a zero count was treated as "no data" and replaced by a sample. Usage, the sidebar meter and the channel count now show your actual numbers, and the plan badge no longer claims "Pro" by default.
• Fixed The sidebar named a sample company instead of your workspace.
• Fixed The phone number collected at sign-up was discarded instead of saved.
• Improved Every new screen and message is translated into all 14 languages, and sign-up, sign-in and password-reset requests are rate-limited.

Video Studio, AI media & new pricing

• New Video Translate — dub any video into any language with the original voice kept, native (transcreated) burned-in subtitles and an AI caption. Editable telop, one-at-a-time bulk queue, and post directly from the tab.
• New Create AI Photo/Video Studio — an ElevenLabs-style Image & Video studio: generate images, AI video, voiceovers, sound effects and music, with per-language dashboards and one-click "use in post".
• New Static → Motion Ad — turn a still ad image into an animated motion ad (zoom/pan, headline, voiceover).
• New Video Downloader — pull videos from YouTube, TikTok, X, Facebook and more; the browser extension downloads Instagram in your own session.
• New Translate tab — localize any text into 15 languages in each country's native talking style (transcreation, not literal).
• Pricing New plans — Starter ¥11,920 · Pro ¥31,920 · Business ¥79,200 (5 / 1,500 / 3,000 credits). Credit packs re-priced, plus dedicated Video Translation tiers with a live price simulator on the pricing page.

Social Analyzer & team roles

• New Social Analyzer — a free Chrome extension that deep-scans any profile (engagement, best posts, business contact, sponsored-rate estimate) and an in-app dashboard with profile photos, clickable to Instagram.
• New AI outreach DM generation and one-click sync from the extension to your dashboard.
• New Sub-admin roles & granular permissions for the admin console.
• Improved Faster, more precise analysis with deeper post sampling.

Global launch: 14 languages & a new landing

• New The entire app and marketing site are localized into 14 languages — English, Japanese, Korean, Chinese, French, Italian, Spanish, Thai, Malay, Indonesian, Hindi, Romanian, Russian and Arabic (with full right-to-left Arabic). Switch from the 🌐 globe next to the theme toggle.
• New Redesigned landing page with a guided cursor walkthrough (create → generate → publish), an animated feature grid, and real per-language dashboard screenshots.
• Improved Sidebar now scrolls independently so every tab is reachable; more bottom scroll room across views.

Eagle AI & Synthia advisor

• New Eagle AI — detects AI-generated feeds, fake news and deepfake/AI-made people, with feed monitoring and a labeled-sample backtest.
• New Synthia — an AI fleet advisor that reads a snapshot of your whole workspace and returns a health read-out, growth plan and chat.
• New Influencer discovery — find and vet creators, folders, rate estimates and outreach.

AI content studio & publishing

• New AI copywriter, idea engine and image generation — grounded in your brand profile, tone and audience, with a fast Z-Image-Turbo preview engine.
• New Autopilot agent swarm, visual calendar & queue, categories, recycling and approvals.
• New 8 platforms via one-click OAuth — Instagram, X, Facebook, LinkedIn, TikTok, YouTube, Pinterest and Google Business — with an encrypted token vault.
• New Analytics, unified Engage inbox, Link-in-Bio studio and the Affiliate program (earn 10%).

Foundation

• New Workspaces & roles (owner, editor, viewer, client), multi-tenant, edit log & one-click revert.
• New Billing & credits — Stripe & PayPal checkout, credit ledger, live FX pricing.
• New Public REST API reference and a built-in changelog.
• Security Encrypted vault, role checks, audited access, and SSRF-hardened outbound fetches.