# Prompting Gemini Omni Gemini Omni makes one continuous shot of up to 10 seconds, with sound. Describe one moment, move the camera inside it, and hold products steady with reference images. Last updated: 2026-10-05 **Key facts** - Gemini Omni in Sloosh makes one continuous shot of 3 to 10 seconds with sound, from a prompt, up to 6 reference images, or one video of up to 10 seconds. - Sloosh has two Gemini Omni models: Gemini Omni Flash and Gemini Omni 1.1 Flash, which adds 360p, 720p, 1080p or 4K and can extend a clip in 10-second steps up to 40 seconds. - Gemini Omni in Sloosh makes 16:9 or 9:16 video only. - A Gemini Omni prompt works best as one beat with its own Audio line: shot, style, light, place, action, then audio. Web page: https://sloosh.ai/docs/prompting-gemini-omni Sloosh has two Omni models, Gemini Omni Flash and Gemini Omni 1.1 Flash. Both make video with audio from a prompt, reference images, or a video you want changed. 1.1 adds resolution choices and can extend a clip. Omni is at its best changing footage you already have: put your product in, restyle a scene, take something out. For a clip from scratch, also try Veo 3.1. ## How it reads a prompt Omni thinks in single shots of up to 10 seconds. It knows a lot about the world, so say what you want to see and let it work out the details. Scripting every frame fights that knowledge and makes the result drift. One clip is one beat. Ask for seven cuts in ten seconds and Omni blends them into one weak shot. ## A prompt that works - Shot: framing and camera move, such as "slow push in", "locked off" or "one continuous shot". - Style: photoreal, cinematic, claymation, flat 2D animation. - Light: the source and its quality. - Place: the setting in a line or two. - Action: what happens, in order. - Audio: its own line for voice, music and ambient sound. **Template** ```text A single continuous [length] shot. [Style]. [Place]. [Light]. The camera [move] as [subject] [action]. [One caption in quotes, and when and where it appears.] Audio: [music, voice, ambient sound]. ``` Omni knows camera vocabulary: push in, tracking shot, over the shoulder, dolly zoom, handheld, 35mm, shallow depth of field. Name the lens and the move. ## Describe what you want, not what you don't Omni has no negative prompt, and a list of "no this, no that" tends to pull those things in. Describe the thing you want instead. "No trousers, no jeans" becomes "an ankle-length white cotton wrap tied at the waist". ## Captions and audio Omni animates text well when there is one caption on screen. Quote it, spell it right, and say when and where it appears: "'NEW' fades in at the three-second mark, bottom left". Several captions in one clip are what garble the type. Audio comes out of the same run. Without an Audio line, Omni guesses. Keep dialogue short and give its tone. ## References and tags Omni takes up to 6 reference images and one video. Use images for anything that must look exact: your product, a face, a mascot. A character described only in words drifts back to something ordinary, so build it first as an image with [Nano Banana](https://sloosh.ai/docs/prompting-nano-banana.md) and bring that in. Refer to reference images by the order you added them: "the bottle in the first image", "the jacket in the second image". Omni's API also understands the tags ``, `` and ``, but plain words work just as well. There is no tag for a video, so call it "the video". Give a person or product that recurs across an ad a short handle, such as @chef, and repeat it in every prompt. ### Where to add them - Studio, Video mode: add reference images the usual way. Attach a video and Studio switches to editing it, so describe the change in words. See [Studio](https://sloosh.ai/docs/studio.md). - Spacelab: the Video (references) node takes Omni's images and one reference video. The Video edit node changes or extends a source video. See [Generate nodes](https://sloosh.ai/docs/nodes-generate.md) and [Edit nodes](https://sloosh.ai/docs/nodes-edit.md). ### Limits on video input - One video per run, 10 seconds or shorter. - A reference video used only for its style or camera motion doesn't work. Describe the look and the move in words instead. ## Before and after ### Product hero shot **Before** ```text Shot 1: wide of kitchen. Shot 2: close up of blender. Shot 3: smoothie pouring. Shot 4: logo. Shot 5: happy family. 15 seconds. ``` **After** ```text A single continuous 10-second shot. Photoreal. A sunny kitchen counter, morning light through the window behind. The camera pushes in slowly on the blender from the first image as a hand pours a thick pink smoothie from its jug into a glass, ending close on the glass. Shallow depth of field, fresh and warm. Audio: the blender winding down, the pour, soft acoustic guitar. ``` Five shots and a logo don't fit in one clip. Make the family moment a second clip if you need it. ### Garment on a model **Before** ```text man wearing kurta, no jeans, no black, no logo, no text ``` **After** ```text A single continuous 8-second shot. A man in his twenties wears the white linen kurta from the first image with loose off-white cotton trousers, walking toward the camera through a sunlit courtyard. Medium shot, the camera tracking backward with him, the fabric moving with each step. Soft afternoon light. Audio: footsteps on stone, distant birds. ``` ### Ad with one caption **Before** ```text sneaker ad with text SALE 50% OFF LIMITED TIME SHOP NOW and logo ``` **After** ```text A single continuous 6-second shot. The sneaker from the first image turns slowly on a concrete plinth under a hard white spotlight, black background. The camera circles it once. One caption, "50% OFF", in bold white sans serif, fades in at the three-second mark, centred below the shoe. Audio: a deep bass hit as the caption lands. ``` ### Edit your own footage **Before** ```text make it better and change the product ``` **After** ```text Replace the bottle in the woman's hand with the bottle in the first image. Keep her movement, the camera, the light and everything else in the video as it is. ``` Attach the clip as the video and the bottle as an image. Name the one change and protect the rest. ## Longer ads For 20 or 30 seconds, make several 10-second beats and join them with Video Concat in Spacelab. Repeat the same handles, style line and light line in every prompt so the beats match. Gemini Omni 1.1 Flash can also extend a clip in 10-second steps, up to 40 seconds, with the Video edit node's Extend task. ## Common mistakes - A shot list in one clip. Write one moment. - A wall of negatives. Describe what you want. - Several captions at once. Keep one. - A product or mascot from words alone. Give it a reference image. - Scripting every frame. State the intent and let Omni fill in the detail. - A reference video for style. Put the style in words. ## Limits in Sloosh - Both let you set 3 to 10 seconds. Gemini Omni Flash picks its own resolution; Gemini Omni 1.1 Flash lets you pick 360p, 720p, 1080p or 4K. - 16:9 or 9:16 only. - Up to 6 reference images and one video of up to 10 seconds. - Extend: Gemini Omni 1.1 Flash only, in 10-second steps, 40 seconds in total. Google's guides describe more input options than these. The list above is what Sloosh accepts today. ## Further reading [Google DeepMind: Gemini Omni prompt guide](https://deepmind.google/models/gemini-omni/prompt-guide/). In these docs: [Video prompts](https://sloosh.ai/docs/video-prompts.md). ## Frequently asked questions ### Does Gemini Omni support negative prompts? No. Gemini Omni has no negative prompt, and a list of "no this, no that" tends to pull those things in. Describe what you want instead: "No trousers, no jeans" becomes "an ankle-length white cotton wrap tied at the waist". ### How do I add a caption to a Gemini Omni video? Use one caption per clip: quote it, spell it right, and say when and where it appears, such as "'NEW' fades in at the three-second mark, bottom left". Several captions in one Gemini Omni clip are what garble the type. ### How do I make a 30-second ad with Gemini Omni? Make several 10-second beats and join them with Video Concat in Spacelab, repeating the same handles, style line and light line in every prompt so the beats match. Gemini Omni 1.1 Flash can also extend a clip in 10-second steps, up to 40 seconds. ### Can Gemini Omni copy the style or camera move of a reference video? No. A reference video used only for its style or camera motion doesn't work with Gemini Omni. Describe the look and the camera move in words instead, and use the video slot for footage you want changed. ### How do I replace a product in my own video with Gemini Omni? Attach your clip as the video and the product as an image, then name the one change and protect the rest: "Replace the bottle in the woman's hand with the bottle in the first image. Keep her movement, the camera, the light and everything else as it is."