Product photo prompts. For Nano Banana.
Build a subject-first prompt for Nano Banana, Midjourney, Imagen, or Stable Diffusion: lighting, camera, color, mood, style reference. Browser-only.
Subject + setting + light + camera.
Paste into Nano Banana or Midjourney.
Three reference outputs.
Subject: a frosted-glass skincare bottle on a travertine plinth. Setting: cream studio backdrop, soft drop shadow. Lighting: soft north-window, gentle wraparound. Camera: 3/4 hero, slightly above eye level, 50mm at f/4. Color: cream, terracotta, soft moss. Mood: warm-organic, minimal. Style: Aesop minimal. Aspect: 4:5. Negative: no hands, no text, no watermark.
Subject: kraft coffee bag with dark-roast beans scattered around the base. Setting: flat-lay on a walnut wood surface. Lighting: overcast diffuse, shadowless. Camera: overhead 90°, 50mm at f/5.6. Color: cream, espresso brown, kraft. Mood: warm-organic. Style: Bon Appetit. Aspect: 1:1. Negative: no extra props, no oversaturation, no AI hyper-gloss.
Subject: a brushed-gold signet ring. Setting: matte black sculptural plinth, abstract studio. Lighting: single-source dramatic, deep unlit shadow. Camera: extreme close-up macro, 100mm at f/8 for sharpness across the band. Color: matte black, brushed gold, ivory. Mood: luxurious, minimal. Style: Vogue beauty. Aspect: 4:5. Negative: no fingers, no plastic finish.
A prompt is a brief for a photographer who reads minds.
A product-photo prompt that produces usable output on the first generation does six things in order. Subject first - the product and its material. Setting second - the surface or environment. Lighting third - one named lighting setup, not three adjectives stacked. Camera fourth - angle plus lens-language for depth of field. Color and mood fifth. Style reference last. The negative prompt sits at the end as a guard rail. Models like Nano Banana and Midjourney v7 weight tokens at the front of the prompt more strongly; lead with the subject and the setting, not the style.
Subject-first ordering
The first 30 tokens of any image-model prompt do most of the work. Lead with the product itself and its key material attribute (frosted glass, brushed metal, kraft paper, ribbed ceramic). Google's Nano Banana (Gemini 2.5 Flash Image) and Midjourney both score subject-first prompts higher on subject fidelity. A prompt that opens "in the style of Kinfolk magazine, soft and slow, a frosted skincare bottle..." will produce a generic Kinfolk image with a bottle inserted; a prompt that opens "a frosted-glass skincare bottle on a travertine plinth, in the style of Kinfolk..." will produce the bottle the brief asked for.
Lighting language
Image models understand lighting language drawn from photography textbooks. North-window light, golden-hour warm directional, three-light key-fill-rim, single-source dramatic, overcast diffuse - all of these map to recognizable training-data clusters. Ambiguous adjectives (beautiful lighting, perfect light) do not. Specify the direction (front, side, top, three-quarter), the quality (soft, hard, diffuse), and one named lighting style. Stability AI's SD3.5 and FLUX both reward this specificity.
Camera language - not Photoshop language
Camera-language tokens (50mm prime, f/4, 100mm macro, three-quarter hero angle) work; post-production language (sharpened, color-graded, retouched) does not. The model is trained on real photographs, so it already knows what a 50mm at f/4 looks like. Asking for "sharpened with high clarity" produces a brittle, AI-tell texture. Asking for "shot on a 50mm prime equivalent at f/4" produces realistic depth-of-field falloff. Lens specificity also helps the model produce believable optical artifacts (slight vignetting, chromatic aberration at edges) that read as photographic rather than rendered.
Color language
Color words work better than hex values for most image models. "Off-white, terracotta, soft moss-green" produces predictable output; "#F4F1EA, #C4623F, #8E9B7E" tends to produce literal color blocks that read as design-software output. The exception is brand-color-led work where exact hex matching matters; in that case, generate with color words, then use a color-grading pass in Photoshop or an AI-edit model like OpenAI's gpt-image-1 to dial in the exact swatch.
Negative prompt as guard rail
Most product-photo failures fall into a small set: extra hands or fingers, AI-tell hyper-gloss, text or watermark hallucinations, distorted product geometry, oversaturated color. List these as negative prompts at the end. Stable Diffusion supports a separate negative_prompt parameter; Midjourney uses --no; Nano Banana accepts negative-prompt instructions in natural language at the end of the brief. The negative block is rarely longer than 10 items; if you need 20, the positive prompt is too vague.
Aspect ratio matters
Aspect ratio drives composition decisions. A 4:5 prompt that asks for a wide environmental shot will crop awkwardly; a 16:9 prompt that asks for a macro detail will leave too much empty background. Match the aspect to the use case: 1:1 for Instagram grid, 4:5 for vertical PDP and Instagram feed, 9:16 for Stories and Reels, 16:9 for hero banners and YouTube thumbnails. Schema.org Product guidance recommends a 1200×630 OG image, which maps to roughly 16:9.
Related tools: Landing page prompt generator for the page the photo lives on. Ad creative prompt generator for paid-traffic versions. Open graph preview for the social card. Favicon generator for the brand mark. Brand identity for the system the photos sit inside.
Five answers.
What is a product photo prompt generator?
A tool that assembles a structured prompt for AI image models like Nano Banana, Midjourney, Imagen, or Stable Diffusion. The prompt orders subject first, then setting, lighting, camera angle, color, mood, style reference, and aspect ratio so the model gets the most important information at the front of the context window.
Which image models does this prompt work with?
Google's Nano Banana (Gemini 2.5 Flash Image), Nano Banana Pro, Midjourney v7, Imagen 4, Stable Diffusion 3.5, FLUX, Recraft, and Ideogram. The prompt structure is model-agnostic; minor tweaks to the aspect ratio tag may be needed (Midjourney uses --ar 1:1, Nano Banana uses natural language).
Why is subject-first ordering important?
Image models attend more strongly to tokens at the start of the prompt. Lead with the product itself (a frosted skincare bottle on a travertine plinth), then add the setting, lighting, camera, and style. Style references at the front of the prompt push the model into a generic aesthetic and lose the specific subject.
What is a negative prompt and when do I need one?
A negative prompt lists what should not appear in the image. For product photography it typically excludes hands, fingers, text, watermarks, blurry-low-quality, distorted-product, oversaturated, plastic-looking. Most modern models support a negative prompt block separately or via a specific keyword like Stable Diffusion's negative_prompt parameter.
Does this tool save my prompts?
No. Every value you enter and every prompt assembled lives in memory for this browser tab only. Nothing is transmitted to a server, stored in a database, or synced across devices. Close the tab and the data is gone.
Photos sit inside a brand system.
Our brand-identity engagements pair art-direction with the prompt library, lighting templates, and PDP layouts your team uses. The system extends; the prompts compound.