Gemini Omni Flash 1.1
by Google DeepMind
Video with sound made in the same pass, from a frame, a prompt or a reference.
What it is
Google describes this as its multimodal video model: it animates a still image into video "with synchronized audio, extending a single frame into coherent motion that reflects the logic of the real world".
The reference ingress is the interesting one. It reasons "across all inputs to produce a single coherent result, with characters retaining their face, clothing, and voice throughout", and it takes images and clips together.
Three ingresses on one model: from a prompt, from a still, or from references. Which one runs is decided by what you attach.
What it is good at
The audio is made with the picture rather than laid over it afterwards, so a footstep lands where the foot lands.
Face, clothing and voice are held across the shot, which is the part that usually breaks when you change scene.
It runs from 360p upward, and the bottom of that range is cheap enough to test an idea properly before you commit to a long one.
Where it struggles
- Google publishes no list of limitations for this model, so none is invented here.
- Three to ten seconds per generation. For longer, build a script so the shots belong together.
- The reference ingress takes images and clips, not audio. If the sound has to come from a clip you supply, this is not the one.
How to prompt it
Say what happens and what it sounds like. Because audio is generated with the picture, naming the sound is part of describing the shot rather than an afterthought.
An example that shows what that means
A woman closes a heavy wooden door behind her and leans against it, the latch clicking and the room going quiet, late afternoon light across the floorboards.
What we offer here
Straight from the catalogue, so this list never goes stale. The price is the provider’s published rate with our markup on top, and it is on the button before you run anything.
Questions
Does it make its own audio?
Yes, in the same pass as the picture. That is the main reason to pick it over a model where you supply a track afterwards.
Can I keep the same person across shots?
Through the reference ingress, which holds face, clothing and voice. Attach a reference and the studio moves you to that variant.
Do I need a starting image?
No. There is a text ingress as well, and the studio picks the one that matches what you attached.
Next
Sources
Try Gemini Omni Flash 1.1 in the studio
The studio opens in stages. We let people in in the order they signed up, and you start with 1,000 free credits.
Get early access