Gemini Omni Flash for AI persona video: what it allows

Gemini Omni Flash makes vertical 9:16 video with sound from your persona’s reference images, for about $0.10 per second at 720p. Google blocks uploads of “certain recognizable people” without defining them. In the EU, Switzerland and the UK you also cannot edit or extend uploaded videos. Every clip carries an invisible SynthID watermark.

So Omni works for an invented persona, as long as her face does not look like a real, recognisable person.

What Gemini Omni Flash is and how you get it

Gemini Omni 1.1 Flash is Google’s video model in the Gemini API. It generates, edits and extends video with sound, and it takes images and short video clips as references.

“Gemini Omni Flash (gemini-omni-1.1-flash) is a high-performance multimodal model designed for high-speed video generation, editing, and cinematic control.” Conversational editing “lets you iteratively refine and edit your videos through natural language conversation.” Last updated 23 September 2026.

Google AI for Developers — Gemini Omni Flash (opens in a new tab)checked 5 Oct 2026

Published 27 August 2026. “You can extend videos in 10-second increments up to a total cumulative length of 40 seconds.” 360p previews come “up to 60% faster* and at a third of the cost compared to Omni 1.1’s standard 720p resolution.”

Google Blog — Build with Gemini Omni 1.1 Flash (opens in a new tab)checked 5 Oct 2026

You use it through Google AI Studio or the Gemini API with a paid key. For a persona, three features matter:

• Subject references. You upload images of her, and the model puts her into the scene you describe.
• 9:16 output. Vertical video for Reels, TikTok and Shorts, without cropping.
• Edits in a conversation. You change the light or the background of a clip in a follow-up prompt, without describing the whole scene again.

Gemini Omni Flash price per second, worked out

Google prices Omni in tokens. At 720p one second costs about $0.10, a 10-second clip about $1.01, and a 360p draft about a third of that.

Gemini Omni Flash, gemini-omni-1.1-flash, “now generally available to developers on the paid tier of the Gemini API”. Free Tier: “Not available”. Paid tier per 1M tokens: input “$1.50 (text / image / video / audio)”, output “$17.50 (video)”. “Billing is based on total output token consumption, calculated at a rate of 5,792 tokens per second of 720p video. Under Standard pricing, this equates to an effective price of approximately $0.10 per second.” Veo 3.1 per second: Standard $0.40 (720p and 1080p), Fast $0.10 (720p), Lite $0.05 (720p). Above them: “Veo 3.1 Preview (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview) is deprecated and will be shut down on October 22, 2026; migrate to Gemini Omni Flash (gemini-omni-1.1-flash) or the GA models available through the Gemini Enterprise Agent Platform to avoid service disruption.” Page last updated 19 October 2026.

Google AI for Developers — Gemini API pricing (opens in a new tab)checked 21 Oct 2026

The sum, from Google’s own numbers: 5,792 tokens × $17.50 per million = $0.1014 per second of 720p video.

Clip at 720p	Output tokens	Output cost	360p draft (a third)
5 seconds	28,960	about $0.51	about $0.17
10 seconds	57,920	about $1.01	about $0.34
20 seconds (one extension)	115,840	about $2.03	about $0.68
40 seconds (the maximum)	231,680	about $4.06	about $1.35

Your reference images add input tokens at $1.50 per million. Google does not publish how many tokens one image counts as on the Omni page, so check the usage figure in your first response. Google also gives no token rate for 1080p or 4K, which it describes as upscaled. Price those from your own bill.

A working routine: draft at 360p until the movement and framing are right, then render once at 720p. Extend the good clip in the same conversation.

1. 360p draftabout $0.17 for 5 s
2. Check her faceagainst your stills
3. 720p renderabout $0.51 for 5 s
4. Extend in the chat10 s steps, 40 s total
5. Label and postSynthID is not a label

Schematicthe shape, not measured data

What generation costs lists the other video models with their dates.

How to use your persona as a reference in Gemini Omni

Upload two or three images of her and refer to them in the prompt with Google’s tags. Ask for one continuous shot, because Omni cuts between shots by default.

“You can generate a video incorporating specific subjects provided as reference images.” Tags: <FIRST_FRAME> “use the image as the starting frame of the video”, <IMAGE_REF_N> “use the image as a reference”. For reference images, add: “Use the given image(s) as references for video generation. The images should not be used as literal initial frames.” “By default Omni Flash will try to create a video with a few different shots. If you need the output video to contain a single scene, you must prompt for that”. aspect_ratio supported values: “9:16”, “16:9”. Default resolution 720p; 1080p and 4K are “upscaled”.

Google AI for Developers — Gemini Omni Flash (opens in a new tab)checked 5 Oct 2026

The steps:

1. Pick two or three stills of your persona: a front close-up, a three-quarter view and a full-body image. ChatGPT AI influencer photos shows which reference images make a stable set.
2. Name each image’s job with a tag: <IMAGE_REF_0> for her face, <IMAGE_REF_1> for her outfit.
3. Describe one scene, with camera, light and sound. Write “one continuous shot”.
4. Ask for 9:16 in the request.
5. Write what you do not want as a sentence. Omni has no negative prompt field. Google suggests “Do not …” in the normal prompt instead.
6. Check the face against your stills before you edit anything. A wrong face is cheaper to regenerate than to fix.

Once one clip holds her face, you can reuse it. Google says video references “work best with likenesses”. Cut a 3-second piece of that clip where her face is clear and add it as a video reference next to her stills. Omni takes up to three such clips.

A request in Python, in the form of Google’s own examples. Google’s quickstart has you set your key as the GEMINI_API_KEY environment variable:

import base64
from google import genai

client = genai.Client()

def b64(path):
    with open(path, "rb") as f:
        return base64.b64encode(f.read()).decode()

interaction = client.interactions.create(
    model="gemini-omni-1.1-flash",
    input=[
        {"type": "image", "data": b64("mara-front.png"), "mime_type": "image/png"},
        {"type": "image", "data": b64("mara-full-body.png"), "mime_type": "image/png"},
        {"type": "text", "text": (
            "A woman <IMAGE_REF_0> wearing the outfit in <IMAGE_REF_1> walks along a sunny "
            "harbour wall and smiles at the camera. One continuous handheld shot, eye level. "
            "Sound: light wind, gulls, no music, no dialogue. "
            "Use the given images as references for video generation. "
            "The images should not be used as literal initial frames."
        )},
    ],
    response_format={"type": "video", "aspect_ratio": "9:16"},
)

with open("mara-harbour.mp4", "wb") as f:
    f.write(base64.b64decode(interaction.output_video.data))

What the parts do:

• genai.Client() reads your key from GEMINI_API_KEY, so the key is not in the script.
• b64() turns each image file into text, because the request carries the images inside it.
• The two image entries come first. In the prompt, <IMAGE_REF_0> points at the first image (her face) and <IMAGE_REF_1> at the second (her outfit).
• The last two prompt sentences are Google’s own wording. They stop Omni from using her photo as the first frame.
• response_format asks for a video in 9:16.
• The last two lines save the returned video as an MP4 file.

To make a clip longer, extend it in a follow-up turn of the same conversation. Omni uses the last 10 seconds as context. Google notes that “Some of the final frames in your input video will be edited to make the transition seamless.”

What Gemini Omni Flash does not allow for an AI persona

Google lists the limits on the Omni page and in its Prohibited Use Policy. Three of them hit persona work directly, and two only hit you in Europe.

Limitations: “Uploading and editing images containing minors is not supported in European Economic Area, Switzerland, and the United Kingdom.” “Uploading and editing images containing certain recognizable people is not supported.” “Editing or extending uploaded videos is not currently available for users in the European Economic Area (EEA), Switzerland, and the United Kingdom (editing or extending videos generated by the model is supported).” “Uploading audio references is unsupported in the current version of the API.” “Video references work best with likenesses; any audio in a video reference is ignored. Video references support a maximum of 3 clips, up to 3 seconds each.” “System instructions, temperature, top_p, stop sequences, and negative prompts are not supported (you can put your negatives in the regular prompt: e.g., “Do not do X”).” “Omni applies content safety filters to both input prompts and generated video (which vary by region).”

Google AI for Developers — Gemini Omni Flash (opens in a new tab)checked 5 Oct 2026

Not allowed: “Impersonating an individual (living or dead) without explicit disclosure, in order to deceive.” “Misrepresenting the provenance of generated content by claiming it was created solely by a human, in order to deceive.” “Sexually explicit content”.

Google — Generative AI Prohibited Use Policy (opens in a new tab)checked 5 Oct 2026

What you want to do	Outside Europe	EEA, Switzerland, UK
Generate from your persona’s images	Yes, if she is not a recognisable real person	Yes, same rule
Use an image that shows a minor	Not stated as blocked	Not supported
Edit or extend a video you uploaded	Yes, up to 10 s input	Not available
Edit or extend a video Omni made	Yes	Yes
Give her a voice from your own audio file	No, audio references are not supported	No

What that means in practice:

• Your persona must not look like a real person. Google blocks “certain recognizable people” and forbids impersonation. The risk guide covers what happens when a persona looks too much like someone real.
• Keep every image clearly adult. In Europe, uploads with minors are not supported at all.
• In Europe, make your edits inside Omni. Generate the clip in Omni and edit it in the same conversation. You cannot bring in a clip from another model and extend it there.
• No saved voice yet. You can describe a voice in the prompt, but you cannot upload her voice sample.
• Keep everything SFW. Google’s policy and filters block sexual content.

Gemini Omni SynthID and your AI label

Every Omni video carries SynthID, an invisible watermark that software can read. It is not a visible label, so you still label the post.

“All generated videos include SynthID watermarking, which is invisible to viewers but can be detected programmatically for provenance verification.”

Google AI for Developers — Gemini Omni Flash (opens in a new tab)checked 5 Oct 2026

Your audience cannot see SynthID, so the visible label is your job. The label matrix shows what each platform asks for. Don’t strip your watermarks explains the marking duty that generators like Google carry, and why you keep such markers in the file. This is not legal advice.

Gemini Omni Flash vs Veo 3.1 for an AI persona

On price per second at 720p, Omni sat level with Veo 3.1 Fast and above Veo 3.1 Lite. But Google shuts the Veo 3.1 preview models down in the Gemini API on 22 October 2026 and names Omni Flash as the model to move to. From then on, Google’s own route to Veo 3.1 is its Gemini Enterprise Agent Platform on Google Cloud, at the same prices. Omni’s advantage for a persona is the workflow: subject references, edits in conversation and extension to 40 seconds.

“Video + Audio generation” at 720p: “Veo 3.1 Lite” “$0.05 / 1 count”, “Veo 3.1 Fast” “$0.10 / 1 count”, “Veo 3.1” “$0.40 / 1 count”. The table does not name the unit; the Gemini API listed the same Veo figures per second.

Google Cloud — Agent Platform pricing (opens in a new tab)checked 21 Oct 2026

At 720p with sound, read on 21 October 2026	Price per second
Veo 3.1 Lite, Google Cloud	$0.05
Veo 3.1 Fast, Google Cloud	$0.10
Gemini Omni 1.1 Flash, Gemini API	about $0.10
Veo 3.1 Standard, Google Cloud	$0.40

Use Omni when one persona has to appear in many scenes and you want to adjust clips by prompt. For a single short image-to-video clip, compare the cheaper models first. Other models fit other jobs:

Your job	Model to try first	Why
Many scenes, adjusted by prompt	Gemini Omni Flash	Subject references, edits in the conversation, extension to 40 s
As many images of her as possible	MiniMax H3	Up to nine reference images in one clip
A real movement, like a dance	Kling motion control	Copies a take you filmed, with a face element
A cheap first test of a scene	Gemini Omni Flash at 360p	A third of the 720p cost, by Google’s own figure

If your persona’s face is refused, Seedance consistent character shows how one strict model handles realistic faces. The best AI video model for an AI influencer has a ten-minute test for any model on your character, and the character consistency guide covers the stills that every video model starts from.

Start with one 360p draft today. Upload her front close-up and full-body image, ask for one continuous 9:16 shot, and compare the face with your stills. If it holds, render the same prompt at 720p.