If you sell at a farmers market, your photos are doing a strange job. They have to look good enough for Instagram, honest enough that nobody feels tricked at the table, and consistent enough that a customer scrolling your feed recognizes you from three stalls away. Most vendors I know are shooting this stuff at 11pm on a kitchen counter with a phone propped on a mixing bowl.
That is a workable starting point. You do not need a studio. You need a repeatable process so the twelfth photo looks like it belongs to the same stall as the first one. This guide is about building that process with an AI image generator filling in the parts a phone camera can't do on a kitchen counter.
Start with a shot list, not a prompt
The mistake nearly everyone makes is opening a generator and typing something vague like "artisan jam jar, rustic, warm light." You get a pretty picture that has nothing to do with your actual product — the jar shape is wrong, the label reads like nonsense, the jam color doesn't match what's inside.
Write the shot list first. For a market stall, five shots usually cover it:
- The hero shot: one product, front and center, nothing competing with it
- The hands shot: your hand holding or pouring the product, which signals scale
- The spread shot: six or eight products together, showing range
- The context shot: product on a table or in a basket, implying where it goes
- The detail shot: texture close-up — crumb, grain, skin, glaze
Once the list exists, every prompt you write has a job. You stop generating random pretty images and start generating the five you actually need.
Shoot your base images flat and boring
Your phone photos should be as neutral as possible before they touch a generator. Flat overcast light from a window, product centered, plain background — a sheet of baking paper works fine. Boring is the goal. You are capturing shape, label, and color accurately so the generator has something correct to build on.
The temptation is to shoot dramatic moody shots and then ask AI to clean them up. That works badly. Hard shadows and colored ambient light confuse the model, and it will invent detail that contradicts your real product. Flat photos give it clean information.
Write prompts in three blocks
A prompt that produces usable product imagery usually has three parts, in this order:
Subject block — what is physically there. "A 250ml glass jar of dark red strawberry jam with a kraft paper label, lid sealed with wax." Be specific about container, size, color, and material. If your label has a distinctive layout, say so.
Light and surface block — how it is lit and what it sits on. "Soft window light from the left, faded linen cloth on a wooden table, gentle shadow falling to the right." This is where consistency across your set gets built. Pick one lighting setup and one surface and reuse them for every product, changing only the subject block.
Camera block — the framing. "Eye level, 50mm lens, shallow depth of field, product fills two thirds of the frame." Keep this identical across the whole batch and your product set will look like it was shot in one session.
The reason vendors struggle with consistency is that they rewrite all three blocks every time. Change one, hold the other two.
Where AI genuinely helps, and where it lies to you
It is excellent at background replacement, surface swaps, adding believable shadow and reflection, and filling in a styled context after the fact. Turning a jar on a kitchen counter into a jar on a linen-draped table takes a couple of minutes and looks honest.
It is unreliable at text on labels. Ask it to place your brand name and you will get plausible-looking lettering that is one letter off or silently misspelled. Never let a generated label go out without a text overlay added afterward in a proper editor. Customers notice, and misspelled jam is a weird thing to be known for.
It is also unreliable at texture honesty. A generated sourdough crumb can look gorgeous and completely unlike bread. If texture is your selling point — and for most market vendors it is — keep the detail shot a real photograph. Use AI for the styling around it, not for the food itself.
Build a small reusable set
Once you have a lighting and surface combination that works, save the exact wording. Keep it in a note on your phone. Every new product gets the same two blocks with a swapped subject, and you regenerate your whole stall's imagery in an afternoon instead of a week.
Do a quick sanity pass at the end: check every jar for label accuracy, check color against the real product, and check that the set looks like one stall. Then post the ones that pass and reshoot the base photo for any that don't.
A small, consistent, honest set beats a large, inconsistent one. If you want to work through this faster, the generator at aiimagegenerators.net handles the styling passes well once you feed it clean, flat source images and a prompt structure you actually reuse.