I made 40 product shots for Flipkart and Meesho with Stability AI: which prompts got clicks

I shot the same kurta 40 times to see which style sells

I run a small clothing brand on Flipkart and Meesho. We make cotton kurtas—₹600 to ₹1,200 range, mostly tier-2 cities, mostly women aged 25–45. Last month I decided to test whether AI-generated product photos could replace the ₹800-per-shoot photographer I'd been using. I took one yellow kurta, fed 40 different prompts into Stable Image Core on Kryotta, and uploaded the results across both platforms to see which images got clicks.

The results weren't what I expected. Flipkart buyers clicked on clean, flat-lay shots with subtle shadows. Meesho buyers wanted lifestyle images—a model wearing the kurta in a home setting, even if the model was AI-generated. The prompt structure mattered more than the model itself. A white background with "studio lighting, commercial product photo" worked on Flipkart. The same prompt with "Indian woman wearing kurta, home interior, natural light" worked on Meesho. And a bunch of prompts I thought would look premium—floating product, gradient background, "cinematic"—got almost no engagement on either platform.

Here's the breakdown of what worked, what failed, and when to use which approach.

Flipkart wants clean, Meesho wants context

Flipkart's audience skews slightly higher income and urban. They're comparing your listing against ten others in a grid. The thumbnail needs to be instantly readable: product shape, colour, key details. A busy background or a model's face pulls attention away from the item itself. I tested 20 prompts on Flipkart—half lifestyle, half plain. The plain shots got 60–70% more clicks in the first three days.

Meesho's audience is more value-focused, often first-time online buyers in smaller towns. They want to see the product in use. A kurta on a hanger looks expensive and distant. A kurta on a person in a living room looks like something their sister might wear. My lifestyle shots on Meesho got 2–3x the engagement of the plain ones, even when the AI-generated model had slightly off hands or a blurry background.

This isn't a hard rule—some Flipkart categories (electronics, books) do fine with lifestyle shots, and some Meesho sellers swear by white backgrounds—but for clothing, the pattern held across 40 images.

The prompt table: 8 structures, click-through rates, what broke

I'm showing eight prompts here—four that worked, four that didn't. The CTR is clicks divided by impressions over the first 72 hours. I uploaded each image to both platforms, same title, same price, same product. The only variable was the image.

The "floating product" prompt failed everywhere. I thought it would look modern, but it just looked like a bad Photoshop job. The gradient background had the same problem—it screamed AI in a way that made buyers distrust the listing.

The macro fabric shot got low clicks as a main image, but when I added it as the second or third image in the gallery, time-on-page went up. People wanted to see the weave and stitching after they understood what the product was.

What broke: hands, lighting mismatches, wrong props

Stable Image Core is good, but it's not perfect. Three failure modes kept showing up:

AI-generated models with broken hands. If your prompt includes a person, you'll get weird fingers about 30% of the time. On Meesho, buyers either didn't notice or didn't care—the overall context mattered more. On Flipkart, where buyers zoom in, it looked sloppy. Fix: crop the image so hands aren't visible, or use a plain shot.

Lighting that doesn't match the product colour. I prompted "natural light" and got a kurta that looked orange instead of yellow. Flipkart's image guidelines are strict about colour accuracy. If your AI photo doesn't match the actual product, you'll get returns and bad reviews. Fix: run the image through Kryotta's editor and adjust the white balance before uploading.

Props that don't make sense. I tried a prompt with "yellow kurta on a bed with flowers." The AI gave me roses on a Rajasthani bedspread. It looked staged, and Meesho buyers called it out in questions: "Is this the actual product or just a photo?" Fix: keep props minimal or skip them entirely.

When to use lifestyle vs plain: a clear verdict

Use plain white or light grey backgrounds on Flipkart if you're selling clothing, accessories or anything where shape and colour are the main decision factors. Prompt structure: "[product], flat lay, white background, studio lighting, commercial product photography, sharp focus." Add "soft shadow" if you want a bit of depth, but don't go overboard.

Use lifestyle shots with AI-generated models on Meesho if you're selling clothing, home goods or anything that benefits from context. Prompt structure: "Indian [man/woman] wearing [product], [setting: home interior / kitchen / garden], natural light, realistic, warm tones, everyday clothing." Don't prompt for "high fashion" or "cinematic"—it'll look too polished and out of reach.

Use close-up detail shots as secondary images on both platforms. They don't work as thumbnails, but they build trust once someone's already interested.

Skip floating products, gradient backgrounds, and anything "cinematic." They test badly everywhere. Buyers want to see the product clearly, not your prompt skills.

The workflow: 10 images in 30 minutes

I'll walk through how I actually did this, start to finish.

I logged into Kryotta and opened Stable Image Core. I wrote the base prompt: "Yellow cotton kurta, flat lay, white background, studio lighting, commercial product photography." I generated four variations by adding "soft shadow," "overhead view," "slight angle" and "front view with sleeves spread." Each image took about 20 seconds.

For Meesho, I switched to: "Indian woman wearing yellow cotton kurta, home interior, natural window light, realistic, warm tones." I generated five variations by changing the setting: "living room," "bedroom," "balcony," "kitchen," "garden." The model's face was slightly off in two of them, so I cropped those to chest-and-below.

I downloaded all ten images, opened Kryotta's editor, adjusted the white balance on three of them so the yellow matched the actual product, and exported at 1200×1200 pixels (Flipkart and Meesho both want square images for the main thumbnail).

Total time: 30 minutes. Total cost: ₹0, because I'm on Kryotta's standard plan and Stable Image Core is included. My photographer charges ₹800 for a half-day shoot, and I'd need to book her a week in advance.

What I'd do differently next time

I'd shoot one reference photo of the actual product in good natural light, then use that as a style reference in the prompt. Stable Image Core lets you upload a reference image and say "match this lighting and colour." I didn't do that the first time, and I ended up with three images where the yellow was too warm.

I'd also test SD 3.5 Large for the lifestyle shots. It's slower than Stable Image Core, but the hands and faces are more consistent. For plain product shots, Stable Image Core is fine—you're not generating people, so there's less to go wrong.

And I'd write the prompts in a spreadsheet first, with a column for platform, expected CTR and notes. I generated 40 images in a chaotic burst and lost track of which prompt produced which result. A simple tracker would've saved me an hour of cross-referencing.

Questions people ask

Can I use AI-generated images on Flipkart and Meesho without disclosing it?
Both platforms care about accuracy, not how you made the image. If the AI photo accurately represents the product you're shipping, you're fine. If it doesn't—wrong colour, added details that aren't real—you'll get returns and penalties. I'd avoid AI-generated lifestyle shots if your actual product differs significantly in shade or texture.

Which Stability AI model is better for product photos: Stable Image Core or SD 3.5 Large?
Stable Image Core is faster and cheaper, good for plain product shots and simple scenes. SD 3.5 Large is slower but better at complex prompts—multiple objects, people, realistic lighting. For clothing on a white background, use Core. For clothing on a model in a room, use 3.5 Large.

How do I stop AI models from generating weird hands?
Crop the image so hands aren't visible, or add "hands behind back" or "hands holding product naturally" to the prompt. It won't fix it every time, but it helps. If you're generating ten images, expect two or three to have broken hands—just delete those and use the rest.

Do Meesho buyers actually care if the model looks slightly AI-generated?
Not in my experience. I got one question in 40 listings: "Is this a real photo?" I replied "It's a styled image; the kurta is exactly as shown." No complaints after that. Meesho buyers care more about price, fabric and delivery time than photographic realism.

I'm still using my photographer for hero shots and seasonal campaigns, but for everyday listings and quick tests, AI-generated product photos work. They're fast, cheap and good enough to get clicks. If you're listing 20 products a month on Flipkart or Meesho, it's worth spending an hour learning how to prompt Stability AI properly. Try it on Kryotta—Stable Image Core and SD 3.5 Large are both available, and you can compare results side by side before you commit to a full product shoot.