Sora 2 Prompt Patterns for Product Videos: 8 Recipes That Actually Ship

Sora 2 shipped to business users in early 2026, and the first wave of "AI product videos" on Instagram and TikTok has been, frankly, terrible. Robotic camera moves. Uncanny hands. Products that morph mid-shot. The problem isn't the model — it's the prompting. Sora 2 is a motion-and-physics model, not a text-to-image model with extra steps, and the prompt patterns that worked for DALL·E 3 or even early Sora don't translate.

This post gives you eight prompt patterns we've tested against real product briefs. Each one ships a video you can actually post — hero shots, texture close-ups, unboxing, flat-to-lifestyle transitions, UGC-style, before/after, motion loops, and Reels-native vertical formats. Swap the product, keep the skeleton.

Why prompt patterns matter more for video than for images

A single image has roughly four variables you have to get right: subject, style, composition, lighting. A Sora 2 clip adds at least four more: camera move, duration, physics (how things fall, drip, shake, bounce), and the cut point. Miss any one and the clip breaks — either it feels like stock footage, or worse, it falls into the uncanny-valley "AI video" aesthetic that viewers now instantly pattern-match and scroll past.

The solution is to write prompts that explicitly name all eight variables. Below are eight skeletons that do exactly that.

Each pattern assumes you're using Sora 2 or Sora 2 Pro via the OpenAI API (`sora-2`, `sora-2-pro`) or the ChatGPT interface with Images 2.0 enabled. Clip lengths are the defaults most product-video teams are using: 5-10 seconds for Reels and TikTok, 2-4 seconds for feed loops and ad creative.

1. The hero shot pattern

Use this when the product is the entire story — a new launch, a premium SKU, a single-product ad. The subject is stationary, the camera does the work.

Skeleton:

``` A [product] sits centered on a [surface material] under [lighting condition]. The camera slowly orbits 90 degrees to the right over 6 seconds, holding the product in the center of frame at all times. Depth of field: shallow, f/2.8 look. Background [background style] is softly blurred. Lighting: [key light direction] with a soft fill. Color palette: [brand colors]. Mood: [mood keywords]. End on a 1-second static hold of the product, square to camera. ```

Why it works: Sora 2 is reliable at orbit moves because they're common in the training data (product videography, jewelry shoots, car reveals). The final static hold gives you a clean cut point for a branded end card.

Example prompt (brass coffee grinder):

A hand-turned brass coffee grinder sits centered on a dark walnut table under warm morning sunlight streaming from a window at frame left. The camera slowly orbits 90 degrees to the right over 6 seconds, holding the grinder in the center of frame at all times. Depth of field: shallow, f/2.8 look. The background, a softly out-of-focus tiled kitchen wall, is blurred. Lighting: directional sunlight from the left, soft fill from a bounce card on the right. Color palette: warm amber, cream, walnut brown. Mood: artisanal, unhurried, tactile. End on a 1-second static hold of the grinder, square to camera.

2. The texture close-up pattern

Use for fabric, food, skincare, anything where the viewer needs to feel the surface. Close-ups are where Sora 2 genuinely out-performs stock footage because it can simulate specific material properties on demand.

Skeleton:

``` Extreme macro close-up of [product surface/material], filling the frame. The camera slowly pushes in 20% over 4 seconds. [Describe the micro-motion: fabric fibers catching light, oil pooling, steam rising, crumbs falling]. Lighting: [side light, rim light, backlight]. Shallow depth of field, one plane of focus. Color grade: [grade keywords]. ```

Why it works: Naming the micro-motion explicitly prevents Sora 2 from adding its own (which is where weird physics glitches come from). "Fabric fibers catching light" is a real, physical description; Sora 2 can render it.

Example (linen shirt):

Extreme macro close-up of washed natural linen fabric, filling the frame. The camera slowly pushes in 20% over 4 seconds. Individual fibers catch raking sidelight from the left, creating soft highlights and shadows across the weave. A single thread stirs slightly as if in a gentle breeze. Lighting: hard side light from the left, no fill. Shallow depth of field, one plane of focus on the fabric surface. Color grade: natural, slightly warm, high detail in the shadows.

3. The unboxing-style reveal pattern

Unboxing videos are the highest-saving format for ecommerce on TikTok. Sora 2 can simulate them convincingly — but only if you stage the shot like a real unboxing, not like a VFX sequence.

Skeleton:

``` Overhead (top-down) shot of a [brand] box on a [surface]. Two hands enter from the bottom of the frame and lift the lid. Inside, [product] is revealed, nested in [packaging material]. The hands gently lift the product out of the box over 4 seconds. Camera: static overhead. Lighting: flat, soft, even (tabletop photography lighting). Background: [background surface]. Color palette: [brand colors]. ```

Why it works: Overhead + static + hands-only is the unboxing grammar viewers recognize. Sora 2 handles this reliably because it avoids full-body human rendering (where it's weakest) and stays in hand-product interaction (where it's strongest).

A note on hands: specify "two hands" and describe what they do. Leaving hands implicit is how you get four-fingered nightmares.

4. The flat-to-lifestyle transition pattern

This is the "mix shot" that performs best on Instagram Reels for fashion, home goods, and accessories. Start static, add motion, end in context.

Skeleton:

``` Shot 1 (0-2s): Overhead flat-lay of [product] on [surface] with [2-3 styling props]. Camera static. Lighting: soft, even. Shot 2 (2-4s): The camera lifts up and tilts to a 45-degree angle, revealing [context or environment] in the background. Shot 3 (4-6s): The camera pushes in on the product at 45 degrees, [styling prop] now in the foreground out of focus. Consistent lighting and color palette throughout. ```

Why it works: Sora 2 handles multi-shot prompts if you explicitly name them as "Shot 1 / Shot 2 / Shot 3" with time codes. Treating the prompt as a storyboard is the unlock. Use the same color palette and lighting keywords in every shot to prevent grade shifts at the cut.

5. The UGC-style pattern

"User-generated content" style is the most scroll-stopping format for DTC brands on TikTok. It looks like a customer filmed it on their phone — not a brand-produced ad. Sora 2 can fake this credibly with the right signal words.

Skeleton:

``` Vertical 9:16 format. Handheld iPhone-style camera, slight natural shake. A [product] is held up to the camera by [unseen person's hand or the frame's POV]. [Brief natural motion: tilting the product, rotating it, squeezing it]. Lighting: natural window light, slightly mixed with warm indoor light. No studio look. Imperfect composition. 4 seconds. Soft autofocus hunt once. ```

Why it works: "Handheld iPhone-style" + "slight natural shake" + "soft autofocus hunt once" are the three details that strip out the "AI video" uncanniness. The word "imperfect" is load-bearing — it tells Sora 2 to skip its default polish.

Pair this with ChatGPT-generated caption copy that doesn't sound like AI, and the whole post reads like a real customer.

6. The before/after transformation pattern

For skincare, cleaning products, laundry, home renovation — any product that demonstrates change. This is where Sora 2's physics engine earns its keep.

Skeleton:

``` Shot 1 (0-2s): [Surface or subject] in its "before" state. [Describe the condition in concrete visual terms]. Shot 2 (2-3s): [Product] is applied to the surface. Describe the application motion. Shot 3 (3-6s): The surface transitions to its "after" state. [Describe the change as a continuous physical process, not a cut]. Camera: static or slow push-in. Consistent lighting throughout. ```

Critical rule: Describe the change as a continuous physical process ("the stain gradually lifts as the foam breaks down, revealing the clean tile underneath over 3 seconds"), not as a before/after cut. Before/after cuts trigger Meta and TikTok ad-policy flags as potentially deceptive. Continuous-process videos do not.

If you're making these for regulated categories (beauty, health, household claims), add a compliance disclaimer overlay in post. See our AI product photography guide for the claim-safe language framework we use with ecommerce clients.

7. The motion-loop pattern

3-4 second loops that play seamlessly on Instagram feed are the highest-ROI format for ad creative. The product has one motion — pouring, drizzling, spinning, shaking — and the clip is designed to loop invisibly.

Skeleton:

``` Close-up of [product] performing [single continuous motion]. The motion starts and ends in the exact same frame position, so the clip loops seamlessly. Duration: 3 seconds. Camera: static. Lighting: dramatic [studio lighting setup]. Background: [solid color or textured surface]. Color grade: [grade keywords], high saturation for feed stopping power. ```

Why it works: Sora 2 respects the "loop seamlessly" instruction when you also specify "starts and ends in the exact same frame position." Without that explicit framing, you get a jump cut at the loop point. Ship these as 1:1 for Instagram feed ads.

8. The Reels-native vertical pattern

9:16 vertical, 8-15 seconds, designed for Instagram Reels and TikTok. This pattern layers everything above into a single recipe for social-first product videos.

Skeleton:

``` Vertical 9:16 format. 12-second clip. Beat 1 (0-2s): Hook — [surprising visual or text overlay prompt]. Beat 2 (2-6s): Product reveal using [Pattern 1, 3, or 4 above]. Beat 3 (6-10s): Lifestyle or in-use shot — [describe]. Beat 4 (10-12s): End card space — static hero shot of the product, leaving top third of frame empty for text overlay in post. Consistent lighting, color palette, and motion pacing throughout. Camera moves should be slow and intentional, no whip pans. ```

Why it works: Naming the beats with time codes forces Sora 2 to pace the clip. Without beats, it tends to rush the first 3 seconds and drag the last 5 — the exact opposite of what Reels rewards.

Leave the top third empty in your final beat so your Instagram post copy and on-video text overlay have breathing room. This small staging choice is the difference between a clip that looks made-for-Reels and one that looks cropped-from-a-commercial.

Quick wins: three things to do today

1. Pick two patterns. Don't try all eight at once. For most ecommerce brands, Pattern 1 (hero) and Pattern 7 (motion loop) cover 80% of feed and ad needs. For fashion or food, Pattern 2 (texture close-up) and Pattern 8 (Reels-native) are usually stronger starters.

2. Build a prompt library. Copy the two skeletons above into a Notion or Google Doc. Fill in the bracketed variables for your three best-selling products. That's six tested prompts — more than most brands ship in a quarter.

3. A/B test against your stock footage. Post a Sora 2 clip and a matching stock or phone-shot clip, same caption, same time of day, one week apart. Watch save rate and completion rate. You'll know within a week whether Sora 2 earns a permanent place in your workflow or stays situational.

Want to pair Sora 2 clips with on-brand still visuals and captions? Start with Adpicto free — no credit card required, 5 free AI images per month plus branded caption generation from your logo and color system.

Ship your first Sora 2 video this week

The teams shipping good Sora 2 product videos in 2026 aren't the ones with the fanciest prompts — they're the ones with a library of 6-8 reusable skeletons they can adapt in 5 minutes. You now have that library.

Start with one product and one pattern. Render, watch, tweak the variables, render again. Three iterations usually get you to a post-worthy clip. From there, the pattern-library approach scales — two hero-shot prompts plus two motion-loop prompts is enough to ship a full week of product Reels, without once staring at a blank prompt box wondering what to type.