FlowWall
FlowWall AI
Home
FlowWall-1.0.11Latest

Highlights

  • SVG animation is now "Code animation". Describe what you want in one sentence and get an animation in all kinds of styles — "UI motion", "Vlog text", "Cinematic title", "Cartoon short" and more — up to 60 seconds long. A new "Style" option offers nine styles to pick from; leave it unset and the model decides from your description. Text and titles meant to sit on top of a video automatically get a transparent background. Claude Opus 5.5 generates it by default, and while you wait you can see whether it is thinking, planning, writing code or checking; when it is done, it plays the animation through once on its own and fixes any problems it finds before handing it to you. In simple text and logo animations you can edit the words directly, and once exported as a video they can feed into "Video composition" and other downstream nodes. Existing SVG animations still play, take text edits and export as before.
  • Split by shot in Video clip and composition. The "Video clip" node gains "Detect shots", which finds every cut in the video automatically; select any one shot to export it on its own. The "Video composition" node gains "Auto-split by shot", which cuts a stretch of video on the timeline into several pieces by shot in one click.
  • Share DetailX before/after with one link. Open the DetailX banner at the top of the home page and drag the slider in the middle to compare the picture before and after upscaling. The comparison page has its own link: send it to friends and they can open it and watch right away, and anyone who wants to use it just clicks "Try it now". You can also open it from the "Video upscale" section of the Tutorial.

Improvements

  • The right-click menu and the add-node menu you pull out of a connection now use the same icon style as the "+" menu.
  • Files downloaded from a Flow are now named after their node (for example "Image 12") instead of a string of numbers, and Japanese, Korean and other names are kept exactly as they are.
  • Full-screen video on the home page and in work details plays more smoothly, without stuttering.
  • The LLM node's version bar now looks the same as the image node's, and it no longer sits off to the left.
  • When you build a Flow over MCP from Claude or ChatGPT, a code animation can be written straight into a node; every auto-assigned node name can be referenced with @ right away; and for results with no picture, such as text, subtitles and motion graphics, the card shows a summary of the content.

Fixes

  • Fixed the LLM occasionally returning a blank response or cutting off mid-sentence.
  • Fixed longer SVG animations failing to export, and exported videos running longer than the set duration.
FlowWall-1.0.10

Highlights

  • Claude Opus 5.5 and GPT-6 Sol are here. Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol, the two newest models, officially land on FlowWall AI today — pick them on any LLM node in a Flow. Claude Fable 5.1 and GPT-6 Luna arrive alongside them.
  • DetailX upscaling, upgraded again. The DetailX video upscaling model takes another step up: upscaled footage now plays smoother and silkier. The new "Light correction" makes lighting look more natural and the picture more true to life.
  • A big MCP update. When you use FlowWall from Claude or ChatGPT, results and canvas previews show up as cards right in the conversation, in 12 languages. You can upload your own files straight from there, and connecting and signing in are simpler too.

Improvements

  • The SVG animation node now lets you choose its model, and the node card has had a round of polish, so it is smoother to work with.
  • Transparent reference images now sit on a checkerboard everywhere, so you can tell at a glance that they are transparent.
  • The download progress bar shows the real percentage.
  • The Changelog page now shares its top bar with the Tutorial, with a language switch at the top right.
  • On a tablet in portrait, the home page shows three event cards per screen at the top; the showcase sections below use one column on phones and two on tablets in portrait.
  • On narrow video nodes on the canvas, the "Run" button no longer crowds the playback bar.
  • Seedream 5.0 Lite reference images that are over the size or dimension limits, and reference images too large for the Claude models, are flagged before you run instead of failing mid-run.

Fixes

  • Fixed Video upscale, Video loop, Frame interpolation and similar nodes failing with "video record not found" in a copy of someone else's shared or public Flow.
  • Fixed untranslated text on the Audio Generation screen in 10 interface languages.
FlowWall-1.0.9

Highlights

  • New "Audio Generation": one prompt, a whole sound scene. Powered by Seed Audio 1.0. This is not text read aloud — write the characters, dialogue, emotion, sound effects and setting, and the model produces the whole thing in one pass. Attach up to three reference voices (pick them from the voice library or upload a clip of your own) and use @ in the text to hand a voice to each character; add timestamps to control which second a line lands on, and tune speed, pitch and volume separately. Stuck on the wording? "Polish prompt" fills it out for you. Templates come in four groups — Text Gen / Reference / Time control / Multilingual — each with a finished sample you can listen to first. It lives on the audio page and on the audio generation node in a Flow, and everything you make stays under "History".
  • New "Grid group": arrange images into one finished picture. Turn a few selected images into a grid group and they drop into a sheet of cells — handy for comparisons, portfolios and storyboards. Pick the number of cells; rows and columns follow the aspect ratio. Long-press an image to pick it up and swap positions, remove one from the corner of its cell, and anything that no longer fits when the grid shrinks goes back to the Flow. The background can be Light / Dark / Auto, and each cell can carry a number. When it looks right, hit "Stitch (2K)" and the whole sheet — gaps included — becomes a single 2K image in your asset library. Images inside a grid sit out of the workflow, so their existing connections are cut; you are asked once before that happens.

Improvements

  • Server-side composition is cheaper: 1 credit per 5 seconds instead of 1 credit per second, rounded up to the next 5 seconds.
  • If GPU encoding fails while rendering a composition, the browser falls back to software encoding automatically — no more restarting the browser. When it genuinely cannot render, you get "Reload / Hand to server" on the spot; choosing the server hands off just that node and the rest of the run carries on.
  • Video upscaling with DetailX has a new "Light correction" toggle (off by default), on the canvas toolbar and on the image & video page; hover it to compare on and off.
  • Video upscaling can now skip frame interpolation, and is priced for what it actually uses.
  • The Flows list keeps loading as you scroll to the bottom — no more paging.
  • Switching models in a Flow keeps the settings you already dialled in, and only folds away what the new model cannot take.
  • Grid split and image annotation re-run with the workflow, so downstream nodes no longer pick up the previous result.
  • You now get a notification when credits are refunded.
  • The default language model in a Flow is now Gemini 3.8 Flash; 3.5 Flash has been retired.
  • The estimate before a run is down to one line: what it costs and how long it may take.

Fixes

  • Fixed the selection box jumping when you drag past the edge of the canvas: scrollable boxes such as text and language-model output no longer drag the selection away with them, and the further out you go the faster it scrolls.
  • Ctrl+A in a Flow now selects all nodes, and "Disconnect" in the right-click menu applies to every selected node instead of only the one under the cursor.
  • Three @ mention fixes: a node name starting with @ is no longer read as a name, an unbound @name shows as a clickable grey chip, and the candidate list redraws on the spot when it changes while you type.
  • Fixed canvas video missing its 480P tier, which turned a 9:16 choice into 16:9.
  • Fixed saving a Flow wiping out a language-model result the server had just written back.
  • Fixed the grid group's "Adjust" button and the grid split divider lines flipping to light colours in dark mode.
FlowWall-1.0.8

Highlights

  • "Settings" and "Help" join the canvas. They sit at the bottom left, and canvas preferences now live in one place — appearance, file names, snap, hide edges and completion notifications, one switch per row. Help holds the keyboard shortcuts, the tutorial and the Discord community. Shortcuts moved out of a dialog into a large panel above the toolbar with three tabs — Keyboard & mouse / Trackpad / Video composition — and key caps that show Ctrl or ⌘ to match your system. Both panels open on phones as well.
  • Uploads can keep their file names. An uploaded image or video can carry its file name as the node name: turn on "File names" in settings and every node you create — picking from the asset library, dragging a file in, or pasting a link — is named after the original file; leave it off and you get "Image 1, Image 2" as before. The toast right after adding can switch just that batch between file names and numbers, and a node's "More" menu can switch at any time — the @ mentions in downstream prompts are rewritten along with it. Screenshots, AI results and anything else without a trustworthy file name stay numbered.
  • Dragging snaps to the grid. As you drag, a node snaps to the background dots, and a multi-selection keeps its formation. "Snap" is on by default and can be turned off in settings.

Improvements

  • When a video finishes while you are on another tab, you are asked once whether to turn on browser notifications; permission is requested only after you click "Turn on", and you are never asked again.
  • The "Select / Move" mode buttons are gone from the canvas toolbar — the V / H shortcuts still work; the comment tool only appears in a shared Flow.
  • Renaming a node by double-clicking its title now offers the original file name as a button: one click fills it in.
  • The transparent background on image nodes is now an icon toggle.
  • Menus on phones open and close on tap instead of hover, and submenus expand in place rather than flying out to the right.

Fixes

  • Fixed short subtitle lines being moved wholesale into someone else's speaking window; they now land where they were actually spoken.
  • Fixed lines carrying a voice effect being dropped and pushed five seconds late, and drawn-out closing lines being squeezed into the previous segment, which left the last line without subtitles.
  • "Naming" no longer shows up on nodes that have no original file name, such as AI results and exports; moving the mouse across a submenu no longer closes it.
FlowWall-1.0.7

Highlights

  • Image2.5 Flare and Sunburst are here. OpenAI's new generation of image models, priced the same as Image2. Flare looks better and renders faster, so it fits everyday high-volume work; Sunburst follows editing instructions more precisely, for campaign visuals and polished product shots. Both are on the image page and on image nodes in a Flow.
  • New "Grid split" node. Cut one picture into cells and keep only the ones you want. Connect an image, pick 4 / 9 / 16 / 25 cells or set the rows and columns yourself (up to 8×8), drag any dividing line to move it, and click a cell to drop it. Hit Export and the remaining cells become real asset images — saved to your library, reusable anywhere — placed as a group to the right of the node.
  • Draw in the "Image annotation" node. It no longer needs a picture connected. Click "Add background" in the middle of the node to lay down a blank canvas and draw straight away: eight background colours, or a transparent one (the export keeps the alpha channel). Seven aspect ratios from 16:9 to 9:16 change the card and the output size together. Colour and stroke width now share one panel, with a dot drawn at the current setting.
  • Composition and video editing render on your own computer. No more waiting in the server queue: anything up to 2K renders right in the browser and costs no credits. For 4K, very long timelines or a machine that cannot keep up, a dialog lets you choose — stay local, or hand it to the server (server rendering is billed by the length of the result, 1 credit per second, rounded up).
  • New "Pick reference" mode. Click "+ Reference" at the bottom of a node and every node that can serve as a reference lights up in blue; one click connects it, with no wire to drag, and anything that cannot connect explains why. Connection rules are now uniform: once a reference slot is full the new link is refused rather than replacing the old one, so swapping a first frame or a reference image means disconnecting the old one first.

Improvements

  • The asset dialog supports multi-select: turn it on and clicking a card ticks it, an action bar rises from the bottom, and "Add (N)" brings them all in at once; hold Shift to select a run of cards. A batch added to a Flow is laid out in a grid instead of stacking on one spot.
  • Uploaded image nodes and finished image generations both gained an "Edit" button: it opens crop / colour / annotation / layer split / grid split, and the node is created and wired up on the right for you.
  • The top of the home page is now a banner track you can page through, and an Image 2.5 showcase was added below the ad templates.
  • "Recent Flow projects" on the home page refreshes itself when you come back from a Flow or another tab — no full page reload.
  • Composition node: drag a whole group of images onto the timeline at once, change the duration of several image clips together, click empty space to deselect, and image layers gained reset and drag snapping.
  • Editor exports now show a "Preparing assets" step, and the exported result node carries a version bar.
  • Runs of blank lines pasted in are collapsed into one — now in the create dialog, voiceover, music and voice design as well as text nodes in a Flow.
  • Buttons release focus after a click, so Space and similar shortcuts are no longer swallowed by a toolbar button.

Fixes

  • Fixed the last few lines of subtitles being dropped.
  • Subtitle recognition now switches to another route when one is unreachable instead of failing outright.
  • Fixed Delete on a node group leaving the nodes inside it behind.
  • Fixed reference images being trimmed down to one after switching models on an LLM node.
  • Fixed rendering and exporting stalling for half a minute while the browser tab sat in the background.
  • Choosing "Exit anyway" when leaving a Flow now really stops local rendering and exports.
FlowWall-1.0.6

Highlights

  • GPT-6 Astra is here. OpenAI's new top tier: steadier and more detailed on long-form writing, hard reasoning and reading an image to write about it. Pick it on any LLM node in a Flow; Fan Theater scripts and "Reverse image prompt" now use it too.
  • LLM nodes write as they think. Run one and it shows "Thinking" first, then the text arrives line by line instead of appearing all at once at the end. While it works you can use the version bar to look back at earlier takes, then return to the one being written and keep following along.
  • LLM nodes follow the prompt's language. With no skill and no quick prompt attached, the reply follows the language you wrote in rather than the interface language. Long answers from the Claude models — a full storyboard written in one go, for instance — are no longer cut off halfway.
  • Claude Sonnet 5 added. Claude Sonnet 4.6, GPT-5.5 and Gemini 3.5 Flash Lite are retired. An older Flow that used them switches to the successor in the same family when you open it — Claude Sonnet 5, GPT-5.6 Sol and Gemini 3.7 Flash respectively — with nothing to change by hand and no "model not found".
  • New model list for Fan Theater. GPT-6 Astra, Claude Sonnet 5 and Claude Fable 5 — every one on the list really runs.

Improvements

  • The model dropdown on LLM nodes is grouped Gemini → GPT → Claude, with the newer ones first inside each group.
  • Switching models keeps the node skill you already picked, so there is no need to choose it again.
  • The "New version available" prompt was redesigned: it lists the highlights of this update and keeps a link to the changelog at hand.
  • The tutorial page gained a chapter path, so you always know which chapter you are reading and can jump back to the top in one click.
  • Template cards on the Flows page drop the "generated N times" line and keep just the title.

Fixes

  • Fixed LLM nodes where picking GPT-5.6 Sol / Terra or Claude Fable 5 / Opus 5 did nothing and quietly fell back to the default model.
  • Fixed the Claude models reporting invalid parameters as soon as Temperature or a similar setting had been touched.
  • Fixed the Claude models failing outright when a reference image was connected.
  • Fixed Fan Theater drafts saved with a retired model refusing to open and reporting the model as unavailable.
  • Fixed LLM nodes that had already produced their answer still showing as running.
  • Gemini 3.7 / 3.8 Flash now retries automatically after an occasional failure instead of giving up on the first try.
FlowWall-1.0.5

Highlights

  • New "SVG animation" node. Describe the animation you want in one line and the node builds it on the spot and plays it back; you can @ a connected image as a reference. Keep asking for changes to refine it, or run it with an empty box for a fresh take — every take stays in the version bar, so you can switch back and compare at any time. Text inside the animation can be edited directly and updates instantly, with no re-generation. When it looks right, export it as a transparent video in one click; it lands on the video node to its right, ready to build on, and costs no extra credits (Chrome or Edge for now).
  • Export added to composition, video-editing and SVG animation nodes. The exported cut becomes a new video node, so you can keep building from it instead of downloading and re-uploading.
  • Public templates run straight away. Hit "Generate" on a template in Discover and it starts working while being filed under Templates in My Flows. The dialog shows live per-node progress and every earlier result, nothing is lost if you close it, and you can pick up where you left off. Templates you no longer want can be removed, optionally along with what they produced.
  • Color schemes. Settings gains "Appearance and color", with several ready-made schemes (a primary, an accent and a matching gradient) that work in both light and dark mode and change the whole site at once.
  • New models for text nodes. GPT-5.6 Sol, GPT-5.6 Terra, Claude Fable 5 and Claude Opus 5.

Improvements

  • Hovering a template card shows the reference images it uses; open it to preview the whole Flow read-only, and save it as your own template if it fits.
  • The login page separates email sign-in from the social options, and the showcase beside it now switches by clip length, playing each one through to the end.
  • Failure messages are more accurate: a whole class of cases that used to read "the upstream service is unavailable, please try again later" now tells you which asset or setting is the problem, so there is no need to keep retrying blindly.
  • Opening an older Flow gives names to nodes that prompts referenced by position, so existing @ references keep working untouched.
  • Building a Flow through an AI assistant over MCP no longer stacks new nodes on top of each other, and template and subject nodes can now be created directly.

Fixes

  • Fixed generations occasionally getting stuck on "generating" forever; interrupted work now ends promptly and refunds the credits.
  • A Flow run whose first dispatch fails now ends immediately instead of sitting there for a long time.
  • Fixed template runs whose progress stopped moving; they now finish on their own even if the page misses an update.
  • Stuck or timed-out generations reach a conclusion faster, and internal wording no longer shows up on screen.
FlowWall-1.0.4

Highlights

  • One-click prompt translation. Image, video and text nodes gain a translate button beside Run: turn the prompt into your interface language with one click, click again to undo. While undone, hover the button to compare with the original. @ references and audio tags are kept verbatim so nothing breaks, and a prompt already in the target language is flagged instead of quietly spending credits.
  • @ references now use node names. Write @node name in a prompt to point at exactly the asset you mean; rename the node and every reference follows. Text can be substituted in place, and several text nodes can feed the same prompt.
  • MCP access does more. AI assistants can now bring assets straight into a canvas (by URL or from your device), build first/last-frame loop videos, and draw on a built-in body of production know-how, so what they make comes closer to the work of someone who knows the tools.
  • New node skill: Storyboard Script Spec. Turns storyboard material in any format into a standard, ready-to-render prompt, binding the element names in it to reference images of the same name on the canvas.
  • Video model updates. Gemini Omni 1.1 Flash is live, with better physics and shot-to-shot consistency plus image and video references; MiniMax H3 adds 2K output.
  • Membership perk change. The discounted-model perk moves from Nano Banana Lite to MiniMax H3 (20% off); the Seedance discount is unchanged.

Improvements

  • Hovering a video node brings up a playback bar: scrub the timeline, adjust the volume and capture the current frame as an image. Clicking the picture plays or pauses it.
  • Uploading and replacing assets on the canvas now opens the asset picker, so you can reuse what you already have. A replacement keeps the node's name, so existing @ references stay intact.
  • With a node selected, hovering a connected reference thumbnail — or an @ reference in the prompt — floats an enlarged preview, and videos play right there.
  • The canvas toolbar is now one "+" menu plus four fixed buttons, with every node type collected under "+", which opens on hover.
  • The sidebar scrolls as a single area; when the window is too short or the bar is collapsed to icons, the docs and social links fold into pop-up menus and the horizontal scrollbar is gone.
  • The three cards at the top of the home page now play short clips, falling back to a full cover image on slow connections and pausing when scrolled away or the tab loses focus.
  • Extending or editing a single video is billed as a fixed 10 seconds, so the estimate shown before you submit matches what you are charged.

Fixes

  • Fixed occasional failures in prompt translation and subtitle processing.
  • Fixed reference images being dropped by Seedance 2 Mini.
  • Video references beyond Gemini Omni's duration limit are now flagged up front instead of failing mid-generation.
  • Fixed occasional navigation glitches when opening a canvas from a home page banner.
  • Fixed parameters displaying incorrectly after switching models in Wallpaper Studio.
  • Failed generations now show a more specific reason, making it clearer whether to retry or change the settings.
FlowWall-1.0.3

Highlights

  • A brand-new home page. Fresh top banners, "Create with Flows" feature cards, a featured Flows showcase and the CPP Creator Partner Program gallery — play standout works, see how they were made and apply to join; your recent Flow projects greet you on arrival, and the creation box is more compact.
  • MCP access is now open to everyone (Beta). Connect FlowWall to MCP-enabled AI assistants such as Claude and ChatGPT, then create canvases, wire up node pipelines and generate images, video and audio just by chatting. The entry point is in the left navigation.
  • Subtitle proofreading upgraded. Drag subtitle blocks freely along the timeline, split a line or insert a new one between two, and right-click to edit or delete; the node now shows the audio waveform, colored as playback progresses.
  • The canvas remembers node settings. Newly added nodes reuse the model and parameters from last time, so there is no need to pick them again.

Improvements

  • The text node supports basic formatting: headings, bold, italics, lists and a light backing panel.
  • The speech node on the canvas gains speed, pitch and volume controls.
  • After rendering, the compositing node can pick any result from its history as the output — going back to editing no longer interrupts downstream nodes.
  • Subscription credits are shown more clearly, with balance and total allowance at a glance.
  • New empty canvases guide you to the upload button, and the sidebar gains official community links.

Fixes

  • Fixed shared canvases exiting unexpectedly when closing an asset preview, and copies losing access to shared assets.
  • Fixed interface language switches occasionally not taking effect until a manual refresh.
  • Fixed preview seams flickering, playhead ghosting, and audio being deleted along with a video clip in the compositing node.
  • Fixed the subtitle node losing its audio waveform after re-entering the canvas.
FlowWall-1.0.2

Highlights

  • The video compositing node now includes editing. Arrange multiple video and audio tracks on the canvas, trim, split, change speed, detach audio, add fades, preview instantly, and view both edit and rendered histories.
  • The subtitle node is now available. Pass in video or audio to recognize and proofread subtitles line by line, translate while preserving timing, create bilingual captions, highlight keywords, and choose subtitle styles.
  • The layer-splitting node is now available. Seedream 5.0 Pro splits an image into a base image and transparent layers for further editing and composition.
  • New video models: Seedance 2.0 Mini and MiniMax H3, available for video generation on the generation page and canvas.

Improvements

  • Added light, dark, and follow-system appearance modes, with readable canvas, asset, dialog, and changelog views in every theme.
  • Asset multi-select now supports selecting whole date groups, swipe selection on touch, and moving items to a folder or creating one in place.
  • Canvas editing is quicker with multi-select, timeline zoom shortcuts, clip speed controls, and spacebar play/pause for the selected node.
  • Added multilingual tutorials and documentation; empty projects now offer visual guidance to create a desktop or phone canvas directly.

Fixes

  • Fixed shared canvases that occasionally failed to open, play media, or create a copy.
  • Fixed transparent images and videos displaying incorrectly in the asset library and canvas.
  • Fixed low contrast for some canvas controls, waveforms, and notices in dark mode.
  • Fixed inaccurate task states, timeline clips, and error messages after a generation failure.
FlowWall-1.0.1

Highlights

  • Music generation is live. The audio page gains a Music tab powered by MiniMax music-3.0, with text-to-music and covers. Upload a reference track to cover it; leave the lyrics empty and the original ones are used.
  • Music and speech return as soon as you submit — carry on with something else instead of waiting on the page.
  • Crop is now a real crop. Once applied, the asset itself is the cropped result, and upscaling or AI editing afterwards works on it.
  • Vietnamese added — 12 interface languages in total.

Improvements

  • Asset multi-select reworked: the selected state is clearer and the desktop grid is more compact.
  • Multi-select supports swipe-to-select on touch, and a new Move to action puts assets into a folder — you can create the folder right there.
  • The filter drawer can be swiped away to the right.
  • Audio assets show a waveform thumbnail with their duration, and a new music detail card lets you play and scrub, in both light and dark themes.

Fixes

  • Music prompt and lyric limits are now tiered per model, and going over is caught before any credits are charged.
  • Fixed failed audio generations sitting in "generating" forever.
  • Fixed audio generations spinning indefinitely after a failure.
  • Audio history no longer shows failed records by default, matching images and video.
  • Error messages no longer carry model names or internal details.
FlowWall-1.0.0

Highlights

  • The MiniMax model family is live: H3 for video, speech-2.8 for speech and music-3.0 for music — all three available at once.
  • Wallpaper Studio editor rebuilt: the canvas is locked to the device form and composition is the crop; markup folded into AI editing, and erasing works either by description or by painting over the area.

Improvements

  • Audio page reworked: the voice library filters by language and gender, remembers your last choice, and gains a recommended section plus a full-screen editor.
  • Portrait-phone support: the sidebar opens with a swipe, and the upload, assets, audio, theatre and expand pages were re-laid out.
  • Faster first paint: the splash screen only fades once the page is actually visible, and a blank screen can be retried.
  • First and last frame became two independent toggles, and reference images and video are picked the same way.
  • Clearing the prompt box now clears the reference images with it.

Fixes

  • Fixed a crash on iOS 16.0–16.3 when opening the canvas page.
  • Fixed two blank-page issues.
  • Fixed seven phone keyboard problems.
  • Fixed specific error messages being replaced by the generic one; failed canvas nodes show the real reason again.
  • Fixed occasional generations failing for no reason.
  • Voice cloning now asks for a different minimum audio length per model.