Gemini Omni Flash
Unless instructed otherwise, Gemini Omni Flash generates video split into multiple cuts. State in the prompt whether to keep one continuous scene or at which second to cut, and point to connected images in your sentences with tags.
Specify scenes and timing
Gemini Omni Flash builds a narrative from the prompt, so by default it generates video split into multiple cuts.
Keep it as one continuous scene
For video that continues as one scene with no cuts, put one of the expressions below in the prompt.
In a single unbroken scene / In a single continuous shot / No scene cuts
Split cuts by time
You can also set times to split the cuts yourself. There is no fixed syntax; write them as plain sentences or as timecodes.
After 3 seconds, a woman enters the scene. At 5s the chorus starts in the background audio. Every 2s cut to a new frame.
[0-3s] A person is walking [3-6s] They stop and turn around [6-10s] They start running
- Split timecodes within the length set in the video node. The length can be set to up to 10 seconds.
- Audio is generated automatically from the prompt, so you can write when a sound plays in the same way.
You can check the supported inputs, length, ratio, and resolution in Video models.
Point to images with tags
An image's role is set by which video node input it is connected to, and in the prompt you point to that image with a tag.
An image connected to the Start frame input becomes the video's first frame, and an image connected to the Reference image input is used only as a reference, not as the first frame. The two inputs can be used together, and you can connect up to 6 reference images.
| Tag | Image it points to |
|---|---|
<FIRST_FRAME> | The image connected to the start frame input |
<IMAGE_REF_N> | An image connected to the reference image input. Images are numbered from 0 in the order connected, and the start frame is not counted |
If you connected 6 reference images, use <IMAGE_REF_0> through <IMAGE_REF_5>.
If you connected a start frame, point to it with <FIRST_FRAME>.
<FIRST_FRAME> a woman is walking
If you connected 2 reference images, point to them with <IMAGE_REF_0> and <IMAGE_REF_1>, in the order connected.
in the style of <IMAGE_REF_0> a woman <IMAGE_REF_1> is walking
Combine tags with timing
Tags can be mixed with timing. The example below is a prompt with 4 reference images connected.
[0-3s] A studio fashion sequence. Starting with woman <IMAGE_REF_0>, she is holding <IMAGE_REF_1> [3-6s] Then we see the man <IMAGE_REF_2> holding <IMAGE_REF_3>
Declarations Klipse adds automatically
When you connect images, Klipse automatically adds a declaration of each image's role to the start of the prompt, and a sentence telling the model to use the images in those roles to the end. Even with many images and complex roles, write only the body with tags in your prompt.
| Connected input | Declaration added at the start | Instruction added at the end |
|---|---|---|
| Start frame | [# Sources <FIRST_FRAME>@Image1] | A sentence telling the model to use the first image as the starting frame |
| Reference image | One entry per connected image, as in [# References <IMAGE_REF_0>@Image2 <IMAGE_REF_1>@Image3]. Without a start frame, counting starts at @Image1 | A sentence telling the model to use the reference images as references, not as the first frame |
For example, connect 1 start frame and 1 reference image, and write the following.
a woman <IMAGE_REF_0> is walking.
The model receives a prompt that begins with [# Sources <FIRST_FRAME>@Image1] [# References <IMAGE_REF_0>@Image2] and ends with instructions to use the two images as the start frame and as a reference, respectively.
Do not write the declarations or instruction sentences yourself.
If you write a [# Sources …] or [# References …] declaration, or an instruction sentence such as Use Image1 as the starting frame., in the prompt, the same content is sent twice. If the numbers you write do not match the actual connections, one image can be sent with two different roles. Set each image's role by how you connect it, and put only tags in the prompt.
What to check when using tags
- Use a tag only when an image is connected to that input. If you write
<FIRST_FRAME>without connecting a start frame, or an<IMAGE_REF_N>numbered beyond the images you connected, there is no image for the tag to point to. - In your sentences, it is better to point to images with tags than to refer to them by position, as in
Image2. The@Imagenumbers in the declarations change depending on whether a start frame is connected, but<IMAGE_REF_N>counts reference images only, so it does not change. - A reference image that you disconnect and connect again takes the last number.
- You can also use the
<<<image_1>>>notation from Kling Omni. These numbers count from 1 and are converted to numbers starting at<IMAGE_REF_0>, so<<<image_1>>>points to the first reference image. - When you use AI Edit with a video input, images cannot be used with it, so write only what to change, without tags.