How to Make a Restaurant Video with AI
The exact 4-step workflow to make a restaurant video with AI: model picks per shot type, routing by cuisine, Italian trattoria walkthrough at $6 in compute, and what breaks.
You can make a restaurant video with AI in four steps: generate dish stills with Nano Banana Pro, animate pours and action shots with Seedance 2.0, cut in ambient ambiance footage with Kling 3.0, and drop music under it. A complete 30-second opening promo for an Italian trattoria costs about $6 in model credits on the 8frame canvas. This guide covers every format, the routing logic by cuisine type, and the three places where food video generation reliably breaks.
TL;DR
- Dish stills: Nano Banana Pro handles plated food rendering at speed; specify surface material, light direction, and angle or the model defaults to flat supermarket lighting
- Motion shots: Seedance 2.0 animates pours, sauce drizzles, and hand-service with multi-reference conditioning; upload the actual dish photo as a reference image
- Ambiance cuts: Kling 3.0 for dining room atmosphere, candlelight flicker, street-entrance exterior shots; it generates 60-second clip runs for about $0.40 each
- Full 30-second promo: $5 to $8 in model credits, 45 to 60 minutes first time through, 20 minutes per variant after
Restaurant video formats and when to use each
Pick the format before you open the canvas. The model choices differ.
Menu promo (15 to 30 seconds): 3 to 5 signature dishes, heavy on close-ups, light on people. Nano Banana Pro for stills, Seedance 2.0 for one motion moment, Kling 3.0 for a dining room establishing shot. The format most restaurants need first.
Opening announcement (30 to 45 seconds): exterior, interior atmosphere, dish previews, reservation invite. Kling 3.0 for atmosphere, Nano Banana Pro for stills, Seedance 2.0 for one hero moment (candle-lit table reveal, pasta being plated).
Delivery service ad (10 to 15 seconds, 9:16): appetite-forward, close-up food in the first 3 seconds. Seedance 2.0 leads with steam, cheese pull, or sauce pour. No atmosphere establishing shots; there's no time.
Seasonal special (15 seconds, social): a single visual anchor, the dish or the decorated space. Nano Banana Pro for the hero still, Kling 3.0 for atmosphere, text overlay for the date and offer.
Ambient mood reel (30 to 60 seconds, homepage or reservation platform): no dishes, no speech. Kling 3.0 handles this alone at $0.40 per clip. Candlelight flicker, window light at golden hour, soft focus on a full dining room.
The 4-step workflow
Step 1: Generate dish stills with Nano Banana Pro
Every restaurant video needs sharp, appetizing dish images. Even if your final output is motion video, you want stills for the freeze frames, end cards, and any platform where video autoplay is disabled.
Upload your actual dish photo if you have one (Nano Banana Pro can use it as a composition reference), or prompt from scratch. The prompt formula that works consistently:
[Dish name] on [surface material], [light source and direction], [camera angle], [mood or color grade], [one texture or action detail]
For an Italian trattoria opening promo:
Pappardelle al ragu on a wide shallow white ceramic bowl, matte linen tablecloth surface, warm candlelight from the left, 3/4 angle at 25 degrees, dark moody trattoria atmosphere, fresh basil leaf placed center, grated Parmigiano over the top
Generated on 8frame in approximately 14 seconds. The linen surface call prevented the default marble fallback. Candlelight direction produced warm specular highlights on the pasta edges without blowing the plate rim. Generate 3 variants per dish, pick the best one.
Cost: roughly $0.08 to $0.12 per still at Nano Banana Pro's standard credit rate. For a 5-dish menu promo, budget $0.50 to $0.60 for all your stills.
Step 2: Animate pours and action shots with Seedance 2.0
One or two motion moments transform a slideshow into a video. The highest-impact motion shots for restaurant content: wine or olive oil pours, sauce drizzles, steam rising from a fresh dish, and hand-service plating.
Seedance 2.0's multi-reference conditioning is the key feature here. Upload the dish still you generated in Step 1 as a reference image. The model animates motion that's consistent with the actual dish rather than generating a generic food video with the wrong plate.
Prompt for a pasta dish animation:
Close-up of a pappardelle al ragu dish, steam rising gently, sauce catching light as it settles, warm candlelit atmosphere, 3 seconds, vertical 9:16, no camera movement, food stays centered
Result on 8frame: Steam resolved as translucent wisps, not a flat white haze. The "no camera movement" instruction kept the dish stable, which matters for social formats where a drifting camera reads as shaky amateur footage. Generation time: approximately 65 seconds per clip.
Prompt for a wine pour:
Red wine pouring into a wide-rimmed wine glass, close-up, dark restaurant background, single soft light source from the right catching the liquid, motion continuous and smooth, 4 seconds, vertical 9:16
Result: Liquid physics read as realistic. The glass rim stayed in place across the full clip. Generate 3 variants at different pour speeds; pick the one that matches your edit pacing.
Cost: $0.45 to $0.65 per Seedance 2.0 clip.
Step 3: Generate ambiance cuts with Kling 3.0
Atmosphere shots contextualize the food. A menu promo that's only dish close-ups looks like a delivery app listing. Two or three seconds of dining room or exterior footage make it feel like a place.
Kling 3.0 is fast enough (approximately 60 seconds per clip) and cheap enough (about $0.38 to $0.42 per clip) to generate volume here. Prompt for specific atmosphere elements:
Dining room at service:
Warm Italian trattoria dining room, full tables in soft focus background, single candle flickering in foreground, brick walls, vintage wine bottles on shelves, amber light, 4 seconds, no people in sharp focus, cinematic shallow depth of field
Exterior entrance at dusk:
Exterior of a small Italian restaurant at dusk, warm window glow, stone facade, hand-painted menu board outside the door, one couple entering, cobblestone pavement wet from recent rain catching the light, 5 seconds
Result on both: Kling 3.0 handled the candlelight flicker without the strobe artifacts that appeared in earlier versions. The exterior wet cobblestone produced correct specular reflections without oversaturating. Generate 2 to 3 clips, cut the best one.
Cost: $0.80 to $1.25 for 2 to 3 Kling ambient clips.
Step 4: Add music and cut
You now have: 3 to 5 dish stills, 2 motion clips from Seedance, 2 to 3 atmosphere clips from Kling. Cut order for a 30-second opening promo:
| Seconds | Shot | Content |
|---|---|---|
| 0 to 4 | Exterior entrance | Location establishing |
| 4 to 10 | Dining room atmosphere | Candles, tables, ambiance |
| 10 to 18 | Dish stills (2 to 3) | Hero menu items |
| 18 to 24 | Motion shot | Pour or steam close-up |
| 24 to 28 | Final dish still or mood shot | The image to leave on |
| 28 to 30 | Text overlay | Name, location, reservation CTA |
For Italian trattoria content, a soft acoustic guitar or piano track with no vocals keeps music from competing with the visuals. Stay under -18dB. 8frame's video workflow templates on /workflows include a restaurant promo template with cut timing and text overlay presets already configured.
Routing by cuisine type
Surface materials, lighting, and atmosphere prompts need to match the cuisine category. The wrong defaults produce results that feel generic or tonally off.
Italian (trattoria, osteria): Dark surfaces, candlelight or warm amber, brick walls, linen tablecloths. Avoid bright white studio light; it reads as fast-casual.
Japanese (sushi, ramen, izakaya): White ceramic or black lacquer surfaces, cooler side lighting. For ramen, steam and broth surface tension are the key motion shots.
Mexican: Vibrant surfaces, warm daylight or terracotta tones, colorful garnishes. Overhead flat-lay works well for taco spreads.
Upscale/fine dining: Matte white plates, single-source directional light, no props beyond plate and cutlery. Motion shots: sauce quenelle, microgreens placement, server hand in gloves.
Fast casual (burgers, bowls): Overhead flat-lay with color contrast, bright diffuse light. Seedance 2.0 for cheese pulls and sauce drips. Aspect ratio 1:1 for Instagram, 9:16 for delivery app.
Italian trattoria opening promo walkthrough ($6 in compute)
Full cost breakdown for the 30-second opening promo example in this guide:
| Asset | Model | Variants | Cost |
|---|---|---|---|
| Pappardelle al ragu still | Nano Banana Pro | 3 generated, 1 used | $0.10 |
| Osso buco still | Nano Banana Pro | 3 generated, 1 used | $0.10 |
| Tiramisu close-up | Nano Banana Pro | 3 generated, 1 used | $0.10 |
| Red wine pour | Seedance 2.0 | 3 variants | $1.65 |
| Pasta steam rising | Seedance 2.0 | 2 variants | $1.10 |
| Dining room candle flicker | Kling 3.0 | 2 variants | $0.84 |
| Exterior entrance at dusk | Kling 3.0 | 2 variants | $0.84 |
Total model credits: $4.73. With a 30% buffer for discarded clips and regenerates: $6.00 to $6.50.
Total generation time on 8frame running clips in parallel: approximately 45 minutes. Final edit using the restaurant promo template: 15 minutes.
Pitfalls
Text on menu boards drifts. If you prompt an exterior shot with "menu board with specials written" or "chalkboard with today's pasta," the text will be illegible and often partially correct in a way that looks worse than no text at all. Keep menu boards empty ("hand-painted wooden sign, no legible text") and composite any real copy in post.
Hand-handling food authenticity. Seedance 2.0 and Kling 3.0 both struggle with hands in direct contact with food: hand-stretching pizza dough, tearing bread, plating with bare hands. Fingers distort, merge, or appear in wrong positions. Use tool-mediated actions instead: tongs placing pasta, a server's gloved hand setting a plate, a ladle pouring sauce. These render correctly because the hand-to-object relationship is mediated.
Wine and cocktail liquid physics. Liquid in a glass can glitch in two ways: the surface flattens and stops looking like liquid, or the pour stream bends mid-air. For wine pours in Seedance 2.0, always add "liquid physics realistic, no bending stream" to the prompt. For cocktail glass contents, avoid prompting "ice melting" because the model handles static ice better than state change. "Single large ice cube, condensation on glass" renders correctly; "ice cube slowly melting in amber liquid" produces artifacts in about 40% of generations.
FAQ
Do I need to disclose AI-generated food imagery?
It depends on the platform and how you're using it. For restaurant social media posts (Instagram, TikTok, Google Business), there's no legal requirement to disclose AI-generated imagery as of June 2026 in most jurisdictions, but it's worth adding a disclosure if the images could be mistaken for photos of your actual dishes (particularly for menu accuracy). Platforms like TikTok and Meta require the AI-generated content label on paid ads. For organic posts, the current guidance is to disclose if the content could mislead customers about what they'll receive. Check your platform's current policy before running paid campaigns, as the rules are updating frequently.
Does Google Business video content affect local SEO?
Yes, video on Google Business Profile is a positive signal for local search. Google's guidance as of 2026 treats video as a strong engagement indicator, and Business Profiles with video content appear more frequently in the "photos and videos" rich result in local packs. Video content doesn't need to be filmed; AI-generated video meets the platform requirements as long as it accurately represents the business and complies with Google's content policies. A 30-second ambient reel or a dish promo uploaded directly to your Business Profile is a practical local SEO move, particularly for competitive restaurant categories.
What aspect ratio works best for Google Business Profile video?
Google Business Profile accepts 16:9 (landscape) and 1:1 (square), with a minimum resolution of 720p and maximum file size of 75MB. For restaurant content, 16:9 is the better choice for Google Business because it fills the video player correctly and gives more horizontal space for establishing shots. For Instagram Reels and TikTok, 9:16 (vertical) is the required format. The practical workflow: generate your clips at 9:16 for social, then use Kling 3.0 to generate one 16:9 version of your dining room establishing shot specifically for the Google Business upload.
The workflow is four steps and under $8 in compute for a 30-second promo. Dish stills via Nano Banana Pro, motion via Seedance 2.0, atmosphere via Kling 3.0, cut together in under an hour.
Clone the restaurant promo workflow on 8frame, load your dish reference photos, and run Step 1 first. The stills take less than 2 minutes to generate across a 5-dish menu. Once you have the food looking right, the motion and atmosphere steps follow quickly. For a deeper look at AI video production across the full ecommerce and food service category, see the AI video for ecommerce complete 2026 guide.