What's New in Visuals
There is a new default video model, and you can now keep a scene going instead of starting a new one.
Gemini Omni 1.1 Flash arrived last week and is now what a new video card starts on. The bigger change is what it lets you do: take a clip you already made and continue it, up to 40 seconds, as one growing file rather than a set of separate shots you have to stitch together.
The short version:
Gemini Omni 1.1 Flash is the new default video model
Extend a clip up to 40 seconds as one continuous file
Pick 360p draft, 720p, 1080p or 4K right on the card
Trim any clip on a timeline before you use it
Long uploads are usable and transcribe while they are still processing
Images download instead of opening in a new tab
A new default video model
Gemini Omni 1.1 Flash is now what a new video card starts on. The model you were using before is still there under a clearer name, Gemini Omni 1.0 Flash, it is just no longer the default anywhere.
What it brings:
4 resolution tiers on the card. 360p for drafts, then 720p, 1080p and 4K. The chip on the card shows the resolution and the menu spells out the tier, so a rough pass costs you a rough pass.
Reference videos. Connect up to 3 clips to carry a character or a motion style into what you generate.
First and last frame. Pin where a clip starts and where it ends.
One limit worth knowing up front: the model cannot currently take a video you uploaded yourself as a reference. That is a fault on the model provider's side, not ours, and the card says so plainly if you try. Extending a clip you generated in Visuals is unaffected and works normally.
Extend a scene up to 40 seconds
This is the part to try.
Before, a longer sequence meant generating separate shots and joining them together, and the joins looked like joins.
Now you can take a finished clip and continue it. Put a card in Extend mode, connect the clip, and it comes back as one grown file rather than a second clip you have to attach. Do it again and it keeps growing, up to 40 seconds in total.
You can start an extension from anywhere the clip already lives:
Right click a finished card.
The Extend Video button in the full screen preview.
Drag from a video card's output.
The card menu, like any other card.
A few rules the app handles for you:
An extension stays at the quality of the clip it continues. Changing tier partway through would make the model return nothing new, so the quality control pins to "Matched to source" and the other tiers switch off.
Additions fit under the 40 second ceiling. Ask for 10 more seconds on a 33 second clip and you get 7, with a note saying so.
A chain survives a reload. You can close the tab and pick the same clip back up later.
You only pay for new footage. A fresh clip bills for what it produced. An extension bills for the seconds it added, not for the whole file.
One note: 4K clips currently download at 1080p. The full resolution file is there, the download path is what caps it, and that is being fixed separately.
Trim any clip before you use it
Every video card now has a Trim control, on any model, not just the new one.
It opens an editor-style timeline with a ruler and draggable edges, and the selection loops while you set it so you can hear and see exactly what you picked. Press Done and the trim is what the run uses.
Long videos are usable while they are still processing
A long upload used to be a dead end. A 2 hour podcast could sit on "Processing Upload" for more than half an hour, could not be dragged onto the canvas even though it played fine, and never produced a transcript at all.
Four separate things were wrong. All four are fixed.
A video you can play is a video you can use. Whether a clip plays and whether it still has work running are now two separate questions. If it plays, you can drag it, click it and build with it. The badge stays and names what is actually still running, so you know the transcript is not ready yet.
Transcripts arrive much sooner. Transcription no longer waits on the largest version of your file, which on a long video arrived about 22 minutes after the audio did. On uploads under the size ceiling it now starts from your own file while the video is still being processed, so the wait is roughly the upload rather than the encode.
A video too long to analyse is skipped, not failed. Content analysis has a 2 hour ceiling, which was written down nowhere. A longer video used to be marked as a red failure after half an hour. Now it is settled in seconds and marked as skipped. The upload worked, the video plays, the transcript is unaffected.
One notification, not two. Uploading raised a toast and a progress panel saying the same thing from the first byte to the last. The panel stays, since it is the one that knows when the file is genuinely ready.
Anything that needs a transcript now checks for one first. Cuts used to read the transcript to choose its clips without ever checking that a transcript existed, so it would run and hand you a result built from nothing. It now tells you whether the transcript is still coming, failed, or was never going to exist.
Also fixed
Images download instead of opening in a new tab. Pressing Download on an image navigated to the file and displayed it, leaving you to save it by hand. It was happening across My Library, the visual canvas, the create panel and share pages. Every download button in the app now runs through the same path and saves the file.
All outputs is there as soon as you have 2 versions. The right click Download and Extract asset menus only offered the all-versions choice once you had 3 or more, and only after a later generation refreshed the page. The count now matches what is actually on the card.
Plus assorted polish on menus, previews and status labels.