This one is about the visual canvas.

It now tells you what every card is doing while it works, puts new cards exactly where you asked for them, and keeps a full record of every version you generate, prompt and settings included. Everything you make there is easier to get out, too.

The short version:

This was the single most requested change, and it is the one to read first.

Before, a card saved every version you generated, but clicking through them only swapped the picture. The prompt, the model, the duration and the aspect ratio all kept showing your most recent attempt. If you generated 5 versions of a shot, you had no way to tell what made any of them.

Now, clicking a version restores everything about it: the prompt, the model, the duration, the aspect ratio, quality, the audio and loop toggles, and every model specific setting. Click the newest version to get back to where you were.

More on the card itself:

One thing worth saying plainly, because it trips people up: image references work on the visual canvas and on the image and video generator pages. Video references, and audio, work only on the visual canvas.

Before, you pressed Run and every card in the flow started spinning at once. A card waiting 3 steps back looked exactly like a card that was genuinely being made, and there was no sense of how far along anything was.

Now every card says which of 5 things it is doing:

A progress ring fills as the card works, measured against how long that model usually takes at those settings. One honest note: that is a time estimate, not a live reading from the model. It holds near the end rather than claiming to be finished, and only completes when the result actually lands.

Credits at a glance. A ring around your profile picture shows how much of your current period's credits are left. There is deliberately no number on it, because a running total sitting on a creative surface is its own kind of stress. Click your picture for the exact figure, the same as before.

Three complaints, one fix.

Before, adding a card dropped it at the far right end of your flow no matter where you had clicked, and then the canvas zoomed onto it and threw away the view you were working in.

Now it depends on how you added it:

And the canvas holds still. Adding a card never re-frames your view again. New cards also arrive already selected, so you can start working on one without hunting for it.

Everything defaults to 16:9 now, images and video both. Video used to default to portrait, so every single card needed correcting before you could use it.

Cards take the shape of what you feed them. Connect a piece of media into a card and the card inherits that media's aspect ratio, snapped to the nearest shape the model can actually produce. Drop a file onto the canvas and it is measured before it even uploads, so its card is the right shape in the first second rather than correcting itself later.

Each flow remembers your last choice. Pick a model, ratio, duration, quality or audio setting by hand and the next blank card in that flow starts there. It only learns from choices you actually made, never from guessing at what is already on the board, and it never carries across to a different flow. It tells you when it has done it, with a label like "16:9 matching this flow", and there is a Reset that drops back to the model's own defaults. A preset that declares its own shape always wins, so picking Movie Poster inside a 16:9 flow still gives you a poster.

Also on the canvas:

Before, dropping a video onto the canvas showed nothing at all until the upload finished, which on a large file is minutes of staring at an empty board.

Now the card appears immediately, previewing straight off your own disk. You can see it, move it and connect it while the file goes up. When it is ready the card swaps to the real asset in place, so nothing you did in the meantime is lost. If it fails, the card is removed and the message names the file.

Run waits for it. The Run button stays off while anything is still uploading or processing, and it tells you which file it is waiting on. It used to come back on in the gap between the upload finishing and the video actually being playable, so a run started there reached a card with no video in it.

Uploaded videos no longer read as unavailable. This was a real bug. A video would upload into My Library perfectly, then show as unavailable on the canvas, and refreshing did not help. The only workaround was to replace it with the same asset from My Library. It is fixed, and the canvas now quietly retries on its own while a video finishes processing instead of sitting on a broken preview until you reload the page.

One notification instead of two. Uploading a video used to raise a toast and a progress panel saying the same thing. The panel stays, because it is the one that knows when the video is genuinely usable.

Before, what you could learn about an asset depended entirely on where you opened it. A card on the canvas knew its own settings. The same asset opened from My Library showed almost nothing.

Now every surface reads the same record, so a card on the canvas, an asset in My Library and the library sidebar all describe an asset identically: model, aspect ratio, dimensions, duration, quality, audio and loop, then location, size, source, visibility, release and created.

References follow the asset into My Library. On the canvas you can see what fed a card, because the connections are right in front of you. In My Library there are no connections, so an asset that plainly started from a frame looked like it came from nowhere. Generations now record what went into them and that record travels with the asset. Most of your existing work is covered too, not just new generations.

The references section has 3 states rather than 2. It can tell you a generation used nothing, or it can tell you there is no record for that asset. Saying the first when the truth is the second would be a lie, so it does not.

Reference videos play. Click a reference and it opens. If it is a clip it plays, and whatever was playing behind it pauses while you watch and resumes when you close it.

Pick a frame from a generated video. The frame scrubber that asset cards have now works on generation cards too. Scrub, click, and that frame becomes the card thumbnail and the frame fed to anything connected to that output. There is also Add frame to canvas, for when you want the video to keep feeding its own last frame while a still branches off somewhere else. It used to mean waiting for the generation, going to My Library, dragging the asset back and only then picking a frame.

ProRes masters run to several gigabytes, and every download path except one was pulling the whole file into your browser first. So the same file that downloaded fine from the sidebar died from the bulk bar, from both menus and from Save all. Only .mov files were affected, because MP4s are small enough never to hit the ceiling.

There is now one download engine behind every download button in the app, and files stream instead of piling up in memory.

Two limits worth knowing. A ZIP holds up to 200 assets. Any single file over roughly 256MB downloads on its own alongside the ZIP instead of going inside it. Both tell you clearly rather than handing you a broken archive.

Download is where you expect it now.

Extract asset. Right click a result on a generation card and place it on the canvas as its own asset card, leaving the generation card where it is. It works per version, which is the whole point: branch off the take you liked without disturbing the card that made it. There is a second option to extract every result at once, stacked and numbered.

Multi select now works in every full screen preview in the app: My Library, the visual canvas, generation results, Explore and the video editor lightboxes.

Whatever you have selected downloads as one ZIP. The action bar sits at the bottom of the browser window, and the preview no longer resizes the moment you start selecting.

Video previews use real controls now. A timeline you can actually scrub, seeking, elapsed and total time, volume and fullscreen, everywhere a video opens.

If you connect Visuals to a chat app, the Visuals panel there now takes on that app's own look instead of fighting it: its type, its surfaces, its buttons.