ChatGPT Image Prompts: 50 Prompts for Realistic, Cartoon and Product Images

The short version: ChatGPT's image tool (built on GPT-4o) can produce convincing realistic photos, clean cartoon art, and usable product shots if you feed it specific camera language, lighting terms and composition instructions instead of vague ideas. Below are 50 prompts split into those three categories, plus the one thing nobody tells you about why "photorealistic" still isn't the same as real.

I run product images through ChatGPT most weeks now, for my own site and for two clients who sell physical goods. I stopped booking a photographer for small product refreshes about eight months ago, and I've saved roughly £1,800 in shoot fees this year alone. That's not a small number for a solo consultant. But I've also had images rejected by a client because a "realistic" hand had six fingers, and I've had a cartoon mascot come out looking like a different brand's mascot because I didn't specify style tightly enough. Both things are fixable with better prompts. Here's how.

How ChatGPT image generation works, in one paragraph

When you type an image request into ChatGPT, it's not pulling a photo from a database, it's generating pixels from scratch based on your text, using patterns learned from millions of images. That means the exact same prompt typed twice can give you two different results. It also means the more specific your words, the less room the model has to guess wrong. Vague prompt ("a woman drinking coffee") gives you generic stock-photo mush. Specific prompt (camera, lens, lighting, angle, mood, setting) gives you something with texture and intention. That's the entire skill here.

Realistic image prompts (17)

These work best when you write them like a photography brief, not a description. Include camera type, lens, lighting direction and time of day.

• "A candid photo of a woman in her 50s laughing at a kitchen table, morning light through a window, shot on a 50mm lens, shallow depth of field, natural skin texture, no retouching look"
• "A UK high street coffee shop interior, rainy afternoon, warm tungsten lighting, shot on 35mm film, muted colour grade"
• "A freelance consultant working from a home office, laptop and notebook on desk, soft window light from the left, documentary photography style"
• "An elderly man reading a newspaper on a park bench in autumn, shallow focus, golden hour backlighting"
• "A busy London street at dusk, motion blur on buses, neon reflections on wet pavement, shot on a wide-angle lens"
• "A close-up of hands typing on a mechanical keyboard, desk lamp lighting, shallow depth of field, product photography realism"
• "A small business owner standing in front of her shop, arms crossed, confident expression, overcast daylight, portrait lens compression"
• "A hiking trail in the Scottish Highlands, misty morning, wide-angle landscape shot, muted green and grey tones"
• "A team meeting in a modern office, five people around a table, natural window light, candid mid-conversation moment"
• "A close-up of a coffee cup with steam rising, top-down flat lay, soft diffused daylight"
• "A father and daughter cooking together in a kitchen, warm evening lighting, genuine unposed expressions"
• "A city commuter checking their phone on a train platform, early morning light, shot on a 85mm portrait lens"
• "An artisan baker kneading dough, flour dust in the air, side lighting through a bakery window"
• "A dog running on a beach at sunset, motion frozen, warm orange and pink sky, shallow depth of field"
• "A close-up portrait of a woman with grey hair, natural wrinkles visible, soft studio lighting, no airbrushing, editorial style"
• "A farmer's market stall with fresh vegetables, midday sunlight, shallow depth of field on the produce"
• "A remote worker on a video call, laptop screen glow lighting half their face, evening ambient light in the background"

One honest note here: even with all that specificity, "realistic" ChatGPT images still have tells if you look closely. Text on signs in the background often garbles into nonsense letters. Hands, especially holding objects, still occasionally come out wrong. Teeth in wide smiles sometimes look too uniform. No prompt trick fixes this reliably yet, it's a known limitation of the model, and anyone telling you they've "cracked it" with a magic phrase is overselling. Always zoom in on hands, text and teeth before you publish anything, every single time.

Cartoon and illustration prompts (16)

Cartoon prompts need a named style reference or the model defaults to a generic "digital art" look that's used everywhere. Naming a specific animation studio style, decade, or art movement gets you something distinctive.

• "A cartoon illustration of a woman running a business from her laptop, flat vector style, bold outlines, warm colour palette"
• "A Studio Ghibli style illustration of a small cottage in a green valley, soft watercolour textures, warm afternoon light" (if you want the exact wording that gets closest to that look without triggering a copyright block, I've written up the full breakdown in my Ghibli-style prompt guide)
• "A 1990s Saturday morning cartoon style illustration of a dog chasing a ball in a park, bright primary colours, thick black outlines"
• "A minimalist line-art illustration of a person meditating, single continuous line style, soft pastel background"
• "A Pixar-style 3D character of a small robot watering plants, soft rounded shapes, warm studio lighting"
• "A retro 1970s travel poster style illustration of a coastal town, flat colour blocks, geometric sun in the sky"
• "A children's book illustration of a fox reading under a tree, watercolour texture, soft muted tones"
• "An isometric flat design illustration of a home office setup, desk, chair, plants, muted pastel colour scheme"
• "A comic book panel style illustration of a superhero landing on a rooftop, bold ink outlines, halftone shading"
• "A cute chibi style character of a cat wearing glasses working on a laptop, oversized head, simple background"
• "A hand-drawn sketch style illustration of a busy street market, loose ink lines, light watercolour wash"
• "A flat vector illustration of a team celebrating a product launch, confetti, bright brand colours, no gradients"
• "An anime style illustration of a girl standing on a cliff overlooking the sea, dramatic wind in her hair, soft cel shading"
• "A whimsical storybook illustration of a treehouse village, warm sunset colours, soft rounded shapes"
• "A bold pop-art style illustration of a coffee cup, Ben-Day dots, high contrast primary colours"
• "A cartoon mascot of a friendly owl holding a lightbulb, simple flat design, suitable for a logo, transparent background"

Product image prompts (17)

This is where I've replaced a photographer, and it's the category most small business owners ask me about. The trick is treating ChatGPT like a studio photographer's assistant, not an artist. Be exact about background colour, angle, and lighting setup.

• "A white ceramic mug on a plain white background, studio lighting from above, soft shadow beneath, e-commerce product shot style"
• "A leather handbag photographed at a 45-degree angle on a light grey background, soft even studio lighting, no harsh shadows"
• "A skincare bottle on a marble surface, natural window light, minimal styling, shallow depth of field"
• "A pair of trainers floating against a plain white background, dramatic side lighting, high contrast, e-commerce style"
• "A candle in a glass jar on a wooden table, warm ambient lighting, lit wick, cosy styling, lifestyle product photo"
• "A stack of three notebooks on a white background, top-down flat lay, soft diffused lighting, minimal shadows"
• "A bottle of olive oil on a rustic wooden surface with scattered herbs, warm natural light, editorial food photography style"
• "A wristwatch photographed close-up on a dark background, dramatic rim lighting highlighting the metal detail"
• "A tote bag hanging on a hook against a neutral beige wall, soft natural light, lifestyle context shot"
• "A protein powder tub on a plain background with a scoop of powder beside it, studio lighting, clean commercial style"
• "A pair of sunglasses on a sandy beach surface, bright natural sunlight, shallow depth of field"
• "A cosmetics palette open showing colour swatches, top-down angle, soft even studio lighting, no colour cast"
• "A wine bottle with label facing forward on a dark slate background, single spotlight from above, dramatic shadow"
• "A children's toy on a white background, bright even lighting, no shadows, e-commerce catalogue style"
• "A wooden cutting board with fresh bread and a knife, natural side light, rustic kitchen styling"
• "A phone case product shot on a plain background with the phone at a slight angle, soft studio lighting, reflective surface below"
• "A set of scented soap bars stacked on a linen cloth, soft diffused daylight, minimal shadow, spa-style product photography"

The step-by-step I use for a product shoot replacement

Here's exactly what I do when a client asks for a batch of product images without booking a photographer:

• Step 1: I take one real photo of the actual product on my phone, even a bad one, and upload it to ChatGPT alongside the prompt, asking it to keep the product's real shape, colour and label accurate while changing the background and lighting.
• Step 2: I write the prompt in this order every time: subject, background, lighting direction, camera angle, mood, then any brand colour constraints.
• Step 3: I generate 3 to 4 versions of each shot, because the model varies its output even from an identical prompt, and one version is usually noticeably better than the rest.
• Step 4: I zoom in at 200 percent on any text, logos, or fine details before approving anything for use.
• Step 5: I run the final image through a second pass asking ChatGPT to "clean up small imperfections while keeping the product accurate" rather than regenerating from scratch, which preserves the parts that already worked.

That whole process takes me about 20 minutes per product now. A studio shoot for the same output used to cost a client between £150 and £400 depending on the number of shots, plus a half-day of my time coordinating it. That's the real saving, and it's why I now recommend this route to almost every small ecommerce client I work with, unless the product needs to be shown in real-world use by a real person, in which case a genuine photo still wins every time.

Where this gets uncomfortable

Here's the bit most guides on this topic skip. A growing number of businesses are using AI-generated "customer" photos and fake "before and after" images to fabricate social proof, dressing up generated people as real testimonials. I've been asked to do this by two separate clients this year, and I said no both times. The Advertising Standards Authority in the UK has already flagged AI-generated imagery used deceptively in ads as a compliance risk, and platforms including Meta and Amazon have started requiring disclosure labels on AI-generated product and lifestyle images in certain categories. Using ChatGPT to build a product shot of your own actual product is fine. Using it to fake a customer holding your product, or to fake a review photo, is the kind of thing that gets accounts suspended and, increasingly, gets regulatory attention. Use the tool for what it's good at: mockups, concept art, product staging, and illustration. Don't use it to manufacture trust you haven't earned.

A few things that will save you time

If you're editing an existing photo rather than generating a new one from scratch, the prompt structure is different again, and I've written a separate breakdown of the exact wording that works for that in this photo editing prompts guide. If you're planning to copy any of the 50 prompts above and adapt them for client work, it's worth reading how to do that without landing on someone else's copyrighted style or a brand's protected imagery, which I cover in this piece on adapting prompts safely. And if you want more unusual, less obvious ideas beyond the standard categories, I tested dozens of odd ones and kept the best in this list of image prompt ideas worth trying.

Also worth knowing before you generate 50 images in one sitting: there are practical daily limits on how much you can send ChatGPT depending on your plan, and I've laid out the real numbers in this breakdown of daily prompt limits, so you're not caught out mid-project.

Save the prompts that work

The single biggest time-waster I see people make is rewriting the same prompt structure from memory every week and getting worse results because they've dropped a detail that mattered. Once a prompt gives you a result you'd use again, save the exact wording somewhere searchable rather than trusting your memory. I keep mine organised by category, the same way the 50 above are split, and I've written up the exact system in how to build a personal prompt library. It's the difference between a five-minute image request and a twenty-minute one where you're re-deriving lighting terms from scratch.

Related tool: my free tool that writes the prompt for you.

If you want the full set, start at the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.

Related reading: chat gpt cartoon image.

See also How to Write Better AI Image Prompts (What I Learned After 200+ Bad Ones), which picks up where this leaves off.

Frequently asked questions

Can ChatGPT generate truly realistic photos, or does it always look AI-made?

ChatGPT can produce convincing realistic images when you write detailed, camera-specific prompts, but it still has weak spots: hands, background text, and teeth are the three places it most often gives itself away. Always zoom in on those before using an image publicly.

What's the best way to write a ChatGPT prompt for a product photo?

Structure it like a photography brief: name the product, the background colour or surface, the lighting direction and type, the camera angle, and any brand colour constraints, in that order. Uploading a real photo of your actual product alongside the prompt keeps shapes and labels accurate.

Why do the same ChatGPT image prompts give different results each time?

ChatGPT generates images from scratch based on patterns rather than retrieving a fixed file, so identical wording can produce different outputs on separate runs. The fix is generating 3 to 4 versions of anything important and picking the strongest one rather than expecting a single perfect result first try.

Is it okay to use ChatGPT images for customer testimonials or "real customer" photos?

No. Using AI-generated images to fake customer photos, reviews, or before-and-after results is a growing area of regulatory scrutiny in the UK and on major ad platforms, and it can get accounts suspended. Use ChatGPT for your own product shots, mockups and illustrations, not to fabricate proof involving real people who don't exist.

Official documentation

• OpenAI's own platform documentation