How to Write AI Image Prompts That Actually Work
By PromptWise Team
Your first AI image prompt usually looks something like this: “a woman in a coffee shop”. The model then fills in all the missing parts with its default settings. The result is another AI image with flat lighting, a centred composition, safe colours and a generic face nobody asked for, just like thousands of other images of women in coffee shops.
No special keyword is required to fix this. What changes the result is giving the model the decisions any photographer, illustrator or art director would make before starting work: what the subject is, what it is made of, where the light comes from, where the camera sits and what the image is for. This guide explains how to write those decisions into your prompts with a repeatable structure and the words that actually change results. There are also more than 30 example prompts you can copy, paste and adapt.
All of the prompts here work with the current generation of image models, including Nano Banana Pro, GPT Image 2.0, Seedream 5.0 Pro, Flux.2 Pro and the other image models in PromptWise, which you can test side by side in one app.
Start your 7-day trial and try every prompt in this guide on 40+ image, video and audio models. 7-day trial with 50 credits on every plan. Card required. Cancel anytime.
The short version
If you only read one section, read this one. A strong image prompt usually covers eight things, in roughly this order:
Subject → Medium → Style → Lighting → Composition → Colour palette → Mood → Detail and finish
Here is the difference that makes.
Weak prompt:
a woman in a coffee shop
Strong prompt:
Editorial portrait of a woman in her thirties reading a paperback at a window seat in a small Tokyo coffee shop, shot on a full-frame camera with an 85mm lens at f/1.8, late afternoon sun coming through the window and falling across the table, warm amber and cream tones, shallow depth of field with the street softly blurred behind her, quiet and unhurried mood, natural skin texture, magazine print quality
This second prompt does not contain any ‘magic words’. Instead it answers a series of questions that the model would normally answer for you. Who is she? What is she doing? Where is she? What camera took this photo? Where is the light coming from? What are the main colours in this image and how should it feel? Answering each of these questions removes a guess, and fewer guesses means fewer generic results.
Why most AI image prompts fail
Image models are trained on huge collections of images with captions. So when you input a short prompt, the model reaches for the average image that matches those few words. “Coffee shop” retrieves the average coffee shop. “Woman” retrieves the average face of a woman. Average lighting, average composition and average colour follow. This is why short prompts often look so generic. It’s not that the model can’t do better. The result is simply the average of everything the model has seen before.
The second reason is vague language. Words such as “beautiful”, “amazing”, “stunning” and “high quality” don’t describe anything you could photograph. They tell the model you want a good image, which it was already trying to make. A photographer can’t light a set for “beautiful”, but they can light it for “soft window light from the left with a white bounce card on the right”.
The third reason is that people change too many things at once. They get an image they don’t like, reword the entire prompt, get another image they like even less, and never figure out which words had any effect. Good prompting is more like keeping a lab notebook than writing creative stories. Keep what works, change one thing at a time and record the results.
The stacking method, layer by layer
Think of each layer in a prompt as one part of a stack. Not every layer is worth your time for every prompt (a simple sketch for a concept search doesn’t need a lens at all), but running down the list before hitting generate will catch the majority of the gaps found in below-average images. This typically takes around 30 seconds.
| Layer | What you’re deciding | How we’d word it |
|---|---|---|
| Subject | Who or what, doing what, plus the details that make it this one | ”a ceramicist in her sixties trimming a bowl on a wheel” |
| Medium | Photo? Illustration? 3D render, oil, collage? | “35mm film photograph”, “gouache illustration” |
| Style | A genre or a magazine look. Not a living artist | ”documentary editorial”, “1970s travel poster” |
| Lighting | Where it comes from, which way, how hard, what time | ”soft north window light from the left” |
| Composition | How close, how high, which lens, where things sit | ”medium close-up, eye level, 50mm, subject on the left third” |
| Colour palette | Two to four named colours (or a grade) | “sage green, terracotta and off-white” |
| Mood | How it should feel, in normal words | ”calm, focused, slightly nostalgic” |
| Detail and finish | Materials, textures, surfaces, the print feel | ”visible clay slip on her hands, fine film grain” |
Subject: be specific about the thing you actually want
This is usually where a prompt is thinnest. “A dog” could be any breed, any age, sitting anywhere. Ask instead for “a grey-muzzled black Labrador lying on a worn kitchen floor with his chin on his paws, looking up at the camera” and you’re limiting the possible images to a single one.
So, a good subject line tells the model four things: what it is, what it’s doing, how it’s this one and not another, and where it is. If you’re adding people to an image request, then include an age range, their clothes, their expression and their hands. Hands are the worst. Give them a job to do (“holding a mug with both hands”) and they tend to behave.
As for objects, you need to provide material, condition and scale. “A watch” is not a useful prompt. “A steel dive watch with a faded blue bezel and a scratched sapphire crystal, resting on a wet black rock” is a product shot with some history to it.
Medium: tell the model what kind of image this is
Creating images of the same subject in different media, such as a phone snapshot, a medium-format editorial photograph, a watercolour or a risograph print, yields four completely different images. The medium largely determines the look of the image. Most models, when left to their own devices, create images in a rather shiny and over-processed digital style, which is what most people think of when they talk of AI images.
Here are a few examples that we find very useful: “35mm film photograph”, “medium-format editorial photograph”, “candid smartphone photo”, “flat vector illustration”, “gouache illustration on textured paper”, “isometric 3D render”, “linocut print with two colours”, “charcoal sketch”, “studio product photograph on seamless paper”. All of these are quite narrow, and “illustration” is a category in itself. We would ask for “children’s book illustration in soft coloured pencil”, for example.
Style: reference genres and publications, not living artists
Skip “in the style of” + a living artist’s name. Describe the look instead, or point to a genre, era, publication type etc. “Interiors magazine spread”, “1990s skate magazine flash photography”, “Swiss modernist poster”, “mid-century travel advertisement”, “museum catalogue product shot”. Each of these is a strong visual concept that doesn’t copy one artist’s work.
This approach has two advantages. First, it’s more respectful than imitating another person’s work. Secondly, it’s more reliable to specify a genre than to say “in the style of…”. The model has seen thousands of images of a genre, so it can repeat the main visual features of that genre across a whole series.
Camera and lens language that changes the result
Learning the camera vocabulary is the fastest way to improve photographic prompts. You don’t need any equipment for this. Just knowing the difference each choice makes to the picture is enough. The model learned these differences from millions of photographs with their captions.
| Lens or camera phrase | What it does | Reach for it when |
|---|---|---|
| 24mm wide angle | Gets more of the room in; depth gets exaggerated and edges can stretch | Environmental portraits, interiors, landscapes |
| 35mm | Feels like you’re standing there. Documentary-ish | Street, lifestyle, behind-the-scenes |
| 50mm | Honest proportions, nothing fancy | Everyday portraits, food from the side |
| 85mm at f/1.8 | Kind to faces; background melts | Headshots, beauty, fashion portraits |
| 135mm telephoto | Squashes the background up behind the subject | Fashion, sport, anything far off |
| 100mm macro | Really close detail, razor-thin focus | Jewellery, skincare textures, food details |
| Medium-format camera | Lots of fine detail, smooth tones | Luxury product, editorial, beauty |
| Smartphone photo | Casual, everything sharp, believable | UGC-style ads, social posts |
| Disposable film camera with flash | Harsh flash, grain, instant nostalgia | Party scenes, youth brands, editorial |
Aperture values can also be used to describe images, with “f/1.4” or “f/1.8” shots featuring blurred backgrounds, while “f/8” or “f/11” shots have everything in sharp focus. Those are particularly suitable for landscape, flat lay and product images, where every last detail needs to be picked up by the viewer.
Then there’s camera angle. A scene shot from “eye level” looks neutral. Shooting from a “low angle looking up” at a person creates a powerful impression of them. A “high angle looking down” makes them look small or even vulnerable. An “overhead flat lay” is the most common way to shoot food and products. The “Dutch angle” is a special case where the horizon line is tilted to create tension. Put the angle near the start of the composition layer so it isn’t lost at the end of a long prompt.
Lighting terms that work
“Good lighting” is not an instruction. Specify the light source, its direction, whether it is hard or soft, and what time of day it is. These are the phrases we use most often.
| Lighting phrase | What you’ll get | Works for |
|---|---|---|
| Soft window light from the left | A gentle fall-off across the face. Feels natural | Portraits, food, lifestyle |
| Rembrandt lighting | That little triangle of light on the shadow-side cheek. Drama | Character portraits, moody editorial |
| Butterfly lighting | Light from above and in front; a small shadow under the nose | Beauty, glamour |
| Rim light with soft fill | A bright edge round the subject, shadows lifted | Product, sport, pulling things off dark backgrounds |
| Hard midday sun | Hard shadows, lots of contrast | Summer fashion, architecture |
| Golden hour backlight | Warm haze; hair and edges glow | Lifestyle, travel |
| Blue hour with sodium street lights | Cool sky, warm orange puddles of light | Urban night scenes |
| Neon practical lighting | Coloured light spilling from signs in shot | Night portraits, cyberpunk looks |
| Overcast diffused daylight | Flat and even. No shadows to fight | Fashion lookbooks, product on location |
| On-camera flash | Flat light from the front, background drops to dark | Party scenes, 90s editorial |
“Practical” in film terms is a light source that can be seen in the shot, such as a lamp, a candle, a phone or even a neon sign. If the practical you add to a scene also lights your subject then the resulting image will look motivated and real because the light has a reason to be there.
One tip that has served us well is to use exactly the same words for the lighting in all the prompts you write for a particular shoot, and then change only the subject and the pose in each image. Images made this way look as if they had all been shot on the same day.
Composition and framing
Composition is how an image is arranged within a frame. This can start with the basic shot size of an image such as extreme close-up, close-up, medium close-up, medium shot, full body or wide shot. This can then be modified by positioning the subject of the image within the frame, such as being centred within the shot, placed on the left third of the frame or even very small within the image with lots of negative space surrounding them. Composition can also be altered to include space for text to be added on top, i.e. “empty space in the top third of the image for a headline”.
Add depth by including foreground framing, shot through leaves of a fig tree in the foreground, shot framed by a doorway etc. Ensure that foreground, middle ground and background are clearly separated. Many models handle this well for landscapes and interiors.
Note that the image’s aspect ratio greatly affects its composition. This should be specified before you start generating images. For example, for Stories images that are 9:16, images should have their subjects near the top of the frame and be cropped very tightly. For banners that are 16:9, images should have their subjects off to one side of the frame and a large headline running along the other side. 4:5 suits most social feeds, so this is usually the safest bet.
Colour palette and mood
Specify the colours used in the image. “Warm colours” is too vague and gets misinterpreted most of the time. Instead specify the exact colours to be used such as “terracotta, ochre and cream”. Try to specify between two and four colours. The model will start to ignore some of the colours if you specify too many.
This is how you describe colour in terms of a ‘grade’ or look: “muted film colours with lifted blacks”, “high-contrast black and white”, “soft pastel colours with soft shadows”, “teal and orange grade”.
Mood, for all these kinds of images, can be indicated by a few words at the end of a prompt. One or two words are enough to get across the kind of feeling you are going for. Note that a prompt asking for several moods at once, e.g. “joyful, melancholy, energetic and serene”, is really asking for four different pictures.
Materials and texture: the difference between AI-looking and photo-looking
Materials, surface detail, skin. To stop the image looking plastic, focus on the materials, what the objects are made of, and what has happened to them.
Generic: “a leather bag”. Specific: “a hand-stitched, vegetable-tanned leather messenger bag with a darkened patina along the strap and scuffed brass buckles”.
Additional phrases to get more realistic images: “brushed aluminium with fine machining marks”, “raw selvedge denim with faded creases”, “hand-thrown stoneware with visible throwing lines”, “linen with natural slubs and soft wrinkles”, “condensation droplets on a cold glass bottle” and “natural skin texture with visible pores and fine lines”. For portraits: “natural skin texture” and “no retouching”.
Putting text in images
Several current models render text much better than previously. If you need to include text in your image, place the exact words in quotation marks and indicate where you would like them to be placed. Describe the type, too.
Minimalist poster for a jazz night, the words “LATE SET” in large condensed sans-serif capitals across the top third, “Friday 9pm, The Cellar” in small type at the bottom, a single saxophone silhouette in the centre, deep navy background with one warm yellow accent, screen-printed texture
It’s much easier to get short text correct than long. Try to limit headlines to 1-5 words. If one word in a quote gets misspelled, re-render the entire image instead of rewording the entire prompt. Be careful not to place two quoted text bits right next to each other. For large amounts of copy, generate the image with a large amount of space and then add the text in your design tool of choice.
Stop using quality keywords
Early image models often needed phrases like “8K, ultra-detailed, masterpiece, best quality, trending” in their prompts. Current models don’t, and these phrases can push an image towards an over-sharpened, over-saturated look.
Describe what makes the picture high quality in specific wording instead: “crisp weave visible in the wool”, “individual droplets on the glass”, “fine film grain”, “accurate hands with relaxed fingers”, “sharp focus on the eyes”. These phrases give the model something it can act on, and specific beats superlative every time.
Say what you want, not what you don’t
Many people try to fix problems by listing what not to include (no blur, no extra people, no text, etc.). Some interfaces have a separate negative prompt field, and that’s typically where these kinds of exclusions go. But inside the main prompt, listing what you don’t want can backfire, because mentioning “text” or “people” puts those ideas in front of the model.
Instead of expressing what you don’t want, try to rephrase it as a positive. So “no people” becomes “an empty street at dawn”, “no blur” becomes “everything in sharp focus, f/11”, and “not cartoonish” becomes “photographic, natural skin texture”.
Start your 7-day trial and run the same prompt on several image models at once. Plans from $19/month after the trial. See models and pricing.
Reference images and editing prompts
It’s also worth noting that many image models can use reference images, and some are built specifically for editing an existing image according to your written instructions. Tell the model what you’d like to keep from the reference and what you want to change.
Editing prompt example: “Keep the model, pose and outfit exactly the same. Change the background to a sunlit concrete rooftop with a soft city skyline, and match the lighting on her face to the new sun direction.” Saying what must stay the same matters as much as saying what should change.
For product work, a clean, well-lit photo of the real product as a reference is usually better than describing the product from scratch. Then prompt the scene, the light and the props around it. Our guide to AI product photography goes through that workflow step by step.
Model-by-model prompting tips
Below are starting points for how to write for each image model in PromptWise, not benchmark results. The only way to really know which is best for a given idea is to run one prompt on three or four of them and compare.
| PromptWise image model | We’d try it for | How to prompt it |
|---|---|---|
| Nano Banana Pro | Busy scenes, text in images, editing with instructions | Write it like a full brief, in sentences. Exact text goes in quotation marks. |
| Nano Banana 2 / Nano Banana 2 Lite | Drafts and variations | Use them for early rounds, then move the winning prompt to a heavier model. |
| GPT Image 2.0 | Detailed instructions and layouts | Plain sentences work, including where each thing goes. |
| GPT Image 2.5 Flare / GPT Image 2.5 Sunburst / GPT Image 1.5 | Seeing versions of the same idea | Don’t touch the prompt; compare outputs first, edit after. |
| Seedream 5.0 Pro / Seedream 5.0 Lite | Stylised, design-led, poster-type images | Spell out the palette, the layout and the type. |
| Seedream 4.5 / Seedream 4.0 | Variations, style tests | Handy for checking a look holds up on another model. |
| Realism Pro (Soul 2) | Photoreal people, lifestyle shots | Think photographer’s shot list: lens, light, skin texture. |
| Flux.2 Pro | Long, detailed prompts | Go long. Be precise. It pays off. |
| Grok Image | Quick concepts, a different take | Run it next to your main model and see how it reads the prompt. |
When you compare models, keep it fair by using the same word-for-word prompt on the first run of each. Also, judge them on the brief you wrote rather than on which picture you like more. If the brief asked for soft window light from the left and one model gave you a dramatic spotlight, that model missed the brief, however nice the image is.
Copy-paste AI image prompt examples
These prompts are written to be adapted. Swap the subject, keep the structure, and you’ll have a solid first draft for most briefs. Each one follows the stacking method, so you can see how the layers combine in practice.
Product and e-commerce
Studio product photograph. A matte black stainless steel water bottle stands on a slab of pale travertine. Soft diffused key light from the upper left, with a gentle rim light separating the bottle from a warm grey seamless background. Fine condensation droplets on the surface. Centred, with generous negative space above. 100mm lens at f/8. Clean, premium mood.
Lifestyle product photo: a linen tote bag hangs from a wooden chair in a sunlit Mediterranean kitchen. Morning light comes in from an open window on the right, and a bowl of lemons sits slightly out of focus in the foreground. Warm white, olive green and lemon yellow. 35mm lens. Relaxed weekend mood.
Overhead flat lay of a skincare routine on a damp white marble surface: a glass dropper bottle, a frosted jar and a folded cotton towel, with a few eucalyptus leaves. Soft even daylight, shot from directly above at f/11, natural shadows. Pale sage and white. Calm spa mood.
Portraits and people
Close-up portrait of a man in his seventies with a white beard and a flat cap, standing in a fishing harbour. Overcast diffused daylight. 85mm lens at f/2, so the boats behind him are softly blurred. Muted blues and greys, broken by his rust-orange jacket. Weathered skin texture with visible pores. Quiet and dignified.
Candid smartphone photo. Two friends laughing at a street food stall at night, lit warm by bare bulbs strung above the stall. Steam rising from the grill, a little motion blur on their hands, deep focus. Warm orange and teal colours. Energetic and authentic, natural skin texture.
Corporate headshot of a woman in her forties in a navy blazer against a plain light grey background. Soft butterfly lighting with a white bounce from below. 85mm lens at f/4, head and shoulders. Relaxed, confident smile. Natural skin texture, no heavy retouching look.
Food and drink
Side-angle food photograph of a bowl of ramen with a soft-boiled egg, chashu and spring onions. Steam shows up against a dark background. One soft light from behind and to the left makes the broth glisten. 50mm lens at f/2.8. Warm amber and deep brown palette, cosy late-night mood.
Overhead photograph of a sourdough loaf torn open on a floured wooden board: deeply caramelised crust, open crumb, scattered flour and a butter knife. Soft window light from the top of the frame. Rustic warm palette. Morning bakery mood.
Interiors and architecture
Wide-angle interior photograph of a small Scandinavian living room, with a low oak sofa, a wool throw and a paper floor lamp. Late afternoon sun casts long window shadows across pale floorboards. 24mm lens at f/8, everything in sharp focus. Oatmeal, warm white and black. Calm and uncluttered, like an interiors magazine spread.
Exterior architectural photograph of a concrete brutalist library at blue hour. Interior lights glow warm through tall windows and the wet pavement reflects them. Symmetrical composition from a low angle, 24mm tilt-shift look with straight verticals. Cool blue and warm amber. Quiet, monumental mood.
Fashion and lifestyle
Full-body fashion photograph: a model in an oversized camel wool coat walks across a zebra crossing in light rain. Shot from across the street on a 135mm lens, so the traffic behind is compressed and blurred. Soft overcast light. Muted city greys, with the camel coat as the only warm colour. Editorial lookbook style.
1990s skate magazine style photo of a teenager mid-ollie over a kerb in a car park at dusk. On-camera flash stopping the motion, fisheye lens, dark blue sky behind. High grain, saturated colours. Raw and energetic.
Social ads and thumbnails
UGC-style smartphone photo. A woman in her twenties holds a small serum bottle up to the camera in her bathroom, with a ring light reflected in her eyes and a slightly messy shelf behind her. Natural skin, no retouching. Casual and believable. Vertical 9:16 frame with space at the top for a caption.
YouTube thumbnail image of a surprised chef holding a burnt tray of cookies in a bright kitchen. Strong key light on the face, high contrast and saturated colours. Subject on the right third, with empty space on the left for large text. Expressive and playful.
Illustration and design
Children’s book illustration in soft coloured pencil on textured paper. A small fox carries a lantern through a snowy forest at night, its warm glow against cool blue snow. Wide composition, with the fox small in the lower left. Gentle, cosy mood.
Flat vector illustration of a cyclist riding through a city park. Limited palette of forest green, coral and cream. Simple geometric shapes, no outlines, plenty of white space. Modern editorial spot illustration style.
Retro travel poster for a fictional coastal town called “Port Avel”. Bold headline text “PORT AVEL” at the top in a wide serif typeface. Below it, a stylised lighthouse and sailing boats against a sunset gradient from coral to deep blue. Screen-printed texture, mid-century advertising style.
Characters and concept art
Character concept sheet of a young mechanic in a desert outpost, shown in front, side and back views on a plain light background. She wears oil-stained overalls, goggles pushed up on her forehead and a toolbelt with labelled pouches. Consistent proportions in all three views. Clean digital painting style.
Concept art of an abandoned greenhouse on a cliff edge at sunrise. Broken glass panels overgrown with vines, mist rolling in from the sea below. Painterly, with visible brush strokes. Soft pink and green palette. Wide establishing shot. Peaceful, and slightly eerie.
How to iterate like a professional
There’s no such thing as ‘write one prompt and accept it’ for professional creators. They run loops: generating small batches, picking the best image and working out why it’s better, making one change and generating again, usually over five or six rounds, until the prompt can be reused.
Keep a simple log of your prompts so you can track how your ideas develop. Write down the initial prompt, the model you used and a few words on how you feel about the results. Then change one layer at a time (light, lens or palette) and keep generating. If things get worse, you’ll know which change caused it. Over a few projects you’ll build up a library of tried-and-tested light and lens phrases.
It also pays to compare models early rather than late. A prompt that struggles on one model may be exactly right on another. In PromptWise, every plan includes all image, video and audio models, so you can run the same prompt on Nano Banana Pro, GPT Image 2.0 and Seedream 5.0 Pro in the same workspace and keep the winner. If a generation fails, the credits are refunded automatically.
A simple iteration checklist
- Write the full prompt using the eight layers.
- Generate a small batch on two or three models.
- Pick the image closest to the brief, not the prettiest one.
- Name the biggest gap: light, framing, subject detail, colour or mood.
- Change only the layer that controls that gap.
- Regenerate and compare with the previous best.
- Save the final prompt and the model in your log.
From still image to video
Good AI video often starts with a strong still image. Use an image model to create that single frame, then animate it with a video model such as Kling 3.0, Seedance 2.5 or Veo 3.1. Describe the camera move and what happens in the scene. You’ll have far more control over composition and colour than if you start from text alone.
The prompting changes when you add motion: you’re now describing a shot, not a picture. Our guide to cinematic AI video prompts covers camera movement, timing and model-by-model tips. In PromptWise, Flows lets you chain these steps on one canvas, from prompt to image to video, so you can reuse the whole workflow for the next project.
The most common image prompt mistakes
Most weak results come down to five habits: short prompts, fuzzy language, no description of the light, too many competing ideas and giving up after the first attempt. Fixing these five habits will have more impact than any trick or keyword.
Another frequent mistake is cramming two images into one prompt. “A chef cooking in a kitchen and a family eating at a table outside at sunset” asks for two scenes. Pick one moment and describe it well. If you need both, write two prompts. We cover these and other problems in more depth in 5 AI prompt mistakes killing your results.
| Mistake | What you get | Fix |
|---|---|---|
| Prompt is one short phrase | Something generic and average | Go through the eight layers |
| Vague adjectives (“beautiful”, “amazing”) | Not much changes, honestly | Describe light, lens and materials |
| Quality keywords (“8K, masterpiece”) | Over-sharpened, plasticky | Name the actual detail you want |
| Several scenes in one prompt | A muddled composition | One moment per prompt |
| Negatives in the main prompt | The thing you didn’t want turns up | Say what you do want |
| Changing everything between attempts | You can’t tell what worked | Change one layer at a time |
| Testing only one model | You miss a better result | Run the same prompt on three models |
Frequently asked questions
How long should an AI image prompt be?
Usually 40 to 80 words, enough to cover the eight layers. Very short prompts tend to return generic results, while very long prompts can bury important instructions in the middle. It’s best to front-load your prompt with the subject and composition, then add the finishing details at the end.
Do camera and lens names really make a difference?
Yes. Because the models learned from captioned photos, phrases like “85mm at f/1.8” or “24mm wide angle” carry real visual meaning: background blur, perspective and how much of the scene is visible. You don’t need to name a camera brand. The focal length, aperture and angle do most of the work.
Should I still add “8K” or “masterpiece” to my prompts?
No. Current models don’t need quality keywords, and they can make images look over-processed. Instead, describe the exact detail you want to see, e.g. “fine fabric weave visible” or “sharp focus on the eyes”.
Can I use an artist’s name in a prompt?
Rather than naming a living artist, it’s more respectful to refer to a genre, era, medium or type of publication, such as “1970s travel poster” or “interiors magazine spread”. Results are usually more consistent across a series with references like these.
Which AI image model is best?
It depends on the job. For text in images and editing with instructions, try Nano Banana Pro or GPT Image 2.0. For realistic people, go with Realism Pro (Soul 2). For stylised or design-led work, try Seedream 5.0 Pro. The best way to pick is to run the same prompt through several models and compare them in one PromptWise workspace.
How do I keep the same character across several images?
Write a fixed character description and reuse exactly the same words for the character’s age, hair, clothing and distinguishing details. If the model allows it, add a reference image of the character too. Keep the lighting phrase the same. For ongoing characters, PromptWise also has Influencer Studio, which is built for consistent on-screen characters.
How many credits does it take to learn this?
Credit use depends on the model you choose, and each model’s cost is shown in the app and on the models and pricing page. Every plan starts with a 7-day trial with 50 credits, which is enough to test the method on a few models. Paid plans start at $19/month for 275 credits.
Do I own the images I generate?
Yes. Everything you generate in PromptWise comes with a full commercial AI licence. You can use your images in ads, on product pages and on social media.
Start writing better prompts today
Good prompts aren’t long for the sake of it. They’re complete. They set out what a photographer or illustrator would decide before a shoot: subject, medium, style, light, framing, colour, mood and finish. Work through the eight layers, make one change at a time and see how different models read your prompt. You’ll get images that look intentional, not average.
PromptWise puts 40+ image, video and audio models in one web app on one subscription, so you can test every prompt in this guide on the models that suit it and turn your best stills into video. Plans start at $19/month, and you can compare plans and per-model costs on models and pricing.
Start your 7-day trial 7-day trial with 50 credits on every plan. Card required. Cancel anytime.