Multi-Image Reference Videos: Turn Your Photos into an AI Video Ad

Introduction

Text-to-video is great at inventing things. It struggles when you need it to show something real: your product, your presenter, your shop. Describe a blue takeaway cup in words and you'll get a blue takeaway cup, but not yours, and not the same one twice.

Multi-image reference video solves that. Instead of describing everything in words, you give the model photos of the product, the person and the place, and it builds a new video around them.

In this article we explain how it works, where it's useful and how to get good results. Then we put four of the leading models through the same test: three photos, one prompt, one coffee ad.

What Is a Multi-Image Reference Video?

Most AI video tools offer three ways to start:

Mode	You provide	What you get
Text to video	A prompt	A video invented from scratch
Image to video	One image and a prompt	That exact image brought to life; it becomes the first frame
Multi-image reference video	Several images and a prompt	A new scene featuring the people, products and places from your images

The key difference is between image to video and reference video. Image to video animates your photo: the framing, background and composition stay as they are, and the model adds motion.

A reference video uses your photos as a cast list. The model learns what the presenter, the product and the location look like, then directs a new scene with them, with new camera angles, poses and action, while keeping them recognisable. That's why it's also called reference-to-video: your images are references for identity and style, not frames to animate.

What Can You Use It For?

• Product ads. Photograph your product once and put it in any setting: in someone's hand on a sunny street, on a kitchen counter or at the gym. This is the classic use case, and the one we test below.
• A consistent presenter or character. Use the same photos of a presenter, mascot or brand character across several clips, so they look like the same person every time.
• Fashion and try-on. Combine a photo of a model with a photo of an outfit or accessory, and show it being worn in motion.
• Places and venues. Bring your café, hotel, shop or venue together with people and products, for a promo that's recognisably your place.
• Social content at scale. Keep your brand's look consistent across a run of Reels, Shorts and TikToks without a new shoot for each one.
• Storyboards and pitches. Turn a mood board into moving footage, and show a client what an idea will feel like before you film it.

How to Get Good Results

• One subject per image. A clean photo of the product, a separate photo of the person and a separate photo of the location work better than one busy photo with everything in it.
• Refer to your images in the prompt. Phrases like "the woman from the first reference image" or "the cup from the third reference image" tell the model which photo is which.
• Use the prompt for action and camera work. The images already show what everything looks like. Use the prompt for what happens: walking towards the camera, a sip, a smile, a slow tracking shot.
• Ask only for the shots you want. Models take camera directions literally. Ask for a close-up of the logo and you'll get one, sometimes as an abrupt cut. If you want a single smooth shot, say "one continuous shot".
• Start from sharp, well-lit photos. The model can only keep what it can see. Logos, text on packaging and faces come out best from clear, front-on images.

How to Make One in Animax

In Animax, open the AI video screen and choose the Multi-image Video tab:

1. Upload 1–10 images of your product, people, logo or location.
2. Write a prompt describing what should happen.
3. Choose Standard or Premium. Standard runs on Gemini Omni 1.1 Flash and Premium on Grok Imagine 1.5 (more on both below). If you pick a length, resolution or aspect ratio that only one of them supports, Animax switches to that model for you.
4. Generate. Every clip includes sound.

You can also do it from the Animax AI assistant. Attach your photos in the chat and ask for what you want, for example "make a short ad for my sneakers with these photos" or "show me wearing this jacket in Paris". The assistant decides whether to animate a single photo or build a new scene from all of them, and picks the model for you.

Multi-image videos are available on paid plans.

The Test: One Coffee Ad, Four Models

Until now, the Multi-image tab ran on Kling O3. For this update we tested four newer models to choose its replacement, giving each one the same three photos and the same prompt:

1. A woman in a red suit, the presenter.
2. A street café with orange awnings, purple flowers and a parked orange Vespa, the location.
3. A blue ZUS Coffee takeaway cup, the product.

The prompt:

A short, upbeat coffee commercial. The stylish woman in the red suit from the first reference image walks down the sunny street outside the café with the orange awnings, purple flowers and the parked orange Vespa from the second reference image, heading towards the camera while holding the blue ZUS Coffee takeaway cup from the third reference image. As she gets closer she takes a sip, then the camera moves into a close-up of the cup with the ZUS COFFEE logo clearly visible, and finally she lowers the cup and gives a big, warm smile to the camera. Cinematic, natural daylight, shallow depth of field, smooth tracking shot, realistic motion and skin. Audio: light upbeat music, footsteps on the pavement, gentle city ambience.

It's a demanding brief. Each model has to keep three separate references recognisable, combine them into one believable scene, follow four beats in order (walk, sip, close-up, smile) and keep the logo legible.

Every clip was generated at 720p, 16:9 and 8 seconds, with one take per model and no retries. As in our previous AI video model comparison, we recorded:

• Cost in Animax credits. For models that aren't available in Animax we show the equivalent credit cost, so the numbers are comparable.
• Generation time, from submission to a finished video.
• Quality score out of 10, for likeness to the references, prompt adherence, motion and overall ad quality.

Grok Imagine 1.5 — 232 credits | 720p | 102s gen | 8s clip | ⭐ 9/10

The only clip that looks like a finished commercial. Its street is the richest of the four, with the café, the Vespa, pavement tables and a zebra crossing. The camera moves naturally from a wide tracking shot to a mid shot as she sips, then settles on a warm close-up smile with the cup in frame.

Her face, the red suit and the cup stay consistent from start to finish, and the lighting has the polish of a real shoot. Instead of cutting to a separate shot of the cup, Grok shows the logo in the close-up of her sip, which reads more naturally. It was the second-slowest model to generate.

HappyHorse 1.1 — 224 credits | 720p | 140s gen | 8s clip | ⭐ 8/10

Cinematic and faithful. HappyHorse follows all four beats in order, including a bold product shot with the cup held up to the lens and the logo crisp, and the closing smile is warm and natural.

The scene is a little plainer than Grok's, with the café on screen only briefly before the camera moves in. It was also the slowest model in the test, and nearly as expensive as Grok.

Veo 3.1 Fast — 160 credits | 720p | 38s gen | 8s clip | ⭐ 7.5/10

The fastest generation in the test, and a good-looking clip: a sunny street that matches the reference, a natural walk and an energetic, believable laugh at the end.

It loses points on the product. In the close-up the cup is turned so the logo is only half visible, which matters in an ad. Veo also has the tightest limits of the four: three reference images at most, and a fixed 8-second length.

Gemini Omni 1.1 Flash — 160 credits | 720p | 50s gen | 8s clip | ⭐ 7.5/10

A very faithful recreation of the reference scene. The café, awnings, flowers and Vespa are all there, the woman and her suit match the photos closely, and the walk and sip look natural.

The weak spot is the product shot. Omni cuts abruptly to the cup filling the frame, then to a tight close-up of her face, which feels like an edit rather than a camera move. That's partly down to our prompt, which explicitly asked for a close-up of the cup; asking for one continuous shot should give a smoother result. For the price and speed, it's still a strong showing.

Results at a Glance

Model	Quality	Credits (8s, 720p)	Gen time	Reference images	Clip length	Aspect ratios
Grok Imagine 1.5	9/10	232	102s	Up to 7	Up to 15s	16:9, 9:16, 1:1, 3:2, 2:3
HappyHorse 1.1	8/10	224	140s	Up to 9	Up to 15s	16:9, 9:16, 1:1, 4:3, 3:4
Veo 3.1 Fast	7.5/10	160	38s	Up to 3	8s only	16:9, 9:16
Gemini Omni 1.1 Flash	7.5/10	160	50s	Up to 10	Up to 10s	16:9, 9:16

The usual caveat applies: every model got one attempt, and any model can have a good or bad day. Treat each result as a data point, not a final verdict.

What We Changed in Animax

The Multi-image tab now runs on two models:

	Model	Best for	Credits per second
Standard	Gemini Omni 1.1 Flash	Everyday product and brand clips: up to 10 images, 3–10 seconds, 720p or 1080p	20 at 720p, 30 at 1080p
Premium	Grok Imagine 1.5	Ad-quality results: up to 7 images, up to 15 seconds, square and other social formats, 480p or 720p	17 at 480p, 29 at 720p

Grok Imagine was the clear winner, and the only clip we'd run as an ad without changes, so it's our Premium model. It also supports clips of up to 15 seconds and square, 3:2 and 2:3 formats for social.

Gemini Omni is our Standard model. It recreated the reference scene closely, costs about a third less than Grok, finished in under a minute and takes up to 10 images, more than any other model here. Its one real slip, the abrupt close-up, is something careful prompting helps with.

HappyHorse came a close second on quality, but it was the slowest model and costs nearly as much as Grok without beating it. Veo 3.1 Fast is quick and good-looking, but three images and a fixed 8 seconds are too limiting for a multi-image tool.

Conclusion

Reference video is the missing piece for anyone who needs AI video to show something real. A few photos are enough to put your product, your presenter and your location into a scene that never existed, and to do it again with a new setting tomorrow.

Model choice still matters. In our test, Grok Imagine 1.5 produced the only clip that was ready to run as an ad, while Gemini Omni 1.1 Flash delivered a faithful, fast and flexible result at about two-thirds of the price. That's why Animax now offers both, and switches between them for you when your settings need it.

About Animax.ai

The clips in this comparison were generated with the same models that power Animax.ai, and generative video is only one part of the platform.

Animax is an all-in-one video creation platform for content creators, marketers and businesses. Alongside AI video, it lets you:

• Build editable video scenes from plain language. Describe what you want, and adjust every detail afterwards.
• Edit every element. Change colours, timing, text, motion and style without wrestling with a timeline.
• Combine AI with your own assets. Mix generated clips with your brand assets, footage and audio.
• Create at scale. Produce ads, social content, product videos and presentations from a single workflow.

Try it at animax.ai.