The workflow is: photograph the real product once against a plain wall, feed that shot in as a reference, and generate the lifestyle and ad variants from it. Text-only prompts are for backgrounds and drafts. On MakeViral an image is 1 to 4 credits depending on model and resolution, so 20 shots cost a few dollars.
What can a generated image do for a product page?
It can do the variants. The one thing it cannot do is the first shot.
Your product exists. Its color, its cap, its label and its texture are facts, and a model that has never seen it will invent all four. So the job is not to describe your product in a prompt. The job is to show the model your product and ask for a different scene around it.
Once you have that reference shot, the variants are where the cost and time went in the old workflow, and they are what a model does well:
- Backgrounds. The same bottle on marble, on wood, on a gradient, in a bathroom, on a shelf at a cafe.
- Lifestyle scenes. The product in a room with props, in the light of a particular time of day.
- Ad frames. The same product sized for a feed placement with space left for a headline.
- Seasonal sets. Autumn, holiday, summer, without shooting four times a year.
- Variant colors and sizes for a catalog where the physical difference is small.
That is most of the images a store needs, and almost all of the images an ad account burns through.
What is the workflow, step by step?
- Shoot one clean reference. The real product, a plain wall or a sheet of paper, soft daylight from a window, the label facing the camera and in focus. A phone is enough. Shoot three angles while you are there.
- Upload it as a reference image, not as a description. Nano Banana 2 accepts up to 14 reference images, so give it the three angles plus anything that shows the finish.
- Generate the scene, not the product. Your prompt describes the surface, the light, the props and the camera. The product comes from the reference.
- Generate in fours. MakeViral makes 1 to 4 images per run, and four gives you something to choose between for the cost of four credits at 1K.
- Check the label at full size. Every output, every time. This is the step people skip and regret.
- Keep the winner, regenerate the rest. One good frame per scene is the target. Cheap drafts are the point of Z-Image.
- Export at the size you need. 1:1 for the product gallery, 9:16 for the feed. Step up to 2K only when the image gets zoomed.
Which model should you use, and when?
Three models, three jobs. The differences that matter are the reference limit, the prompt ceiling and what happens to text inside the image.
| Z-Image | Nano Banana 2 | GPT Image 2.5 | |
|---|---|---|---|
| Credits at 1K | 1 | 2 | 2 |
| Credits at 2K | not available | 3 | 2 |
| Credits at 4K | not available | 4 | 4 |
| Reference images | none, text prompt only | up to 14 | one |
| Prompt ceiling | 900 characters | 4,000 characters | long prompts accepted |
| Typical wait | 5 to 15 seconds | 10 to 30 seconds | 20 to 60 seconds |
| Best at | Backgrounds, drafts, volume | Products from a reference, people, edits | Legible text inside the image |
Z-Image is the draft model. One credit, 1K only, text prompt only, and a 900-character prompt ceiling that forces you to write tightly. Use it to find the scene: generate eight background ideas for 8 credits, pick the one that works, then rebuild it properly with your product reference on another model.
Nano Banana 2, Google's image model, is the workhorse for a store. The 14-reference limit is the reason: you can show it the product from three angles, a color swatch, a previous photo you liked, and a room you want the mood of, all at once. It is also the one to use for edits, such as changing the surface under a product you already shot.
GPT Image 2.5, OpenAI's image model, is the one to reach for when words have to be readable in the picture: a price on a frame, a headline beside the product, a sign in the background. It accepts one reference image, so drive the product from that single best shot.
Open them from the photo page or go straight to Nano Banana 2, GPT Image 2.5 or Z-Image. Each model also has its own page with the limits written out: the Nano Banana 2 image generator and the Z-Image generator.
AI Product Photo Generator
Describe the product and the scene, or upload the shot you already have. Get back up to four finished images in under a minute, with no studio and no photographer.
What does a catalog of images cost?
Every number is a MakeViral credit price. A credit is about 7 cents on the 49.99 dollar plan for 700 credits, and about 6.7 cents on the 199.99 dollar plan for 3,000 credits.
| Job | Credits | On the 49.99 plan | On the 199.99 plan |
|---|---|---|---|
| One draft background, Z-Image 1K | 1 | 0.07 | 0.07 |
| One product shot, Nano Banana 2 1K | 2 | 0.14 | 0.13 |
| One product shot, Nano Banana 2 2K | 3 | 0.21 | 0.20 |
| 20 product shots, Nano Banana 2 1K | 40 | 2.86 | 2.67 |
| 20 product shots, Nano Banana 2 2K | 60 | 4.28 | 4.00 |
| 8 draft backgrounds, Z-Image | 8 | 0.57 | 0.54 |
| A 10-product launch, 6 images each, 1K | 120 | 8.57 | 8.00 |
Against a photographer, be honest with yourself about what you are comparing. A studio shoot gives you verified photographs of the real item, accurate color, correct fabric behavior and someone to blame when a shot is wrong. Generated images give you volume and speed. Get a quote for your actual catalog, then decide which images need the photographer and which are variants that do not.
The common answer for a small store: pay someone once for a clean set of reference shots of every product, then generate the scenes, the seasonal sets and the ad frames from those references all year. Pricing is on the pricing page. Cancel anytime, 14-day refund window on unused credits.
Twelve prompts you can copy
Each of these assumes your product reference image is attached. Replace the bracketed words with your own product. Keep them short on Z-Image, where the prompt ceiling is 900 characters.
1. Clean packshot
The attached product centered on a continuous white backdrop, soft studio light from the upper left, a faint contact shadow under the base, no props, square framing, sharp focus on the label.
2. Marble bathroom counter
The attached bottle standing on a white marble counter, morning daylight through a window on the right, a folded linen towel blurred in the background, shallow depth of field, vertical 9:16 framing.
3. Hand holding the product
A hand holding the attached product at chest height, phone camera look, a kitchen out of focus behind, natural window light, slight motion in the frame, vertical framing, no text.
4. Wooden table outdoors
The attached product on a weathered wooden table outdoors, late afternoon sun, long soft shadows, a few green leaves in the corner, warm color, square framing.
5. Flat lay with props
Top-down flat lay of the attached product on a sand-colored paper background, a small ceramic dish and two dried stems beside it, even diffused light, generous empty space on the right.
6. Kitchen morning scene
The attached package on a kitchen counter next to a cup of coffee and an open notebook, warm morning light from a window behind, steam visible, shallow depth of field, horizontal framing.
7. Gradient studio with reflection
The attached product on a glossy dark surface against a soft blue to charcoal gradient, one rim light from behind, a clean mirror reflection below, centered, square framing.
8. Scale reference
The attached product standing beside a standard coffee cup on a plain grey surface, even light, straight-on camera angle at product height, both objects fully in frame, neutral color.
9. Seasonal set
The attached product on a dark green surface with pine needles and a small red ribbon nearby, warm low light from the left, winter mood, no snow, vertical framing.
10. In-use lifestyle
A person in a light sweater using the attached product at a bathroom sink, seen from the shoulders down, face not visible, soft daylight, realistic skin, vertical 9:16 framing.
11. Ad frame with a headline space
The attached product on the lower third of a warm beige background, plenty of empty space above it for a headline, soft even light, one subtle shadow, vertical 9:16 framing.
12. Texture detail
Close macro of the surface texture of the attached product, filling the frame, raking side light to show the finish, neutral white balance, no background visible.
Two habits make these work better. Name the light before the props, because light decides whether an image reads as a photograph. And ask for empty space explicitly when the image is going into an ad, because a model will otherwise fill the frame and leave you nowhere to put a headline.
For the creator-style version of these shots, the shot that looks like a customer took it, the UGC photo generator is the same workflow with a person in the frame.
How do you keep the label right?
Small text is where these models fail. Not sometimes: predictably. Treat every generated label as wrong until you have read it.
- Zoom to full size and read the brand name, the variant name and any number on the pack. A misspelled brand name on a product page is worse than no image at all.
- Use GPT Image 2.5 when the text is the point. It is the model to reach for when a label, a price or a sign has to be legible.
- Crop rather than fix. If only the label is wrong, a tighter crop that cuts it out is faster than five more generations.
- Keep one real photograph per product in the gallery, and make it the one where the label is readable. That is the shot a careful buyer zooms into.
- Check ingredient panels, certifications and claims by hand. Never let a model generate those at all. A generated ingredient list is a compliance problem, not a design choice.
What should you never generate?
Three lines worth keeping.
Anything a buyer will compare against the parcel. Color, finish, size, fabric drape. If the generated image is more flattering than the product, you have bought a return and a bad review.
A person who looks like someone real. Do not steer a model toward a celebrity, a creator or a customer. Use the model's own faces.
Proof. A screenshot of results, a certificate, a lab panel, a review. A generated version of any of those is a fabrication regardless of how the image was made.
Everything else, the backgrounds and the scenes and the seasonal sets, is a legitimate use and it is where the money is anyway.
Nano Banana 2 Image Generator
Google's image model, in a browser tab, with the reference-image dropzone that makes it useful: up to 14 photos in, one new image out.
What to read next
The AI photo generator hub covers every model, the aspect ratios and the reference-image workflow in one place, and the AI product photo generator is the screen that does the reference-driven product shots. When the product page is ready and you want it moving, URL to video builds a vertical video from the product URL itself, and how to make UGC ads with AI covers the ad that the product still goes into. For organic reach off the same images, see the TikTok slideshow strategy.