From product photo to video: an OmniArt AI workflow

On June 30, Google showed off a new pairing inside Gemini: Nano Banana 2 Lite for fast image edits and Gemini Omni Flash for short video generation, chained together through the Interactions API. The flagship demo, Omni Product Studio, took one static product photo, edited it into a clean e-commerce shot, then animated that shot into a short video — all in one pipeline. Google showed the same shape twice more, with "Anywhere" (drop a photo into a new location, then animate it) and "Space Lift" (interior-design stills turned into cinematic walkthroughs). The pattern in all three: edit the image first, then hand it to a video model.

You don't need Google's specific pipeline to run this workflow. OmniArt already has the two pieces that matter — an image model built for this kind of editing pass, and a video lineup that turns a still into motion. Below is the same photo-to-video shape end to end — prep, edit, animate, export — using models confirmed live on OmniArt today, with example prompts for every step.

What's actually available on OmniArt today

Nano Banana 2 Lite is brand new, and Gemini Omni Flash's developer API and chaining pairing with it just opened up — neither is inside OmniArt's workspace yet, and this tutorial won't ask you to use them. What's live right now is Nano Banana 2 (not the new Lite tier) for the image-edit step, tagged new and hot in the image model picker, plus a full video lineup — including Seedance 2.0, the model paired with it here — for the animation step. Every prompt below runs on models you can open today.

What you'll need

• One product photo — yours or a clean catalog shot
• An OmniArt account with image and video workspace access
• Nano Banana 2 for the edit step
• A video model for the animate step — this guide uses Seedance 2.0 for how well its reference system holds a product's shape and color through motion
• Optional: OmniArt's audio models for a sound pass before export

Step 1: shoot or source a clean product photo

The edit and animation steps can only work with what's already in frame one, so a few minutes of prep here pays off twice.

• Plain, evenly lit background. White, gray, or seamless paper — Nano Banana 2 edits more cleanly off a simple base than a busy one.
• Diffused lighting. Hard shadows and blown-out reflections are what a video model is most likely to warp once the product starts moving.
• Fill the frame. Keep the product as the dominant subject; the video model follows whatever gets the most visual weight in frame one.
• Shoot your straightest, most representative angle. It becomes frame one, and every later frame gets compared against it.
• Use the highest resolution you have. Labels and fine details hold up better through both steps with more source pixels to work from.

Starting from an existing catalog photo instead of a fresh shoot? Same rules apply — flatten harsh shadows and crop out clutter before Step 2, since Nano Banana 2 edits what's already there rather than inventing a new product.

Step 2: edit and stage the photo with Nano Banana 2

Open Nano Banana 2 in OmniArt's image workspace and upload your source photo as a reference image — you're editing it, not describing a new scene from scratch. That's the part of Google's demo worth borrowing directly: the image step is a targeted edit. Nano Banana 2 is tuned for fast, balanced edits rather than slow, maximal renders, which suits generating a few staging variants per product instead of one hero shot.

Two prompts worth running:

Clean studio background:

"Edit this reference photo. Replace the background with a seamless soft lilac-gray studio backdrop. Keep the product's shape, proportions, color, and label text exactly as shown in the reference. Add soft, diffused studio lighting from the upper left and a subtle reflection on the surface below. Do not change the product itself."

Lifestyle variant:

"Edit this reference photo. Place the product on a light oak counter near a window with soft morning light. Keep the product's size, color, and label unchanged from the reference. Add a shallow depth of field with a softly blurred background. Natural, editorial product-photography style, no added text."

Generate two or three variants of each — you'll want both later, since the studio pass becomes the "official" shot and the lifestyle pass gives you a second clip to test. Pick whichever variant stays closest to the original in proportion and color; Step 3 will animate whatever drift is already in this frame.

For more on Nano Banana 2 versus other image options, see GPT Image 2 vs Nano Banana 2: which AI image model in 2026?. For prompt-writing technique, see How to write better prompts for AI generation.

Step 3: animate the shot with Seedance 2.0

Take the Nano Banana 2 output into OmniArt's video workspace and open Seedance 2.0. Upload the edited photo as your source frame, then — since Seedance 2.0 accepts up to nine tagged reference images — upload it again as @image1 inside the prompt. That gives the model a first-frame anchor plus an identity reference to hold as it generates motion. Describe the camera and environment, not the product — it's already locked from the reference, and anything you write about it risks the model "helpfully" changing it.

Hero push:

"@image1 is the product reference. Keep the product's shape, color, and label exactly as shown in @image1 throughout. Camera slowly pushes in on the product over 5 seconds. Soft light shifts gently across the surface. Static background, no camera shake, shallow depth of field."

Turntable:

"@image1 is the product reference. The product rotates smoothly 360 degrees in place on a seamless studio backdrop, camera locked in position. Lighting stays consistent through the rotation. No added props, no background elements, no text overlays."

Start in Fast mode to see the motion cheaply, then re-render your keeper in Standard once you like the prompt and seed — Standard costs more per clip, so spend it only on the version you're exporting. Keep clips short on the first pass; 4–5 seconds holds product fidelity better than jumping straight to 15.

For more Seedance 2.0 prompt patterns beyond product shots, see Seedance 2.0: prompt patterns and six use cases for AI video.

Step 4: keep the product consistent from photo to video

The riskiest moment in this workflow is the handoff between image and video — it's where color, proportions, and framing are most likely to drift.

Generate two or three seed variants and compare them side by side rather than accepting the first result — a usable clip versus a warped one is often just which seed you kept. Judge each variant on the frame where the product is furthest from its starting angle, since that's where drift shows up first.

Step 5: polish, add sound, and export

A silent product clip reads as unfinished. Since OmniArt keeps image, video, and audio in one workspace, add a sound pass before you export instead of routing to a separate tool — a soft whoosh on the camera move, light room tone, and a short music cue from the audio models. Trim to what the placement needs: 5–8 seconds for a paid-social hero clip, longer only if the extra seconds show something new. Export per channel — square or 4:5 for feed, 9:16 for Reels, TikTok, and Shorts, 16:9 for a product page or YouTube.

Save the edited studio photo from Step 2 in your reference library — it becomes the starting frame for every future video of this product, so the next SKU update or seasonal variant starts from a locked look instead of a blank page.

Picking a different video model

Seedance 2.0 is the pick here because its reference-tagging system is the most direct way to hold a product's identity through motion, but it isn't the only reasonable choice — see the FAQ below for alternatives by job. All of them live in the same video workspace, so swapping models means changing a dropdown, not rebuilding the workflow.

For a full tour of the lineup, see All AI video models in one workspace: the OmniArt lineup. For more ad-format ideas once you have a clip you like, see Turn product photos into video ads with OmniArt.

FAQ

Can I use Nano Banana 2 Lite or Gemini Omni Flash on OmniArt today?

Not yet. Nano Banana 2 Lite and Gemini Omni Flash's developer API and chaining pairing were announced June 30, 2026 (Omni Flash itself debuted earlier, at Google I/O in May), and neither is integrated into OmniArt's workspace as of this writing. This tutorial uses Nano Banana 2 (the current, non-Lite model) and Seedance 2.0 instead — both are live now. OmniArt is evaluating the new pipeline for a future addition; once it lands, see our Nano Banana 2 Lite prompt guide for the details.

Why edit the photo before animating it instead of animating the raw photo?

Because the edit step controls the variables hardest to fix once motion is involved — background, lighting, staging. Fixing those in a still with Nano Banana 2 first means the video model only has to add motion, not fix a busy background while holding the product steady.

Do I need professional photography equipment for the source photo?

No. A phone photo on a plain, evenly lit background works — the requirement is consistency, not equipment. Nano Banana 2's editing step is what turns a decent phone photo into a studio-looking still.

How long should the final product video be?

For paid social and feed placements, 5–8 seconds is the workable range — long enough to show motion, short enough to hold attention. Product pages and YouTube can run longer if the extra seconds show something new.

Which video model should I use if not Seedance 2.0?

It depends on the job: Veo 3.1 for native 4K broadcast-style output, Kling 3.0 for cost-efficient output across many SKUs, PixVerse for fast, lifestyle-leaning social clips. All are available in the same OmniArt video workspace as Seedance 2.0.

Can I add music or sound effects to the final video on OmniArt?

Yes. OmniArt's audio models sit next to the image and video workspaces, so you can add a music cue, sound effects, or a voiceover to the exported clip without leaving the platform.

Getting started on OmniArt

Pick one product, one clean photo, and fifteen minutes. Edit it with Nano Banana 2 in the image workspace, animate the result with Seedance 2.0 in the video workspace, and add a short music cue before you export. That's the same photo-to-video shape Google previewed for Gemini, running today on models already inside OmniArt.