Making Videos with Gemini 4 Argon

Can Gemini 4 Argon make videos? Based on what Google has announced, no: Argon writes text. What it can do is understand video, which makes it a strong assistant director. Here's the honest picture, the Google models that do generate video, and a workflow for a finished clip today.

What Gemini 4 Argon can do for video

Google reports Argon scores 91.7% on LVBench, a long-video understanding benchmark. In a video project, that makes it good at the thinking work:

• Planning: a script, a shot list and one prompt per shot.
• Reviewing: spotting a product that changes color between shots.
• Logging: finding the best moments in long footage, with timestamps.
• Packaging: titles, captions and descriptions.

Argon isn't open to most people yet (see how to get access and the release date). Meanwhile, gemini-3.8-flash already accepts video in the Gemini API, and Claude can use it through our Claude and Gemini setup.

Google's models that do make video

Gemini Omni

Announced at Google I/O 2026, Gemini Omni Flash makes short clips (about 10 seconds each at launch) from any mix of text, images, audio and video, and by default it tries to add a fitting audio track. You edit by conversation: ask for a red jacket and it keeps the rest of the scene. Google rolled it out in the Gemini app and Google Flow for Google AI Plus, Pro and Ultra subscribers, free on YouTube Shorts and YouTube Create, and it's in the Gemini API.

Veo 3.1

Veo is Google's dedicated video line. Veo 3.1 makes clips of up to 8 seconds with native audio (dialogue, sound effects, ambience) at resolutions up to 4K. You can guide it with reference images or first and last frames, and extend clips into longer scenes.

The workflow: plan, generate, finish

1. Plan. Use Argon once you have access, or any capable language model today, to write the concept, shot list and prompts.
2. Lock the look. Make one clean product or character still (for example with Nano Banana 2) and use it as the start frame or reference in every shot.
3. Generate. Run each prompt in Gemini Omni or Veo 3.1, and try Seedance 2.5, Kling O3 Video or MiniMax H3 Video. Different models win different shots.
4. Review. Give the clips to a video-understanding model and ask for continuity problems and cut points.
5. Finish. Upscale, add lip sync if someone speaks, lay one music track under the cut and add subtitles.
6. Assemble. Add titles and your logo in your editor, where you control the spelling.

Worked example: a 30-second product teaser

Here's a full plan for Driftline, a made-up travel mug, in 16:9. Four shots of 8, 8, 8 and 6 seconds fit Veo 3.1's clip lengths (4, 6 or 8 seconds in the Gemini API) and Gemini Omni Flash's roughly 10-second clips.

1. The planning prompt

You are a commercial director. Plan a 30-second teaser for Driftline,
an insulated travel mug that keeps drinks hot or cold all day.
Audience: commuters and weekend hikers. Format: 16:9, no voiceover.
Give me: a one-line concept; 4 shots of 8, 8, 8 and 6 seconds; for each
shot a video prompt covering camera, subject, action, setting, lighting,
style, audio and what to avoid; the end-card text; a music brief.
Describe the mug word for word the same in every prompt.

2. The shot list

3. The shot prompts

Shot 1 (8s): Macro close-up, slow drift left. A sage-green steel travel mug
on a wet wooden dock, thin steam rising. Misty lake at dawn, soft blue light,
shallow focus. Audio: water lapping, one low synth note. No dialogue.
Avoid: scene cuts, people, on-screen text.

Shot 2 (8s): Handheld tracking shot. A cyclist pulls the sage-green steel
travel mug from a bike cage, sips, rides over wet cobblestones. Overcast
morning, cool tones. Audio: tires on stone, light percussion. No dialogue.

Shot 3 (8s): Top-down. A hand opens the sage-green steel travel mug on a
sunny picnic table; ice cubes still float. Warm light, crisp shadows.
Audio: ice clinking, music building. Avoid: scene cuts.

Shot 4 (6s): Slow push-in. The sage-green steel travel mug on a mossy rock
at golden hour, framed left, empty space right. Audio: music resolves,
soft whoosh. Avoid: on-screen text.

Feed the same product still into every shot as a reference or start frame. That matters more for consistency than prompt wording.

A video prompt template you can copy

The fields follow Google's Veo prompt guidance and its Gemini Omni tips: describe the sound you want, and say "no dialogue" or "no scene cuts" when you mean it.

[Shot type and camera move]. [Subject, described the same way every time]
[action], in [setting].
Lighting: [time of day, light quality, mood].
Style: [look, color palette, lens, e.g. shallow focus, 35mm film].
Audio: [ambient sound], [sound effects], [music mood].
Dialogue: ["exact words in quotes"] or no dialogue.
Avoid: [scene cuts, on-screen text, extra people, logos].
Format: [length] seconds, [16:9 or 9:16].

• Keep text out of the video. Add titles in editing, where spelling is guaranteed.
• Edit instead of re-rolling. With Gemini Omni, ask for the one change you want.

Use Argon to review your footage

A video-understanding model is a fast second pair of eyes. Upload your clips in order with:

These are four clips for a 30-second teaser, in order.
1. Check the product in every clip: color, lid shape, proportions.
   List any frame where it looks different, with timestamps.
2. Flag warped hands, flicker, or objects that appear or vanish.
3. Suggest in and out points so the cuts total 30 seconds.

Until you have Argon, gemini-3.8-flash can do this job in the Gemini API.

Frequently asked questions

Sources: Google (Argon), 9to5Google, Google (Omni), The Next Web, Gemini API: Omni, Google DeepMind: Veo, Gemini API: Veo 3.1, Gemini API: video understanding.