Meesho bulk image generator: catalog images for 100 SKUs without chaos.

Producing images for one product is a craft. Producing them for a hundred is logistics. The sellers who do it calmly are not faster photographers; they have a process where every file knows its own name and every step runs once.

The bulk problem: 100 SKUs, 500 images, one shoot

The arithmetic is what makes bulk image work feel impossible. Fifty to five hundred SKUs, four to five images each, and one weekend to shoot them before a season starts or a bulk sheet is due. At a hundred SKUs that is four to five hundred finished files, every one of which has to be a 1:1 square at 1000x1000 pixels, sharp, on a clean background for the first slot, free of text and watermarks, and matched to the right row of the upload sheet. Done one image at a time, with each file opened, cropped, cleaned, saved and hunted for later, the job takes a week and produces a catalog that looks like it was shot by six different people.

The first move is to shrink the number. Variations of one design share most of their images. A kurti set in three colours and three sizes is nine SKUs, but it needs three colour-accurate front shots and, in most categories, one shared back, one shared detail and one shared in-use image. Nine SKUs, six unique photos, not forty-five. Count the unique shots per design before shoot day and you will usually find that a hundred SKUs is closer to a hundred and fifty photographs than five hundred. That list is also your checklist for the rest of the process, because every file you finish can be ticked against it. If your SKU scheme is not yet clean enough to group variations reliably, fix that first with the SKU ID guide; a batch built on messy SKUs mislabels itself.

Shoot on a fixed setup so every frame matches

Consistency across the batch is worth more than quality on any single frame, because the storefront grid shows your products side by side and a buyer reads a mismatched grid as a careless shop. Pick one spot near a window, tape down one white sheet or board, mark the phone position on the floor or a tripod with tape, and lock the exposure and white balance once the first product looks right. Then do not touch any of it until the batch is done. Shoot every variation of a design back to back so the colours are compared under identical light, and shoot in one or two sessions at the same time of day rather than across a week of changing weather. If the shoot has to span two days, photograph the same reference product at the start of each session and compare the two frames before you edit anything; if they differ, the whole second day needs a separate white balance correction. The product images best practices guide has the full single-shot setup; here the only extra rule is that nothing moves between SKUs.

One cheap addition earns its place in a batch: a reference card. Put a plain white card and, if you have one, a grey card in the first frame of every session and every design, then remove them for the real shots. When you batch the white balance later, you correct to the card rather than to memory, and every session lands on the same white. It also settles arguments a fortnight later, when a buyer says the colour was different and you can open the reference frame and see exactly what the light did that day. Sellers who shoot in a room with a coloured wall or a warm bulb overhead need this most, because the cast is invisible in the moment and obvious in the grid.

Name every file by SKU before you edit anything

This is the step sellers skip because it feels like admin, and it is the step whose absence causes almost every mismatched-image complaint. A phone names photos IMG_4821, IMG_4822 and so on, and an hour after the shoot nobody can say which of them is the blue medium and which is the pink large. Rename at import, before a single edit, using the Seller SKU ID followed by the image slot: KURTISET-BLU-M-1 for the front packshot, KURTISET-BLU-M-2 for the back, through to slot five. Uppercase, hyphens, no spaces, slot number last so files sort by SKU and then by slot in any folder view. For shared images, copy the file under each variation’s name rather than trying to reference one file from several rows; storage is cheap and ambiguity is not.

Once the name is on the file, every downstream tool carries it. Background removers, resizers and compressors keep the file name by default, so the finished KURTISET-BLU-M-1.jpg is still called that after four passes, and the bulk sheet can reference it without a lookup. Better still, a spreadsheet can build the reference from the SKU ID column with a formula, which means the image columns of a hundred rows fill themselves. The one thing that breaks all of this is renaming after the sheet is filled. Name once, at import, and treat the name as permanent.

A convention worth copying in full: the Seller SKU ID exactly as it appears in your sheet, a hyphen, then the slot number one to five. Shared images at the design level can be shot once and stored under the style code, for example KURTISET-2 for the shared back view, then copied to each variation as KURTISET-BLU-M-2, KURTISET-PNK-M-2 and so on before the batch runs. If you ever reshoot a slot, append a version rather than overwriting, KURTISET-BLU-M-1-v2, so the sheet can be pointed at the new file deliberately and the old one is still there if the new one is worse. Keep raw files in a separate RAW folder under the same names, because the day a compression pass goes wrong you will want to re-run from the originals rather than re-shoot. None of this needs software; it needs the decision to be made once, before the first edit.

Batch the mechanical steps, one pass each

With named files, the middle of the pipeline is a series of single passes across the whole folder. First the colour-true edits: straighten, exposure and white balance, as one preset applied to everything, followed by a fast scroll to catch the handful of frames the preset got wrong. No filters, no saturation, nothing that moves the colour away from the product in your hand. Second, background removal, and here the rule is slot one only. The first image needs the clean white or light neutral background; the back, detail and in-use shots do not, and running them through a remover wastes credits, flattens the context those shots exist to give, and multiplies the halo edges you then have to inspect. Third, the square. Use a pad-to-square preset that centres the product and adds white margin rather than a crop preset that trims to a square, because a crop built on a kurti will cut the ends off a saree and the handle off a bag. Export at 1000x1000 pixels, never below the 500x500 minimum. Fourth, compression, to the file size the current template states, checked on a dark fabric and a fine print where artefacts show first.

Build every preset on your hardest SKU, the tallest, the widest, the darkest and the shiniest, and the rest of the batch will pass through it without incident. Building it on an easy SKU and hoping is how a preset that looked perfect on a t-shirt trims a hundred sarees.

QC risk at scale: one wrong crop copied to 100 listings

A batch copies mistakes as faithfully as fixes, and this changes what a mistake costs. A single mis-cropped image is one rejection, one re-upload, one lost listing day. The same crop preset run across a batch is a hundred rejections, and in the panel you will usually see them arrive as a hundred separate QC flags rather than one, each needing its own corrected file. Colour drift is worse because it passes QC: a white balance that ran warm on the second shoot day makes every product from that session a shade more orange than the one that ships, and the returns arrive a fortnight later with “colour not as shown” on every one. Mixed backgrounds, half the batch on a white sheet and half on a grey wall, do not fail review either; they simply make the grid look like two different shops and quietly cost clicks on every listing. The common thread is that a batch error is invisible at the file level and obvious at the catalog level, which is why you inspect the batch as a batch. For what QC itself will flag, the Meesho QC errors guide lists the reasons; for the doorstep failures that pass QC, the image generator guide covers the truth checks.

The answer is sampling. You cannot open five hundred files at full zoom, and you do not need to. Sample one file in ten across the finished batch, chosen evenly rather than from the top of the folder, and open each at full zoom on a phone-sized screen, which is where a buyer will see it. Check the outline for halos, the corners for trimmed product, the colour against the reference frame from the shoot, and the first image for a genuinely plain background. Then add the risky ten: the tallest and widest SKUs for the crop, the darkest and shiniest for compression and halo edges, any lace, fringe or glass for the mask, and the first and last SKU of each shoot session for colour drift. If the sample is clean, the batch is almost certainly clean. If it finds a fault, do not fix the files; fix the preset or the step that produced them and re-run only the affected SKUs, because a fault in a preset is in every file that went through it, not just the one you opened. Twenty minutes of sampling is the cheapest insurance in the whole process, and the tallest, widest, darkest, shiniest rule is what makes it reliable.

From folder to bulk sheet

The last stage is where the naming pays off. Move each SKU’s five finished files into a folder named by the SKU ID, so every product is a self-contained set you can re-upload, replace or hand to someone else without explanation. Then open the bulk template. It has one row per variation and a column per image slot, and whatever form of reference the current template asks for, a hosted link or a file reference, the SKU-slot file name means the column can be built from the SKU ID with a formula rather than filled by hand. Verify a handful of rows point at the right files, pick the tallest and the darkest SKU again for the check, and upload. Watch the QC status on the first few SKUs before you assume the rest are fine.

The verification is worth doing properly rather than by eye. Open the sheet and the folder side by side, sort both by SKU ID, and step through five rows spread across the batch: the first, the last, and three from the middle. For each, open the file the image-one column points to and confirm it is that SKU in that colour, not the neighbouring variation. A blue medium pointing at a pink medium is the classic failure, and it passes QC cleanly, because the pink image is a perfectly good image. It fails at the doorstep, twice: the buyer who ordered blue and received it returns it because the picture was pink, and the buyer who wanted pink never ordered because the blue listing showed it. If your batch is large enough that five rows feel thin, check one row in twenty. The bulk catalog upload guide covers the rest of the sheet, the catalog upload walkthrough covers the upload itself, and catalog upload frequency explains why a clean batch every few weeks beats a chaotic one every quarter.

Where the Robnu AI Catalog Studio fits

If your products are already in an OMS, the bulk problem changes shape, because the catalog is the source of the batch rather than a folder on a laptop. Robnu’s AI Catalog Studio generates packshots and short product reels from products that are already in your catalog, per product, inside the same system that runs your orders, returns and payouts. It is a feature of the OMS rather than a stand-alone bulk generator, credits are included to start, and the output goes through the same one-in-ten sampling check as any other batch before it is uploaded, because a generated packshot is still a packshot and the catalog rules do not know or care who made it.

Sources & further reading

Bulk template formats, image column conventions and file size limits change; confirm the current template in your own Supplier Panel before a large batch.

• Meesho Supplier panel, bulk catalog upload
• Meesho Supplier, Learning Hub
• Batch processing, general background