ChatGPT image prompts are the fastest way to turn a rough idea into a usable image: you write a short brief that specifies subject, style, lighting, camera, and aspect ratio, then iterate. The prompts below are copy-paste ready and built for Gemini / nano-banana style image models, where natural-language instructions and reference images work better than keyword piles.
This guide gives you concrete prompt examples, a repeatable structure, and the exact levers that change the output. If you want a library of ready-to-use prompts instead of writing from scratch, NanoPrompts is built for that.
What makes a ChatGPT image prompt different
ChatGPT image prompts are conversational. You can say "keep the face, change the background to a rainy Tokyo alley at night" and the model will attempt an edit rather than a fresh generation. That means your prompt should read like a director's note, not a tag list.
Three things matter most:
- Subject specificity. "A woman" is weak. "A 34-year-old woman with a blunt bob, freckles, and a linen blazer" is usable.
- Lighting language. "Soft window light from camera left, 45-degree angle" beats "nice lighting."
- Output constraints. Aspect ratio, lens, and depth of field control composition before you ever see the image.
If you are new to prompt syntax, start with the prompt structure and syntax guide and then come back to the examples here.
The 6-part prompt structure
Use this order. It maps to how image models parse instructions.
- Shot type — close-up, medium shot, wide establishing shot.
- Subject — who or what, with 2–3 defining details.
- Action or pose — what they are doing right now.
- Environment — location, time of day, weather.
- Lighting and mood — source, direction, color temperature.
- Camera and format — lens, aperture, aspect ratio, film stock or render style.
A prompt that hits all six reads like this:
Medium close-up of a 40-year-old fisherman with a salt-crusted beard and yellow rain jacket, hauling a rope on a wooden dock, overcast North Atlantic morning, soft diffused light from above, 85mm lens at f/2, 3:2 aspect ratio, muted teal and gray palette.
That is a complete brief. The model has no room to invent a different scene.
ChatGPT image prompt examples you can copy
Each example below is written for a specific outcome. Swap the subject and keep the structure.
1. Corporate headshot
Professional headshot of a 29-year-old man with short dark hair and a light stubble, wearing a navy suit jacket over a white shirt, neutral gray studio background, soft key light from camera left with a subtle rim light on the right, 85mm lens at f/2.8, 4:5 aspect ratio, sharp focus on the eyes, natural skin texture.
For more variations on this format, see the headshot prompts guide and the professional headshots prompts collection.
2. Cinematic film still
Wide shot of two strangers sharing a booth in a late-night diner, rain streaking the window behind them, warm tungsten interior light mixed with cold blue street light from outside, 35mm anamorphic lens, shallow depth of field, 2.39:1 aspect ratio, muted orange and teal grade, film grain.
The cinematic film stills prompts page has more framing options if you want to build a consistent look across multiple shots.
3. Product photography
Top-down product shot of a matte black ceramic coffee mug on a concrete surface, a single hard light from the upper right casting a crisp shadow, 100mm macro lens at f/8, 1:1 aspect ratio, no props, clean negative space on the left for text.
Product prompts live or die on lighting direction. The product photography prompts guide covers softbox, hard light, and reflective surface setups.
4. 3D collectible figure
A 3D collectible figure of a cyberpunk samurai with a neon visor and a katana, standing on a hexagonal base, studio lighting with a soft gradient background, vinyl toy aesthetic, slightly oversized head, 1:1 aspect ratio, high detail on the armor plates.
This style works well when you want a physical-object look rather than a photo. More examples are in the 3D action figure prompts library.
5. Editorial fashion portrait
Full-body editorial shot of a model in an oversized wool coat and leather boots, standing on a concrete staircase, harsh midday sun creating hard shadows, 50mm lens at f/4, 4:5 aspect ratio, desaturated palette with one red accent, confident posture, looking off-camera.
Fashion prompts benefit from a clear pose and a single color accent. The fashion editorial prompts page breaks down pose, garment, and location variables.
6. Fantasy landscape
Wide establishing shot of a floating stone monastery above a sea of clouds at sunrise, warm golden light hitting the eastern towers, cool blue shadows on the western side, 24mm lens at f/11, 16:9 aspect ratio, volumetric mist, epic scale with a tiny human figure on a bridge for scale.
Landscape prompts need a scale anchor. The fantasy landscape prompts guide explains how to place one without cluttering the frame.
How to write ChatGPT image generation prompts that hold up
ChatGPT image generation prompts fail for predictable reasons. Fix these before you blame the model.
Be explicit about what you do not want
Negative instructions work better when phrased as positive constraints. Instead of "no blur," write "sharp focus across the entire frame." Instead of "no people," write "empty street, no figures visible." The negative prompts guide covers when to use explicit negatives and when to rephrase.
Control the camera, not just the subject
Lens and aperture change the image more than adjectives do. A 24mm lens at f/11 gives you deep focus and wide context. An 85mm lens at f/1.8 gives you compression and background blur. State both.
Lock the aspect ratio early
If you need a 9:16 vertical for a story, say so in the first sentence. Models that default to square will crop your composition badly if you ask for vertical at the end.
Use reference images when the model supports them
Gemini / nano-banana style models accept image references. A prompt like "match the lighting and color grade of the attached reference, but replace the subject with a golden retriever" is more reliable than describing the grade in words. The multimodal creative workflow article covers reference-image workflows in detail.
Lighting, skin, and mood: the three levers that change everything
Most prompt improvements come from three places.
Lighting direction. "Soft light from camera left" is a different image than "hard light from above." The portrait lighting techniques guide has a full breakdown of key, fill, rim, and practical sources.
Skin texture. "Natural skin texture with visible pores and a slight sheen on the forehead" prevents the plastic look. The skin texture and makeup page shows how to keep detail without over-sharpening.
Mood and atmosphere. "Cold, quiet, slightly tense" steers color and contrast. The scene atmosphere and mood control article explains how to pair mood words with concrete lighting terms so the model does not guess.
Iterating: how to edit instead of regenerate
When an image is 80% right, do not rewrite the whole prompt. Give a targeted edit instruction:
Keep everything the same, but change the background to a rain-soaked Tokyo alley at night with neon signage, and shift the color grade warmer.
This preserves the face, pose, and lighting you already approved. For more on this workflow, see the realtime image editing guide.
If you are working on a portrait and the face drifts, lock it with a reference image and specify "preserve facial features exactly." The face swap prompts page covers identity preservation techniques.
Common mistakes and quick fixes
- Too many subjects. Fix: one primary subject per prompt, then add secondary elements in a follow-up edit.
- Vague style words. "Beautiful" and "stunning" do nothing. Fix: name a film stock, a lens, or a specific artist-adjacent look like "1970s Kodachrome."
- No aspect ratio. Fix: state it in the first line.
- Conflicting lighting. "Golden hour" plus "harsh midday sun" fights itself. Fix: pick one source and one direction.
- Overloaded prompts. If you have 12 clauses, split them into two generations and composite. The image composition techniques article covers when to split.
Where ChatGPT image prompts fit in a larger workflow
ChatGPT image prompts are best for fast iteration and edits. For a consistent series, you want a reusable template with fixed lighting and camera language, then swap only the subject. That is the same principle behind NanoPrompts: a library of ready-to-use AI image-generation prompts and templates for photo transformations such as portraits, headshots, style transfers, and 3D collectible figures.
If you want to go deeper on style consistency, the AI art styles complete guide and the style fusion techniques article show how to combine two looks without muddying the result.
FAQ
Do ChatGPT image prompts work with Gemini / nano-banana models? Yes. Natural-language prompts with subject, lighting, camera, and aspect ratio transfer well. Reference-image instructions are model-dependent, but the structure holds.
How long should a prompt be? One to three sentences for a simple subject, four to six for a full scene. If you exceed eight clauses, split the generation.
Should I include camera settings? Yes. Lens and aperture are the fastest way to control depth of field and compression.
Can I reuse one prompt across many subjects? Yes. Keep the lighting, camera, and format fixed, then swap the subject block. That is how you build a consistent series.
Try it on NanoPrompts
If you would rather start from a tested prompt than write one from scratch, browse the NanoPrompts library and copy a template for portraits, headshots, style transfers, or 3D collectible figures. Pick one, swap the subject, and generate.
