What's New

What changed in each Remix Studio release. For the full technical record, see CHANGELOG.md or the GitHub Releases page.

Spotted a bug or have an idea?

Please open a ticket on GitHub Issues — your reports and feature requests directly shape what ships next.

---

1.25.0 — Civitai projects, auto-tagging, and pages that keep up with your agents

Generate images and videos with any model on Civitai, let a vision model tag your albums and libraries, see what an agent changes without reloading, and watermark or split a single image in a couple of clicks.

Added

• Civitai projects — A new project type generates images on Civitai with any checkpoint, LoRA or embedding from its catalog, paid with your account's Buzz. Add your Civitai API key as a provider, then build the workflow: a checkpoint item and LoRA items — found by searching Civitai or by pasting a link — text that feeds the prompt or the negative prompt, and an image for image-to-image. Several checkpoints, LoRAs or strengths in one item are tried one per job, like a library, so a batch can compare them. Every job and every result keeps the settings and seed it ran with, and MCP agents can build and run Civitai projects too. See Civitai Projects.
• Civitai video — A Civitai project can make videos instead of images, with Wan 2.1, 2.2, 2.2 5B, 2.5, 2.7 or 3.0, or Hunyuan. LoRAs made for the engine and a negative prompt work as for images, and the first image becomes the start frame (Wan 2.7 and 3.0 also take an end frame). See Civitai video.
• Pages that keep up with your agents — What an MCP client, the assistant or another tab changes now shows up on the page you have open, without a reload: new libraries and prompts, new and renamed projects, a project's workflow, campaigns, posts, and posts the scheduler publishes. You only ever receive updates about your own work. See live updates.
• Quick watermark for one image — A stamp button on image album and library cards opens a new tab to watermark just that picture: adjust the text, position, size, opacity and color with a live preview, then Create downloads the watermarked JPEG. Nothing is saved to your storage, and your saved watermark settings stay as they were. See watermarking a single image.
• Split an image into layers — Dola Seedream 5.0 Pro Layer Decomposition on RunningHub takes one image from the workflow and returns a base image plus up to 16 transparent PNG layers, one per element, each saved to the album on its own. A prompt is optional: name what to separate, or leave it out to split every major element.
• Model explanations — Every model now explains itself: what it is for, how it differs from its neighbours, and limits worth knowing, like a prompt cap or a required source image. The explanation shows under the model in the project settings and on each card in the model selector, whose search now finds models by it too. In all six languages.
• Import a workflow from another project — Import Workflow in the workflow panel's menu opens your albums across projects of the same type — Civitai with Civitai, regular with regular; pick a result and its workflow and settings replace this project's after you confirm. See reusing configuration.
• Auto-tagging — A vision model can now tag your albums and libraries for you. Choose the provider, the model and how many requests run at once under Account → Preferences → Auto-tagging, then pick Auto-tag from an album's Actions menu, or the wand button in an image or text library. Each item gets 5–10 tags in your interface language, reusing the tags the collection already has where they fit, and items that are already tagged are skipped unless you say otherwise. Images and text for now.
• Results keep their sources' tags — Results drafted from tagged library items start with those items' tags, so the album filters by the same tags as the libraries behind it.
• Copy an ID for AI — Library items, album items and posts have a Copy ID for AI button. Paste the reference into the assistant or another AI chat and it knows exactly which item you mean.
• More for MCP agents — Agents can read one album or library item straight from a pasted reference, and rename many album files at once, by exact name or by swapping, adding or removing a filename prefix. See MCP Support.

Improved

• Remix a ComfyUI prompt — A ComfyUI text input took one typed text or one library, so a prompt could not be assembled from several materials the way a regular project's workflow builds one. A text input now offers Input to type the text, or Remix, which opens the familiar workflow cards right under it: add typed text and libraries — say a photo of, a library of subjects and a library of styles — drag them into order, and every combination of library items becomes its own job. See ComfyUI Projects.
• Copy or move results to a library or a project — Copy to… and Move to… in the album's Actions menu share one dialog: choose copy or move, then a library or a project, existing or new. Results can now also be copied into another project — with the job and workflow behind them, so Reuse workflow still works there — or moved into a library. You stay on the album when it is done, and the confirmation links to where the items went, in a new tab.
• The album remembers its page size — Each project's album opens at the page size you last picked for it.
• Faster projects, pickers and exports — Opening a project no longer loads every item of every library it uses, and Add to Draft builds the drafts on the server. The media picker opens without loading every project and library, the projects list loads faster for big albums, and the Exports page no longer downloads each export's full item list every few seconds.

Fixed

• Images from Gboard in the assistant chat box — On Android, a copied image, GIF or sticker picked in Gboard now lands in the assistant chat box as an attachment, in versions of Chrome that let the keyboard insert images.
• The assistant's Stop button — Stop now stops the running turn, and the conversation takes a new message straight away.
• Civitai and ComfyUI results in posts — Results from Civitai and ComfyUI projects can be attached to posts and campaigns.
• Uploads that vanished after a reload — A file imported into a ComfyUI input, files dropped onto a workflow, a new workflow item or an edited image could be gone after a reload.
• ComfyUI over a slow connection — A rented ComfyUI behind a slow link stays busy: the next job is sent as soon as ComfyUI finishes one rather than after its result has downloaded, slow downloads are no longer restarted after five minutes, and reference files are uploaded once per batch.
• Files with non-Latin names — An album file renamed with Chinese or other non-ASCII characters could no longer be renamed again, moved or copied.
• Selecting library media on a touch screen — Image, video and audio cards in a library have a checkbox, like album cards.
• Chrome on iPad — The app no longer runs off the bottom of the screen under Chrome's toolbar.
• Videos restored from the trash — They keep their length and resolution.

Changed

• Dola Seedream 5.0 Pro on RunningHub — The RunningHub entry called Seedream 5.0 Pro is now Dola Seedream 5.0 Pro, as RunningHub lists it, so it is no longer confused with Seedream V5 Pro, a separate RunningHub listing. Projects keep their selection.
• Nano Banana 2.1 on Google AI and Vertex AI — Google's newest image model replaces nano banana 2, which Google stops serving on October 29, 2026. Projects set to nano banana 2 now generate with nano banana 2.1, with nothing to change on your side.

1.24.0 — ComfyUI workflows, albums on your TV, and a notification when a batch is done

Run your own ComfyUI workflows in batches, watch project albums on a TV or open them in a file manager, hear back when a project's queue empties, and let an MCP agent upload files of its own.

Added

• ComfyUI workflow projects — A new project type runs a workflow on a ComfyUI instance you host — a local machine, a box on your network, or a rented GPU — so custom checkpoints, LoRAs and node chains can be batched like any other project. Export the workflow from ComfyUI with Workflow → Export (API), and every input is listed by node: numbers and toggles are edited in place, seeds are random for every job unless you pin one, and text and file inputs can draw from a library, so they combine and shuffle like regular workflow steps. The address lives on the project, so a rented GPU that comes back on a new address only needs it pasted in. Instances protected by ComfyUI-Login take an optional access password. See ComfyUI Projects.
• Albums on a TV — Project albums can now be watched on the living-room TV or opened in a file manager, all read-only and managed under Account → TV & devices. TV mode at /tv is a full-screen viewer built for a remote control, with slideshows and video playback; link a TV by scanning the code it shows with your phone. WebDAV opens albums in Finder, Windows Explorer, Infuse, Kodi or rclone. DLNA, off by default, lets TVs on your network find Remix Studio by themselves and browse it in their own photo and video app. See TV & Devices.
• A notification when a project finishes — Turn on Notifications under Account → Preferences and your browser tells you when a project has worked through its queue — how many jobs finished and how many failed — even with Remix Studio closed. Tap it to open the project. On iPhone and iPad, add Remix Studio to the Home Screen first. No third-party account is needed.
• File uploads over MCP — An external agent could organise your media but never add any. It can now upload images, videos and audio and put them in a library, a project workflow or a campaign, including one draft post per file. The file goes straight to the server rather than through the conversation. See file uploads.

Improved

• Edit and create a project from a dialog — Editing a project no longer takes you away from it, and creating one no longer leaves the list it will land in. On a phone the dialog is full screen and keeps the keyboard down until you tap a field.
• Moving album items, in a dialog that remembers where they went — Move to Project opens over the album instead of on a page of its own, so your selection is still there if you back out, and it proposes the project you last moved items to.
• Quieter TV slideshows — Only the pictures are shown; the caption and progress appear for a few seconds when you touch the remote. A running slideshow no longer lets the TV's screensaver start.
• Archived albums on the TV — TV mode has Active, Archived and All tabs.
• Album cards always show their date — The date on an album card disappeared whenever the card got narrow, so it came and went as the window was resized. Narrow cards now show it without the year. Albums also show at most four cards per row, and a wider window makes the cards bigger instead.

Fixed

• Sharing a photo into the app from Android — On some phones a shared photo arrived as Share Failed or as an empty preview, though the image had reached the app. It now comes through.
• Projects switched between models — A project set up on one model and moved to another could send the new model a size or quality it does not support, and the job failed. Each model now gets the closest setting it offers, keeping the picture's shape.
• GPT Image 2.5 on RunningHub — The aspect-ratio picker offered five ratios RunningHub's API rejects. It now offers the ten it accepts plus auto.
• RunningHub reference images — Jobs that uploaded a reference image to RunningHub failed with an unrecognised API key.
• Prompt dialogs under the sidebar — Edit Prompt and the album prompt dialog opened partly hidden behind the sidebar.
• Sign-in keeps your place — After an expired session, signing in returns you to the page you were opening rather than the home page.

Changed

• Qwen Image 3 replaces Qwen Image 2 on RunningHub — Qwen Image 3 Pro and Qwen Image 3 take the place of Qwen Image 2 Pro, and projects set to the old model move to Qwen Image 3 Pro. 21:9 is no longer offered; projects using it switch to 16:9.
• Solid cards and dialogs — Cards, panels, menus and dialogs are opaque now instead of translucent, so what is behind them no longer shows through their text.

1.23.0 — GPT Image 2.5, an assistant that survives a refresh, and a searchable album

OpenAI's newest image models arrive on both OpenAI and RunningHub, a chat no longer loses its turn when you reload the page, and the album gains search, a tidier toolbar and a way to move results into another project.

Added

• GPT Image 2.5 — OpenAI's newest image models are selectable on any OpenAI provider. They ship as two models rather than one: Sunburst is the base model tuned for quality, for work where editing precision and subject preservation matter, and Flare is the small, speed-tuned one, at roughly GPT Image 2's quality with up to half the latency. Both take the same sizes and aspect ratios as GPT Image 2, from 1024x1024 up to 3840x2160. Transparent backgrounds are back — GPT Image 2 had dropped them — and two quality tiers, xhigh and max, sit above high on the 2.5 models only.
• GPT Image 2.5 on RunningHub — The same two models are available through RunningHub's wallet instead of an OpenAI key, in an economy tier and an official-token tier. The official tier is not just a price difference: it takes 32,000-character prompts rather than 20,000, up to sixteen reference images rather than ten, the full quality range including xhigh and max, and a Background picker — the first RunningHub image model to offer one.
• The assistant survives a refresh — A chat turn used to live and die with the page that started it, so reloading, locking your phone, or a proxy timing out a long run left work still going server-side with nothing watching it — and a confirmation card that vanished left the conversation stuck waiting on a question it no longer asked. Turns now run independently of the page. Come back to a working conversation and it reattaches, replays what you missed, and follows the run to its result; a pending confirmation is still there waiting for you.
• Conversation resources in the assistant sidebar — What a conversation had worked on was announced with a big card in the message stream, repeated per tool call and lost as soon as you scrolled past it. The sidebar now lists them: one compact row per library, project, campaign or post, with the reason it is listed underneath (Create Prompt, Tag Album Items, Update Post Text), most recently touched at the top, each linking into the app. The list survives a reload.
• Move album items to another project — Results could be copied into a library but never moved into another project, so a batch generated in the wrong place could only be exported and re-imported, leaving the originals behind and losing their history. Move to Project now takes the items and everything behind them — the files, the job that produced each one, and its workflow snapshot with the reference media — so Reuse workflow still works from the destination. It opens a confirmation page showing what is selected, its size, what travels with it, and a picker for an existing project or a new one created on the spot.
• Album search — The album toolbar has a search box that matches an item's name and its prompt, across the whole album rather than the page on screen.
• A paged media picker — The picker loaded a flat 500 items per source and filtered them in the browser, so anything past the 500th was silently unreachable and every open pulled the whole slice down. It pages now, with the same pager the album uses, and searching and sorting cover the whole set.
• A searchable tag filter in the library preview — In a library with more than about three rows of tags, the rest were scrollable but invisible and read as missing. The filter now leads with your selection and the most-used tags, and All tags (N) opens a panel with search, a most-used / A-Z sort, counts, and a real scrollbar — a bottom sheet on a phone.
• The same library in several workflow steps — A library a step already used was greyed out, so a project could never draw two independent items from one library: two subjects from one subject library, or a foreground and a background from one image library. Every library is selectable now, and the Added badge counts the references (Added ×2) instead of blocking them.

Improved

• A tidier album toolbar — With items selected the toolbar carried nine controls. Export, copy to library, move to project, tag and delete now sit behind one Actions menu, which is what makes room for the search box.
• The post editor on phones — Attached media was a list of full-width rows that overflowed a narrow screen and left the thumbnails tiny. It is a grid of square tiles now — two columns on a phone, four on a large screen — with the controls as overlays, and left/right buttons so reordering works by touch. Dragging to reorder on the desktop is unchanged.

Fixed

• Moving album items reported a failure after it had worked — The move finished, then its cleanup step failed and the whole operation reported an error, so it looked as though nothing had happened when everything had.
• The album loading overlay covered the grid — While the album reloaded, a dimmed panel and a spinner were drawn over the results you were still looking at, and once the list had been scrolled they drifted into the middle of the grid. Refreshing now shows a small spinner in the toolbar instead.
• Album card overlays painted over the selection toolbar — A row scrolling under the toolbar left its checkbox and delete button drawn on top of the bar.
• Select-all checkboxes sat left of the row checkboxes — The header checkbox was a few pixels out of line with the column below it in the project viewer, on Exports, and in the text-library editor.
• The Rename and Tool approvals dialogs were muddy in light mode — Both resolved to grey instead of white. Dark mode is unchanged.

Removed

• GPT Image 1.5 — OpenAI shuts gpt-image-1.5 down on December 1, 2026, and it is retired from the catalog ahead of that date. GPT Image 2.5 Sunburst and Flare, GPT Image 2 and GPT Image 1 Mini are unaffected.

1.22.0 — Album tags, upload history per export, and a faster Campaigns page

Tag anything in a project album and filter by it, see every release a finished export has been through, and page through campaigns instead of waiting for all of them to load.

Added

• Album tags — Album items can now carry tags. Tag a single item from the control beside its filename, or many at once from the toolbar, where one dialog offers Add, Remove and Replace. With nothing selected, a batch applies to everything the current filters match rather than just the page on screen, so you can filter to one tag and add another across the whole slice in a single step. The toolbar's tag filter lists every tag in the album with a count beside it and a Match All / Match Any switch, and the filters live in the URL so a filtered album can be linked. Tags survive a delete and restore, travel inside project bundles, and are copied along when album items are copied to a library.
• Album tags in the assistant and over MCP — The assistant can list what a project is tagged with, ask for a slice by tag, and tag items for you — either a set you name or a whole filtered slice. Writing tags always asks for your confirmation first, and the prompt spells out whether it is adding, removing, or replacing.
• Upload history per export — Each finished export now has a history button beside its download and release actions, opening a dialog with just that archive's releases: where it went, which account, when, the resulting link, and the full error text if it failed. Previously the only record was the account-wide history page, where one archive's release had to be found among everyone else's.
• New models — Claude Fable 5.1, Gemini 3.8 Flash (on both Google AI and Vertex AI), Qwen3.8 Max, and Grok Imagine Image 2.0 join the catalog. The models they succeed stay listed, so pinned projects keep working.
• Scheduled post totals — The Scheduled Posts page now shows how many posts match, in the header, next to the search box as Showing X-Y of N, and in the pagination footer; the calendar view also reports how many fall in the month on screen. The Campaigns page shows the same total beside its Scheduled Posts heading.
• Version and commit in the sidebar — The bottom of the sidebar shows which version and which build you are running, with the commit linking to GitHub.

Improved

• The Campaigns page loads a page at a time — An account with dozens of campaigns used to load all of them, and every thumbnail in each, before the page could paint. Campaigns now arrive 12 at a time with a pager beneath the grid, and the page number lives in the URL so a page can be linked or survive a refresh. Search is now handled by the server: the box matches campaign names and descriptions across your whole account rather than filtering whichever campaigns had already loaded, and the command palette searches the same way.
• Confirmation before every destructive action — Batch Send Now published every selected post the instant it was clicked; it now confirms first, naming the campaign and the number of posts, and shows progress as the batch runs. The remaining gaps closed with it: sending or re-sending a single post, removing media from a post, batch unschedule, removing a passkey, deleting a project import, and granting or revoking admin.
• Tidier campaign cards — The overflow menu moved to the top-right corner and the post progress sits beside the title, which now gets the full width of the card. The dates lost the time of day, so they fit on one line even on a narrow phone; the campaign detail page still shows the time.
• Pages open at the top — Changing page in any list used to leave you wherever the pager was, partway down the new results. Every paginated list now returns to the top.

Fixed

• Campaign dates that cross a year — A campaign running December 2026 to February 2027 read as though it ended two months before it started, because the year was missing. Campaign, scheduled-post, and Recently Posted dates now show the year whenever it matters.
• Campaign dates and calendar counts on phones — The dates on campaign cards were clipped mid-value, the calendar's month summary collided with the next-month button, and the per-day post count was hidden entirely on a small screen. All three are fixed, and the desktop layout is unchanged.
• Long prompts on Grok Imagine Quality (RunningHub) — A prompt over 4,000 characters was rejected only after the job had been submitted. The limit is now known up front, so you get the same truncate/keep/cancel choice every other limited model offers. Projects set to the auto aspect ratio also work on text-only jobs, and auto is now actually offered in the ratio list.

Changed

• Sora 2 and Sora 2 Pro are gone — OpenAI is discontinuing the Sora API on September 24, 2026 and has published no successor, so both models have been removed and OpenAI no longer appears under video generation. A project pinned to Sora needs another provider's video model. OpenAI's text and image models are unaffected.
• Grok Imagine video moved to 1.5 — Existing projects keep working and gain 1080p alongside 720p, and durations from 4 to 15 seconds rather than just 6 and 10.
• Gemini preview models now use their final names — Three Gemini IDs still pointed at preview endpoints that Google has since moved or shut down. Saved project selections still resolve; nothing to change on your side.
• Grok Imagine Pro on RunningHub is now called Grok Imagine Quality — Only the display name changed; it always called the quality tier, which is what the Grok provider lists under that name too.

1.21.1 — New models, one-click voice send, and a calmer interface

Grok 4.6 and Gemini 3.7 Flash arrive, the mic and Send button finally work as one, and two new switches let you tame the interface if your browser struggles with it.

Added

• Grok 4.6 & Gemini 3.7 Flash — Both new releases are available for text projects and the assistant. Grok 4.6 brings a 500K context; Gemini 3.7 Flash brings 1M tokens in and 64K out, on both Google AI and Vertex AI.
• Glass effects & Hover animations switches — Two new appearance switches in Account preferences, both on by default. Glass effects controls the blur behind translucent panels and dialogs; Hover animations controls whether cards lift and zoom under your pointer. If the interface ever flickers or tears as you move the mouse across a grid, turning off Glass effects is the first thing to try.

Improved

• Send while you're still talking — Press Send while the mic is recording and it now stops, transcribes and sends in one go, instead of making you stop, wait, then press Send. Enter and Cmd/Ctrl+Enter do the same. If the transcript fails or picks up no speech, nothing is sent and your draft stays put.
• A calmer project pane — The tab bar, the selection bar and the job rows used to sit at three slightly different shades. They now share one flat colour, top to bottom.

Fixed

• White bars in dark mode — Sticky bars turned white in dark mode when Glass effects were off. They now match their surroundings.

Changed

• Grok Imagine Pro now points at Grok Imagine Quality — xAI deactivated the Pro tier in May 2026, so projects pinned to it had no working model upstream. Existing selections keep working and need no changes on your side.

1.21.0 — MiniMax, a posting trend chart, and assistants that can run your projects

A new provider across all four modalities, new image and video models, a chart that shows what actually went out, and an assistant that can draft, start, and clear generation jobs for you.

Added

• MiniMax — MiniMax is now a provider of its own, covering all four modalities: eight M-series text models led by MiniMax-M3 and its 1-million-token context, Image 01 with an optional portrait reference for a consistent character, Hailuo H3 video at 768P or 2K for 4 to 15 seconds, and Music 3.0 for both instrumental and vocal tracks. The API URL can be pointed at the mainland China endpoint, and MiniMax works with the in-app assistant.
• More image and video models — Seedream 5.0 Pro joins the BytePlus provider with its 1K, 1.5K, and 2K tiers and up to 10 reference images, and Seedream 5.0 Lite gains 4K. GPT Image 2 Official sits beside the economy tier on RunningHub, so the cheaper and the full-price route to the same model can be picked per project. MiniMax Hailuo H3 arrives on RunningHub too, in a frame-to-video form and a multimodal reference form that takes images, videos, and audio clips.
• A chart of what actually went out — A daily posting trend chart appears above Recently Posted on the Campaigns page and as a view on campaign history, with a red line that shows up only when the range contains failures. Ranges run from 7 days to all time; longer spans roll into weeks or months so a year stays readable, and the range you pick lives in the URL for linking and bookmarking.
• Assistants can run a project, not just build one — New tools let the in-app assistant and connected agents draft jobs from a project's own workflow, start them, read the job counts, watch the queue across every project and provider, and clear failed jobs so a stalled project resumes. Agents can also export a whole project bundle or an album archive, list the posts in a campaign, and check which account they are signed in as. Every result now carries a link back into Remix Studio, so an assistant can point you at what it just touched instead of quoting an id.
• Set the same text on many posts — A Set Text action beside AI Generate writes one block of text to every selected post in a single step, with no model involved.
• Bulk export cleanup and bigger library pages — Exports can be selected and deleted in bulk, and the library item list offers 25, 50, 100, or 200 items a page, remembered in the URL. A failed release's full error message now opens in a dialog instead of being cut short.
• The media picker remembers where you were — Picking a workflow image or campaign media reopens on the library or album you used last, per feature and across projects, rather than always starting at the first library.
• One-step sharing — Sharing into the app from your phone now offers Save to Library and Save to Project directly, instead of asking a second time on the import page.

Improved

• Batch text generation is a full-screen editor — The prompt field is no longer pushed below the fold: configuration lives in a left rail and the editor fills the rest at full height, phones get Prompt and Options as tabs, and Cmd/Ctrl+Enter generates.
• Release history and project import moved — Release history is now a sibling of Releases under Exports, and importing a project bundle has its own page reached from the Projects header. Old links redirect.
• Up-to-date dependencies — Every dependency was upgraded to its latest release, clearing the outstanding high and moderate security advisories.
• Clearer documentation — Added a guide for calling the MCP tools over plain HTTP, for clients without MCP support.

Fixed

• Light mode is readable again — The model selector and several other dialogs composited into a muddy grey in light mode, hover faded text to invisible in 14 places, and hover borders snapped to near-black in 12 more. Dark mode is unchanged.
• Campaign batch generation no longer fails on the model — A batch could die against the provider with an unhelpful "model not found". Both the stored and the upstream model id are now accepted, the Gemini model map is current, and when a batch does fail the actual reason reaches you instead of just a count.
• A project's jobs pane fits its column — On a narrow desktop window the pane's toolbar buttons overlapped and album cards shrank to about 95px. The pane now sizes its contents against its own width.
• One album card per row on phones — Portrait phones were showing two cramped cards side by side.
• Number fields can be cleared — The last digit in a number box can be deleted so a value can be retyped from scratch.
• Drawing on a workflow image keeps its size — A draw-only edit was saved back upscaled to an arbitrary size instead of the image's own dimensions.
• The history date filter applies to both views — The list and the chart now read one shared range, changes apply immediately, a single bound stays open on the other side, and days are cut in your own timezone.

1.20.0 — Portable projects, more places to release, and a phone-ready workspace

Pack a whole project into one file, send finished exports to any number of cloud drives, reuse a setup straight from the album, and work comfortably from a phone.

Added

• Export and import a whole project — Export Project on a project card's menu packs a project into a single portable ZIP: its settings, workflow, and album metadata plus every file they reference. The bundle lands on the Exports page like any other archive, so it can be downloaded, released to a drive, or sold. Drop a .zip back onto the Exports page to import it — always as a new project, so a bundle can sit beside the project it came from. Bundles carry finished work, not generation history.
• Releases: every destination in one place — The store settings screen is now Releases, reached from the Exports header. It holds storefronts and cloud drives together, and drives are no longer a single slot: connect any number of them, including several accounts on the same provider. OneDrive joins Google Drive, existing connections carry over automatically, and release history covers drive uploads as well as storefront publishes.
• Kimi (Moonshot AI) — Added Kimi as a provider, with the Kimi K3 text profile: a 1-million-token context window and native vision, so reference images can be attached to a text workflow. The API URL can be pointed at the mainland China endpoint, and Kimi providers work with the in-app assistant.
• Seedance 2.5 video — BytePlus adds Dreamina Seedance 2.5, which accepts video and audio references alongside image frames — up to 30 images, 10 videos, and 10 audio clips per job. Audio generation now follows the project's sound setting.
• Reuse a workflow from the album — Album, text, and audio entries each carry their own reuse control, and the image lightbox has one too (press R), so a setup can be picked while looking at the finished piece. If a result has no stored snapshot, the workflow is rebuilt from its own prompt and references, and the confirmation says so.
• Bulk text edits from an assistant — Connected agents can rewrite the content of many library items in one batch instead of one call per item, and approval prompts now name how many items are affected and what is changing.

Improved

• A workspace that fits a phone — The media picker is a full-height sheet with a two-column grid, dialogs use the real viewport height so their buttons stay reachable, page headers are compact and scroll away with the content, and workflow image references are shown whole instead of cropped to a letterbox slice. Desktop layouts are unchanged.
• A calmer, faster interface — Fixed the pointer-move flicker in Chrome and Safari, removed backdrop blur from cards, the main layout, the album, and the recycle bin, and made progress bars update without forcing the page to re-lay-out.
• Clearer documentation — The README and the docs site were rewritten and audited against current behaviour.

Fixed

• Reuse no longer reports "no longer available" — Adding a second batch of drafts erased the saved setup of every draft, pending, and failed job already in the project, so results from those jobs could not be reused even though nothing had been deleted. Saved setups now survive bulk changes, and they no longer carry links that expire.
• Text library actions on desktop — Copy, tag, edit, and delete buttons on text library rows are visible without hovering.

Removed

• MEGA as a release destination — MEGA has no OAuth API, so connecting an account meant storing its email and password in full: a secret that cannot be limited to one folder, cannot be revoked without changing the password, and grants whoever holds it the entire account. MEGA connections and their stored passwords are deleted on upgrade, queued MEGA releases are marked failed, and past MEGA releases stay in Release History. Google Drive and OneDrive, which authorize over OAuth, remain.

1.19.0 — Smarter assistants & cleaner workflows

Hide workflow clutter, let assistants work reliably through bigger batches, and move around long lists more easily on every screen size.

Added

• Hide disabled workflow items — Use the workflow's three-dot menu or press H to hide or reveal disabled items. The menu shows how many are hidden, and your choice is saved with the project so it follows you across reloads and devices. Hidden items keep their real positions when you reorder the workflow.
• Claude Opus 5 — The Claude provider now offers Opus 5 with a 1-million-token prompt limit and up to 128K output tokens.
• Assistants can browse and fetch album files — New MCP tools let the in-app assistant and connected agents page through a project's album, inspect image details, and turn internal media keys into temporary view or download links. Links are issued only for media owned by the signed-in user.
• Numbered pagination — Projects, libraries, exports, chats, campaigns, scheduled posts, administration pages, and project tabs now offer direct page-number navigation, first/last jumps, and compact five-page skips.
• Automatic truncation recovery — When a model reaches its output limit in the middle of a response or tool call, the assistant can now ask it to continue automatically. Larger bulk jobs also get more room to finish and clearer progress reporting.

Improved

• Phone-friendly lists — Pagination condenses to fit narrow screens, uses larger touch targets, and wraps cleanly beside page totals. Text library entries now show a useful two-line preview and keep edit actions available on touch devices.
• Quicker dashboard navigation — Select the Recent Projects, Libraries, or Campaigns heading to open the corresponding full list.
• Clearer documentation — Refreshed the release history and expanded the MCP and model-maintenance guides, including documentation for album browsing and temporary file URLs.

Fixed

• Reliable Gemini batch work — Gemini 3.5 Flash Lite now preserves every parallel tool-call ID and uses a more suitable thinking level for multi-step work, so campaign batches and repeated calls no longer fail or stop part-way through.
• Fewer unexpected sign-outs — Concurrent refreshes from multiple tabs and responses lost during a network interruption no longer end an otherwise valid browser session.
• Resilient OAuth connections — MCP and other OAuth clients can recover safely when a rotated refresh-token response is lost, while stale-token reuse is still detected and revoked.
• Visible mobile chat controls — The conversation drawer no longer slips underneath the fixed phone header, so search and new-chat buttons remain accessible.
• Drafts visible after fullscreen work — Adding workflow items as drafts now closes the expanded workflow after a successful save, taking you straight to the new drafts.

1.18.0 — A wave of new models & snappier project tabs

New text and image models across five providers, project tabs that load on demand and update instantly, and quality-of-life upgrades throughout the workspace.

Added

• New text models — Google AI and Vertex AI add Gemini 3.6 Flash and Gemini 3.5 Flash Lite; OpenAI adds GPT-5.6, GPT-5.6 Terra, and GPT-5.6 Luna; Anthropic adds Claude Sonnet 5; and xAI adds Grok 4.5. Gemini 3.6 Flash is now the default Gemini text model.
• New image models — RunningHub adds nano banana Pro, Seedream 5.0 Pro, Seedream V5 Pro, and Wan 2.7 Pro. Google AI and Vertex AI add nano banana 2 Lite.
• Auto aspect ratio — nano banana 2 on RunningHub gains an "auto" option that lets the model pick the output ratio itself.
• Optimized or original — When picking album images in the media picker, choose whether you want the optimized version or the original file.
• Filter media by ratio — Narrow album images and videos in the media picker to one or more aspect ratios, with counts for every available ratio.
• Save to library — Send a workflow item's text or image straight to a library with one click.
• Reorder cover images — Arrange product cover images in the order you want on the sell page.
• Lossless library transfer — Export and import text libraries as JSON, so prompts with newlines, colons, or list-like lines survive round-trips byte-for-byte. The simpler plain-text format remains available.

Improved

• Faster project tabs — The Draft, Queue, Done, and Album tabs load their content only when needed and remember content you've already viewed, making tab changes faster.
• Instant album updates — Deleting album items now updates the album, item counts, and pagination immediately.
• Ready-to-view drafts — The Draft canvas shows your newest images as soon as the project opens.
• Safer confirmations — Confirmation dialogs show progress and prevent double-submission while their action is running.
• Library editor polish — Cleaner typography, a refined toolbar, and localized timestamps.

Fixed

• Deleting the current image in the album lightbox now shows the next image immediately instead of leaving a stale one on screen.
• Wan 2.7 prompts longer than 2,048 characters are shortened to the supported limit instead of causing the job to fail.
• Drawing and cropping in the image editor now land exactly where you point, and saved edits keep the image's full resolution.
• Assistant message actions and attachment removal buttons are now visible and usable on phones and tablets.

1.17.1 — Light mode polish

A cleanup pass for light mode, plus tidier fullscreen workflow cards and smarter pagination.

Improved

• Case-insensitive tags — Library tag filtering now matches tags regardless of case, so Portrait and portrait behave as one tag.
• Smarter pagination — The pagination bar in the Album and Done tabs stays out of the way when everything already fits on one page.

Fixed

• Light mode, everywhere — A full audit fixed unreadable labels on colored buttons, invisible controls over image overlays, leftover dark-only text and hover states, and an inconsistent login card.
• Softer shadows — Toolbars and filter dropdowns no longer cast heavy black shadows in light mode.
• Centered draft canvas — The empty draft canvas now sits in the middle of the tab instead of hugging the top.
• Fullscreen workflow cards — In the fullscreen workflow view, cards now fill their grid cells cleanly: text scrolls inside its card, and images and videos fill the available space.
• Library hover border — Removed the harsh border that flashed when hovering library items in light mode.

1.17.0 — Fullscreen workflow view & wake lock

Give your workflow the whole screen, and keep your screen awake during slideshows.

Added

• Fullscreen workflow — A new fullscreen button in the workflow panel header (next to the assistant button) expands the workflow across the entire project view, tucking away the Draft, Queue, Done, and Album tabs. Your workflow items spread out into a grid of equal-sized cards, each scrolling on its own when its content runs long. Click the button again to return to the split view.
• Wake lock for slideshows — Your screen will no longer go to sleep while playing an image slideshow in the lightbox.

Improved

• Model availability docs — Supported text, image, video, and audio models are now listed in separate tables, with a provider summary to make compatible services easier to compare.

Fixed

• Posts that should be skipped are now identified correctly during campaign processing.

1.16.1 — A more immersive lightbox

Fullscreen viewing gets out of the way, and the delete prompt plays nicely with it.

Added

• Immersive fullscreen — In fullscreen, the image fills the whole screen and the controls fade away after a few seconds of inactivity. Move the mouse or press a key to bring them back.
• Confirmation shortcuts — When the image delete prompt is open, press Escape or D again to close it without deleting the image.

Fixed

• The delete confirmation dialog now shows correctly while viewing an image in fullscreen.

1.16.0 — Image slideshow & a documentation site

The image lightbox grows into a full viewer with slideshow playback, and the project gets its own documentation site.

Added

• Image slideshow — Play through your images hands-free with play/pause controls and a circular countdown. The interval is adjustable and remembered for next time.
• Slideshow transitions — Choose how slides change: fade, slide, zoom, blur, or an Android-style ripple. Your choice is saved across sessions.
• Fullscreen & keyboard shortcuts — Toggle fullscreen and drive the lightbox from the keyboard (playback, fullscreen, delete, and interval adjustment), with hotkey hints shown on hover.
• Documentation site — A new documentation site provides guides to Remix Studio's concepts, integrations, deployment, and operation.

1.15.0 — Edit images in place

Crop and draw on your generated images without leaving the app.

Added

• Image editor — A modal for cropping and drawing directly on workflow images, with a reset option to revert your edits.

Improved

• Assistant settings navigation — Returning from assistant settings now lands you back where you started.
• Prompt editor — Cleaner rendered markdown and a simplified, single-pane editing experience.

1.14.1 — Faster, smoother workflow building

Get media into your workflow faster, and keep it stable under rapid edits.

Added

• Drag-and-drop into workflows — Drop media files straight into the workflow list.
• Paste support — Paste text and media into the workflow with Cmd+V / Ctrl+V.
• Auto-scroll — The workflow list scrolls to the newest items automatically.

Fixed

• Resolved a memory leak when working with media items in long sessions.
• Fixed an issue where rapid workflow edits could collide and fail to save.

1.14.0 — Threads, watermarking & sharper campaigns

Publish to Threads, protect your exports with watermarks, and work with more flexible libraries.

Added

• Threads support — Connect Threads as a campaign channel and publish to it, with clear error messages when something goes wrong.
• Album export watermarking — Add watermarks to album exports from a dedicated settings panel.
• Product cover watermarking — Apply per-product watermarks to listing covers automatically.
• Library tag match modes — Filter libraries and workflows by matching all tags (AND) or any tag (OR).
• Switch a workflow item's library — Change the source library on an item without rebuilding it.
• Album page size selector — Control how many album items load at once.

Improved

• Campaign media now generates in the background with live status, so the UI stays responsive.
• Shareable album views — Album filters and sorting live in the URL, so views persist and can be shared.
• Refreshed Library Preview modal with a responsive layout.

1.13.0 — Live project updates

Projects update themselves in real time as work completes.

Added

• Real-time project status — Project pages update automatically as work progresses, with no manual refresh needed.
• Social profile refresh — Re-sync social account profiles and avatars, with automatic recovery when an image fails to load.

Improved

• More reliable job starts and reduced unnecessary background refreshing.

1.12.1 — Smoother project viewing

A polish pass on the project viewer and album browsing.

Improved

• Faster job reuse — Details for a previous job load only when you choose to reuse it, keeping the project viewer responsive.

Fixed

• Album counts and thumbnails update immediately after deleting items.
• Fixed cases where navigating or deleting images in the project lightbox could show the wrong image or lose your position.

1.12.0 — Pagination for big projects

Large projects load quickly thanks to paginated albums and job history.

Added

• Faster large projects — Albums and completed jobs are split into pages, with controls for choosing how they are sorted.
• Clearer confirmations — Confirmation dialogs show when a slower action is still running.
• Visible campaign dates — Campaign pages now show their scheduled date range.

Fixed

• Completed jobs are no longer lost when partial project saves happen.
• For administrators: Server health checks no longer stall during startup, so containers report their status promptly.

1.11.0 — Custom storage domains

Serve exported assets from your own domain.

Added

• S3 custom domains — Self-hosting administrators can serve exported assets through a public S3 endpoint or their own domain.

Improved

• Reusing a workflow now also restores the matching provider and model.

1.10.3 — More models, better watermarks

New image models and cleaner watermark rendering.

Added

• New and updated models — Added Grok Imagine Pro and updated the Google AI and Vertex AI Gemini model profiles to Gemini 3.5 Flash.
• Job media in Completed tab — See the context media behind each finished job.

Improved

• Watermarks render more consistently and position more accurately.
• Orphan files—stored project files that are no longer connected to a live project item—now display in a responsive grid, making them easier to review and clean up.

1.10.2 — Send to chat & mobile sharing

Bring content into Remix Studio from your browser and your phone.

Added

• Reuse a job's exact setup — Restore a past job's full workflow snapshot and generation settings back into the active project.
• Send to chat (browser extension) — New right-click options send an image or text to the Assistant and start a new conversation.
• Android share target — The installed app now appears in Android's share sheet. Share text or media to save it to a library or project, or to start a new assistant conversation.

1.10.1 — Smarter extension imports

Imports remember your preferences.

Improved

• The browser extension remembers separate destinations for text vs. image imports.

1.10.0 — Browser extension polish

A more capable, more reliable browser import flow.

Added

• Smarter image names — Imported images use the description supplied by the source website when available, falling back to the image URL.
• Remembered destinations — Your import destination is saved automatically.
• The browser extension is now attached to each release for easy download.

Fixed

• No more infinite loading when opening the import page without extension data.

Improved

• Refreshed import page design to match the rest of the workspace.

1.9.0 — Sell your exports

Turn finished exports into products you can sell, plus a more capable assistant.

Added

• Digital store integration — Connect Gumroad and manage products to sell your exports, with a publishing history view and a publish-immediately option.
• Assistant tool approvals — For each assistant action that writes data, choose whether it should ask every time or run automatically in that conversation.
• New models — GPT Image 2, GPT-5.5, and Grok 4.3.
• Google Drive delivery confirmation — Review and confirm Google Drive uploads before delivery begins.
• Focused search — Search libraries and projects by name without matching their contents.

Fixed

• The command palette now closes with Escape.

Improved

• Easier export browsing — Export tasks are split into pages, and album images use a space-efficient masonry layout.

1.8.0 — Command palette

Find or create core workspace items without leaving the keyboard.

Added

• Command palette — Press ⌘K on macOS or Ctrl+K on Windows and Linux to find projects, libraries, and campaigns, or create a new one without leaving the keyboard.

1.7.5 — Deeper campaign tooling

A richer campaign workspace with analytics and batch watermarking.

Added

• Post detail view with scheduling, AI generation, and management actions.
• Campaign analytics — Post status counts and summaries with URL-synced filtering.
• Batch post watermarking with a live preview.
• Campaign assistant tools for managing campaigns and posts via the assistant.
• Universal media picker for consistent asset selection across projects and campaigns.

Fixed

• Media display prefers full-resolution sources over thumbnails.
• Replaced browser confirm dialogs with the in-app confirmation dialog.

1.7.4 — Easier media selection

Find the right assets faster, while inactive campaigns remain safely paused.

Added

• Search and filter by source — In the media picker, search your libraries and project albums to find the right source before choosing an item.

Fixed

• Posts in inactive campaigns are no longer processed or published.
• Long campaign post links now truncate cleanly.

1.7.3 — Memory monitoring

New diagnostics help self-hosting administrators investigate server memory usage.

Added

• Server memory diagnostics — Self-hosting administrators can inspect memory usage through a dedicated status endpoint and server logs. See Memory Monitoring for setup and interpretation guidance.

Improved

• Refined campaign list layout.

1.7.2 — Bulk selection & reliability

Select long ranges of items more quickly and get clearer results from batch uploads.

Added

• Shift-click range selection for project jobs and media picker items.
• Architecture overview — A new architecture diagram gives contributors and administrators an overview of Remix Studio's features and system structure.

Fixed

• More reliable batch uploads with per-item error handling.

1.7.1 — Theme polish

Theme changes feel smoother and stay in sync with your device.

Added

• Smooth theme transitions with a circular ripple animation.
• Theme-aware thumbnail borders.

Fixed

• The UI now reliably follows your OS light/dark preference.

Improved

• Enabling or disabling an assistant provider now saves immediately, without a separate Save action.

1.7.0 — A media-first home

A redesigned home and cards put your media front and center, with full campaign localization.

Added

• Media-focused home with horizontal media carousels.
• Redesigned project and library cards with image backgrounds and quick-action menus.
• Campaign localization in English, French, Japanese, Korean, and Chinese (Simplified & Traditional).
• Project deletion directly from list and card menus.

Fixed

• Portrait image covers now crop from the top instead of the center.
• New installs default to the "System" theme.

Improved

• Renamed "Prompt Fragments" to "Items" throughout for clarity.

1.6.0 — Batch generation & polish

Generate library text in batches, track its progress, and manage queue problems more easily.

Added

• Batch AI text generation — Generate multiple text library items in one operation and follow progress while the batch runs.
• Optional model reasoning — Provider settings can request supported models' reasoning output, while internal reasoning markers are removed from generated campaign posts.
• Faster library copying — Copy library content to the clipboard with one click.
• Clearer queue controls — The Queue Monitor is easier to scan and now includes an action for clearing queued work.

Improved

• Consistent rounded corners across the UI.

1.5.3 — Backups & X (Twitter) rebrand

New maintenance tools protect self-hosted data, while the social integration moves to X's current branding and endpoints.

Added

• Database backup and restore tools — Self-hosting administrators can create and restore database backups, with optional automatic cleanup of older backups. See Backup & Restore for instructions.

Improved

• Migrated Twitter endpoints and branding to X.

WARNING

Versions 1.5.0 and 1.5.1 are known to be broken. Upgrade to version 1.5.2 or later before using campaign features. See the CHANGELOG for technical details.

1.5.0 — Social campaigns arrive

The first version of the campaign workspace for planning, scheduling, and publishing social posts.

Added

• Social campaign management — Campaign lists, detail pages, history, channel configuration, scheduled posts, and post creation.
• X/Twitter channel integration for social posting.
• Batch post creation, AI generation, and scheduling flows.
• Campaign media imports from libraries and projects.
• Campaign assistant tools.

1.4.x — Queue monitoring

Visibility into what the background queue is doing.

Added

• Queue Monitoring system with a dedicated UI for tracking projects and providers.
• Detailed generation options (resolution, quality, aspect ratio) in the expanded job view.

Fixed

• Failed task errors are no longer cut off — click to expand the full message.

1.3.0 — Reliable queueing

Background generation recovers more reliably, and library ordering is consistent for everyone.

Added

• More reliable generation queues — Provider concurrency limits are managed centrally, and interrupted jobs can be recovered instead of remaining stuck.
• Consistent library sorting — Libraries now use server-managed sorting instead of manual drag-and-drop, so item order stays consistent across sessions.

Improved

• Modernized the RunningHub video generator integration.

1.2.0 — More providers & bulk library actions

Generate with Alibaba's Qwen models and organize multiple library items at once.

Added

• Alibaba Cloud DashScope — Connect DashScope and use its supported Qwen models in Remix Studio.
• Batch library actions — Copy or move multiple library items in one operation.

Fixed

• Prevented an error on the assistant page when no providers are configured.

1.1.1 — Descriptions everywhere

Add context to projects and libraries so they are easier to recognize and organize.

Added

• Project and library descriptions — Add a description that appears alongside each project's or library's name.

Improved

• Cleaner project and library card layouts.

1.1.0 — Assistant & album improvements

Find assistant conversations faster and manage large project albums more precisely.

Added

• Assistant tools overview — See which workspace actions the assistant can perform.
• More precise album selection — Filter project albums by aspect ratio and limit bulk actions to the current selection.
• Faster assistant access — Search assistant conversation history or start a conversation from a specific library.

Improved

• Workflows trigger automatically after clearing failed jobs.

1.0.0 — Initial release

The first release of Remix Studio — a self-hosted AI workspace for batch content generation.

Added

• Self-hosted AI assistant workspace for orchestration and batch content generation.
• Project workflows built from reusable text, image, video, and audio libraries.
• Draft generation through combination and shuffle workflows.
• Background generation queue with provider-specific execution.
• Provider credential, model profile, alias, and concurrency management.
• S3-compatible storage and ZIP export workflows.
• Built-in assistant and MCP support for operating libraries, projects, albums, models, and storage.
• Authentication, admin controls, 2FA, passkeys, and user storage limits.
• A UI localized in English, Simplified Chinese, Traditional Chinese, Japanese, Korean, and French.