Why it matters

Why Brands Generate Instead of Shoot

Great imagery used to require a great budget — a studio, a crew, and a new production for every visual. Here is what changes when images become something you generate.

No studio dependency

Every new visual used to mean booking a studio, setting up lighting, and coordinating a production team. Now it is a prompt and a few credits.

One photo, endless variations

A single product photo becomes dozens of unique images — different scenes, poses, and moods — so you can test what converts instead of guessing.

Any setting, any season

A beach in winter, a Parisian street, a rooftop at golden hour. Describe the location instead of scouting it — no travel, no weather delays.

Consistent across your catalog

Repeat the same lighting style and aesthetic across every product and every market. Your imagery stays on-brand as the catalog grows.

Step by step

How Image Generation Works

Every image starts from a prompt. Add reference images when you want to edit or restyle an existing photo; leave them out to create from scratch.

Reference photo uploaded for AI image generation

Step 01

Describe your image

Type what you want to see in the prompt bar. To edit an existing photo instead, upload up to 5 reference images — a product shot, an on-model photo, even a quick phone photo — and describe the change.

AI model and settings selected for image generation

Step 02

Pick your AI model and settings

Choose the AI model, aspect ratio, and resolution for the destination — product page, feed post, or banner. Turn on Enhance to let AI expand your prompt with photographic detail automatically.

Step 03

Run and refine

The exact credit cost is shown on the Run button, and your image lands in History within seconds. Reuse any result as the next run's input to change one thing at a time until it is exactly right.

Real results

Before & After

Each example below started as a single reference photo and one prompt — no set, no reshoot. This is the exact transformation the studio performs on every image-to-image run.

On-model photo used as the reference image — Reference photo

Clean product photo of the brown jacket generated by AI from the on-model reference — Generated product shot

Model photo used as the reference for an AI scene change — Reference photo

AI-generated image of the same model doing yoga in brutalist architecture — Generated result

Coverage

What Can You Create?

The same prompt bar covers every visual a fashion brand ships — from a clean packshot to a full campaign scene.

Product shots

Turn an on-model photo into a clean packshot on a white background, or restage a product without reshooting it.

Lifestyle imagery

Place your product in real-world scenes — street corners, apartments, rooftops — that would take a location shoot to produce.

Campaign visuals

Editorial-grade hero images with art direction baked into the prompt: mood, lighting, and styling on demand.

Model & pose variations

Change the pose, the framing, or the setting while keeping the same face and outfit across every variation.

Social content

Feed posts, Stories, and ad creatives in the right aspect ratio from day one — no cropping afterwards.

Art & concept visuals

Pure text-to-image creation for moodboards, lookbook covers, and brand concepts that don't exist yet.

Generation engines

What Are the AI Models?

The model picker decides which AI engine generates your image. All of them handle both text-to-image and image-to-image — they differ in fidelity, speed, supported formats, and credit cost.

Best quality

Nano Banana Pro

Google's flagship generation model. The highest fidelity, the most realistic lighting and texture, and 1K, 2K, and 4K output. Use it for final assets that go on product pages, ads, and campaigns.

Ratiosauto, 1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 21:9
Resolutions1K, 2K, 4K
Reference imagesup to 5

Widest formats

Nano Banana 2

Google's newest generation model with the widest format support — including ultra-wide ratios like 4:1 and 8:1 for website banners and hero strips, at up to 4K.

Ratiosauto, 1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 21:9, 4:1, 1:4, 8:1, 1:8
Resolutions1K, 2K, 4K
Reference imagesup to 5

Best instruction following

GPT Image 2

OpenAI's generation model. Excellent at following long, precise instructions, with selectable quality tiers so you can trade speed for fidelity per run: Low is the fastest and cheapest, Medium balances speed and detail, and High delivers the best visual fidelity.

Ratios1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 21:9
Resolutions1K, 2K, 4K
QualityLow, Medium, High
Reference imagesup to 5

Fast & affordable

Nano Banana

Google's standard generation model. The fastest and cheapest per run — ideal for exploring ideas and iterating on prompts before committing credits to a final render.

Ratiosauto, 1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 21:9
Resolutionsfixed output size
Reference imagesup to 5

A good workflow: draft with Nano Banana until the composition feels right, then re-run the winning prompt with Nano Banana Pro at a higher resolution.

Framing

Choosing an Aspect Ratio

The aspect ratio sets the shape of your image. Pick it based on where the image will live — cropping afterwards cuts into your composition, so it pays to decide upfront.

Auto

Matches your reference image's original framing

1:1

Product grids, marketplace thumbnails, and catalog tiles

4:5

Instagram feed posts and lookbook pages

9:16

Stories, Reels, TikTok, and vertical ads

16:9

Website heroes and presentations

21:9

Ultra-wide banners and headers

4:1 & 8:1

Full-width hero strips — Nano Banana 2 only

Every classic photographic format is supported too — 5:4, 4:3, and 3:2, their portrait counterparts 3:4 and 2:3, and the vertical ultra-wides 1:4 and 1:8. Auto is available on every model except GPT Image 2, and the ultra-wide ratios are exclusive to Nano Banana 2.

Output size

Picking a Resolution

Resolution sets the pixel size of the generated image. Higher resolutions cost more credits per run, so generate at the size the image will actually be displayed — you can always upscale a winner later.

1024px

1K

Prompt exploration and social feeds. Iterate cheaply until the composition is right, then spend credits on the final render.

2048px

2K

Product detail pages and paid ads, where shoppers zoom into texture, stitching, and print.

4096px

4K

Campaign heroes, lookbook covers, and print — maximum detail for the imagery that anchors a launch.

Nano Banana Pro, Nano Banana 2, and GPT Image 2 offer all three sizes. Nano Banana generates at a fixed output size, so the resolution picker appears only for the models that support it.

Direction

Writing Prompts That Work

The prompt is the whole brief: it plays the role of the photographer, the stylist, and the location scout at once. The models respond best to plain, specific sentences — no keywords, no magic words.

A strong prompt usually answers four questions. You don't need all four every time, but each one you answer takes a decision away from the AI and gives it to you.

Subject

Who or what is in the frame, and what are they doing? "A model in a charcoal wool coat, walking" beats "a coat".

Setting

Where does it happen? Name the location, the background, and the time of day: "a rooftop in SoHo, New York, at dusk".

Lighting & mood

"Natural daylight", "soft studio lighting", "golden hour", "nice shadowing" — light is what makes a generated image feel photographed.

Style & framing

The photography language: "editorial fashion photography", "e-commerce catalog shot", "wide angle", "full-body shot", "waist-up".

Text-to-image result generated from a short creative prompt

Text-to-image prompt

"Give me a visual like the Starry Night painting but in a fashion niche."

Image-to-image result generated from a reference photo and an edit prompt

Image-to-image prompt

"Remove the jacket from this model. She should be on a rooftop in her blue shirt. Only blue sky visible. Standing. Waist-up shot. Don't change her face, add nice shadowing."

When editing a reference image, state what must stay the same: "don't change her face or outfit". The AI protects everything you pin down.
Refer to your uploads directly — "the reference garments", "the woman in the black overcoat" — so the AI anchors on the right image.
Give constraints, not just wishes: "no one else in the frame", "only blue sky visible", "full-body shot".
Iterate one change at a time. Reuse the result as the next input and adjust the pose, then the background, then the light.
Not sure how to phrase it? Turn on Enhance — it rewrites your short prompt into a detailed photographic brief before generating.

Get it right

Common Mistakes to Avoid

Most disappointing results trace back to vague prompts or weak reference images, not the AI. These are the patterns we see most often — and what to do instead.

Don't

Write one-word prompts

"Jacket" or "fashion photo" leaves every decision to the AI — subject, scene, light, and style all become random.

Do

Describe the photo you'd brief

Subject, setting, lighting, style — a two-sentence brief in plain language is enough to control the result.

Don't

Upload blurry or cluttered references

Low-resolution photos and busy backgrounds hide the details the AI needs, so faces, fabric, and colors drift in the result.

Do

Use sharp, well-lit reference photos

A clear photo on a clean background — even from a phone — preserves the subject faithfully through every edit.

Changing everything at once? Split it into runs — pose first, then background, then lighting.
Details looking soft? Re-run the same prompt with Nano Banana Pro at 2K or 4K.
Result ignored part of a long instruction? Try GPT Image 2 — it follows complex, multi-step prompts most closely.

Keep creating

What to Do With Your Results

A generated image is usually the start of the pipeline, not the end. Every result in your History can flow straight into the other studios.

Image Upscale — scale your result up to 4K for print and campaign use.
Reframe — extend the image to a new aspect ratio for banners and stories.
Background Changer — swap the scene without regenerating the subject.
Video Studio — animate your image into a short fashion video.
Try-On Studio — dress a model in your garments with purpose-built virtual try-on.

Terminology

Text-to-Image, Image-to-Image, Prompt — What Do They Mean?

Text-to-image

Creating an image from a written description alone. You describe the scene; the AI invents every pixel from scratch.

Image-to-image

Editing or restyling an existing photo. The AI keeps what you tell it to keep and regenerates the rest according to your prompt.

Prompt

The written instruction that directs the generation — the subject, setting, lighting, and style you want in the final image.

Reference image

A photo you upload as input for image-to-image. Up to 5 per run — a product shot, a model photo, or a previous result.

FAQ

Frequently Asked Questions

How many credits does an image generation cost?

The cost depends on the AI model, resolution, and quality you pick. The exact price is always shown on the Run button before you generate, so there are never surprises.

What is the difference between text-to-image and image-to-image?

Text-to-image creates a brand new image from a written description alone. Image-to-image starts from up to 5 reference photos you upload — the AI edits, restyles, or rebuilds the scene while keeping what you tell it to keep. You can use both from the same prompt bar.

Is a phone photo good enough as a reference image?

Yes. A sharp, well-lit phone photo works fine as a reference — no DSLR, studio, or lighting rig required. The AI reads the subject, colors, and details from it; just avoid blurry or heavily compressed shots.

Which AI model should I choose?

Nano Banana Pro gives the highest fidelity and up to 4K output for final assets. Nano Banana 2 adds ultra-wide ratios like 4:1 and 8:1 for banners. GPT Image 2 follows complex instructions closely and offers quality tiers. Nano Banana is the fastest and cheapest for drafts and iteration.

How do I write a good prompt?

Describe the subject, the setting, the lighting, and the photography style in plain sentences. When editing a reference image, say what should change and what must stay the same — for example "change the background, don't change her face or outfit". The Enhance toggle can expand your prompt automatically.

Can I edit an existing photo instead of generating from scratch?

Yes. Upload the photo as a reference and describe the edit — swap the background, change the pose, remove a garment, or turn an on-model shot into a clean product photo. The AI keeps everything you don't ask it to change.

Can AI generate product photos from a single image?

Yes. Upload one photo of your product — even a phone shot or an on-model image — and describe the product photo you need: a clean packshot on a white background, a lifestyle scene, or a new angle. The AI rebuilds everything around the product while keeping it accurate.

How long does it take to generate an image?

Most images are ready in seconds and appear in your History as soon as the AI finishes. Higher resolutions and quality tiers take a little longer, and you can keep working while a generation runs in the background.

Do I own the images I generate?

Yes. Every image you create is fully yours to use on product pages, social media, ads, and any other commercial purpose.

Are my uploads private?

Yes. All uploads are encrypted and stored privately. Your images are never shared, sold, or used for training.