Kling Omni vs Gemini Omni: two different models with the same name
"Omni" now means two different things in AI video. Kling Omni is Kuaishou's reference-video line, whose newest version is Kling 4.0 Omni Reference. Gemini Omni is Google's model family, whose current video model is Gemini Omni 1.1 Flash. Both can build a new scene around reference images of a person or product. Gemini Omni costs less per second and can edit video. Kling 4.0 takes far more references and makes longer single shots, and its API opened on fal.ai on October 23, 2026. Below is the comparison, with sources, as of October 23, 2026.
4 reference photos
New scenes, same face
What each "Omni" is
Kling Omni is a mode family from Kuaishou's Kling. The Omni models take several inputs at once, such as images, short videos and, in Kling 4.0, voice profiles, and treat them as a description of who or what must appear. Kling 3.0 Omni is available through Kling's API and fal.ai. Kling 4.0, with Omni Reference, became available in Kling's own app on October 15, 2026, and through fal.ai's API on October 23.
Gemini Omni is a Google model family announced in May 2026, starting with video. Gemini Omni Flash came first; Gemini Omni 1.1 Flash followed on August 27, 2026. It runs in the Gemini app, Google Flow, YouTube Shorts and the Gemini API, and it has replaced Veo 3.1 in the Gemini app and the Gemini API. It generates from text, images and video, and it is built to edit video as well.
Kling Omni vs Gemini Omni: the numbers
The two differ most in how much you can give them and how long a single generation runs. Kling 4.0 Omni Reference takes up to 15 inputs and makes up to 30 seconds in one go. Gemini Omni 1.1 Flash takes up to 10 reference images and makes up to 10 seconds, which you can extend in 10-second steps up to 40 seconds.
| Compared on | Kling 4.0 Omni Reference | Gemini Omni 1.1 Flash |
|---|---|---|
| Reference images | Up to 10 | Up to 10 in the API, 5 in the Gemini app |
| Reference videos | Up to 5, 30 seconds combined | Up to 3, short clips |
| Voice references | Up to 3 voice profiles | None |
| Inputs per task | Up to 15 | Images and videos as above |
| One generation | 3 to 30 seconds | 3 to 10 seconds, extendable to 40 |
| Resolution | 720p, 1080p, 4K | 360p to 4K (1080p and 4K upscaled) |
| Aspect ratios | 21:9, 16:9, 1:1, 9:16 | 16:9, 9:16 |
| Sound | Native stereo with lip sync | Native audio |
| API today | Yes: fal.ai since October 23, 2026; Kling's own API in early access | Yes: Gemini API, Google Cloud, fal.ai |
| API price at 720p | $0.157/s on fal.ai ($0.14/s Kling preview) | About $0.10/s (fal.ai $0.10/s) |
Keeping one character or product consistent
Both read references as identity, not as a first frame, so the subject can appear in scenes and angles the photos do not show. In Gemini Omni you tag each image in the prompt; in Kling you group images of one subject into an element and name it.
Kling 4.0 has more room for a cast. Up to 7 elements, each with its own images, plus voice profiles, means a person, a second person and a product can each keep their look, and the people their voices, in one shot. Gemini Omni's ten images are shared across everything in the prompt, and Google's documentation says it does not edit videos of recognizable people.
For a single subject in a short shot, the difference is smaller. Three to five clean images, a front view, a three-quarter view and a detail, work for both.
Mia: 4 photos
NOIR 07: 3 photos
Access and price
Gemini Omni 1.1 Flash is the one you can build on today. In the Gemini API it costs about $0.10 per second of 720p video, on the paid tier only; fal.ai lists $0.03 at 360p, $0.10 at 720p, $0.15 at 1080p and $0.30 at 4K per second. In the Gemini app and Flow it needs a Google AI Plus, Pro or Ultra plan.
Kling 4.0 is open to everyone in Kling's app, on plan credits, and its reference-to-video API opened on fal.ai on October 23, 2026, at $0.157 per second at 720p and $0.235 at 1080p. Kling's own API is early access, at preview prices of $0.14 and $0.21. A 10-second 720p clip is about $1.00 on Gemini Omni and $1.57 on Kling 4.0 through fal.ai.
Which to pick
Pick Gemini Omni 1.1 Flash if you want the lowest price per second, work in short shots, or want to edit existing video in the same model.
Pick Kling 4.0 Omni Reference if a shot needs several recurring subjects, consistent voices, or a long take of up to 30 seconds without stitching. It is in Kling's app and, since October 23, 2026, on fal.ai's API.
RefVid runs Kling 4 through fal.ai by default, with Kling 3.0 Omni as a cheaper choice. You save a character or product once, with 1 to 5 photos, and every clip starts from it; your saved assets carry over when the model changes.
Frequently asked questions
- Is Kling Omni the same as Gemini Omni?
- No. Kling Omni is Kuaishou's reference-video line in Kling (Kling 3.0 Omni, Kling 4.0 Omni Reference). Gemini Omni is Google's model family, currently Gemini Omni 1.1 Flash. They share a name and both accept reference images, but they are separate models from separate companies.
- Kling Omni vs Google Omni: which takes more references?
- Kling 4.0 Omni Reference: up to 15 inputs, including up to 10 images, 5 videos and 3 voice profiles. Gemini Omni 1.1 Flash takes up to 10 images and up to 3 short videos per prompt in Google's API, and 5 photos in the Gemini app.
- What is Gemini Omni 1.1 Flash?
- Google's current Omni video model, released on August 27, 2026. It makes 3 to 10 seconds per generation, extendable to 40, with native audio, from text, images or video, and can edit video. It is in the Gemini app, Flow, the Gemini API and on fal.ai, and it replaced Veo 3.1 in the Gemini app and the Gemini API.
- Which is cheaper?
- Gemini Omni 1.1 Flash, at about $0.10 per second of 720p in the Gemini API and on fal.ai. Kling 4.0 costs $0.157 per second at 720p on fal.ai, and Kling's preview API price is $0.14.
- Can I use Kling 4.0 Omni Reference through an API?
- Yes. Since October 23, 2026, fal.ai's API runs Kling 4.0 reference-to-video at 720p or 1080p. Kling's own developer API is still early access on request. RefVid runs it by default for saved characters and products.