Back to the journal

Krea 3 for Product Photos: Fast Scene Changes, Fragile Fine Print

Krea 3 promises fast, low-cost 1K and 2K product edits. Here is the safest workflow for scene changes, plus the label and jewellery details to inspect.

Lilac desk fan in a sunlit room, shown in Krea 3 launch materials
Example image from Krea’s Krea 3 launch materials. Lustra Studio did not independently test this output.

Krea 3 is a new image-editing model aimed at fast production work. For product sellers, its clearest use is turning an approved packshot into several scene ideas without waiting a minute for every attempt. Krea says most edits return in under eight seconds, with native 2K output and current support for up to eight reference images.

The attractive number is the advertised $0.06 price for a 2K image. The important limitation is smaller: fine label text can still change. Krea itself says small print is a weak spot. Treat Krea 3 as a fast scene editor, not as automatic proof that the pictured SKU is accurate.

What Krea launched on 15 October

Krea released Krea 3 on 15 October 2026 as an image-editing model for product scene changes, object removal, cleanup and campaign variants. The company says most edits complete in under eight seconds at native 2K. Its launch pricing is $0.03 per 1K image and $0.06 per 2K image.

Those are Krea's published claims and prices, not results from a Lustra Studio test. Krea's speed comparison includes queue time but was measured on its own platform, and Ultra mode can take several minutes. Web subscriptions use compute units, so the cost shown in a plan may not map as neatly as the API-style per-image figure.

There is also a small documentation change worth knowing. The launch post said Krea 3 accepted up to six references per edit. Krea's current Krea 3 product page and its 20 October guide now say up to eight. If reference count matters to a production job, check the live interface or API documentation before building the workflow around it.

Start with scene changes, not product reconstruction

Krea 3 is most convincing when the physical item already exists in the input and the requested change happens around it. A white-background fan can move into a sunlit room. A bottle can move onto a styled table. A handbag can be placed into a campaign scene using a separate model or location reference.

This is different from asking the model to invent a rear view, open a locket it has only seen closed, or turn a silver ring into a different setting. In those jobs, the model must guess unseen geometry. Fast output does not make the guess more factual.

If exact geometry matters, keep the verified product as a separate asset and build the scene around it. Our guide to placing a real product into an AI background without redrawing it explains the safer compositing version of that method.

A practical first workflow for sellers

  1. Choose one low-risk SKU. Start with a rigid product that has a clear silhouette and little fine text. Avoid launching the test with a chain, engraved pendant, transparent bottle or reflective multi-part item.

  2. Prepare the source at full quality. Correct colour and perspective first. Do not expect the model to repair a blurred logo, clipped edge or compressed marketplace thumbnail.

  3. Use 1K for exploration. At Krea's published rate, four 1K attempts cost the same as two 2K attempts. Test scene, crop and lighting before paying for a larger keeper.

  4. Write the brief in two parts. First state what must remain unchanged, including silhouette, colour, logo, label, cap, hardware and camera angle. Then describe the one change, such as the background, prop arrangement or light.

  5. Assign every reference a job. Say that image 1 is the product, image 2 is the room and image 3 is the lighting reference. More references do not help if the model cannot tell which details belong to which source.

  6. Use a regional edit for a local correction. Krea says regional editing is available in Fast and High modes, but not Ultra. Mark only the unwanted prop, damaged background or isolated area that should change.

  7. Render the chosen direction at 2K, then compare it beside the source. Inspect at normal size and at 100 percent before approving it for a listing or campaign.

A useful prompt structure is: “Keep the product in image 1 unchanged, including [five identifying details]. Change only [scene element]. Match [lighting and camera direction]. Do not add text, accessories or product parts.” Specific preservation instructions are more useful than repeating “accurate” several times.

Fine print is the failure case to plan around

Krea's own Krea 3 comparison guide says large lettering and logos hold better than fine print. Its bottle example kept the prominent branding but misspelled words in smaller ingredient and research-note columns.

For packaging, keep the approved label artwork and legal copy outside the generative step. If the scene is good but the label has drifted, composite the original label back onto the product or use an editable mockup. Our editable packaging label workflow shows why factual text should remain a separate layer.

Do the same for serial numbers, barcodes, safety marks, measurements and medical claims.

Jewellery needs a stricter pass

Krea 3's launch examples include product and fashion edits, but that is not evidence that it will preserve every jewellery detail. Before using an output, count stones and prongs, trace the chain, inspect the clasp, compare the engraving, check the hallmark and confirm that the metal colour has not shifted with the new lighting.

Use it first for low-risk tasks such as extending a background, changing a surface, removing a prop or exploring a campaign mood. Avoid relying on it for a new angle, virtual try-on scale, metal conversion or gemstone change unless you can verify the result against the physical piece. The SKU mistake checklist for AI product photos is a useful final gate.

Measure kept images, not generated images

At the advertised list price, 100 Krea 3 outputs at 2K cost $6. That is inexpensive only if enough outputs are usable. Record the number generated, the number rejected for product drift, the time spent repairing labels and the final number published.

A contact sheet makes small inconsistencies easier to spot across a set. Compare the same product side by side rather than approving images one at a time. If the logo width, cap shape or metal tone changes between variants, the campaign is not consistent even when every individual frame looks polished.

Commercial use still depends on the plan and the asset

Krea's current pricing page lists a commercial licence across its plan descriptions, including Free. Its Terms of Use also say that rights can depend on the specific plan and that users remain responsible for third-party copyright, trademark and legal compliance.

That means a commercial licence does not grant rights to somebody else's logo, model, artwork or product photograph. Use source assets you own or are authorised to use, and recheck the live plan terms before producing client work at scale.

The practical verdict

Krea 3 makes the iteration stage more interesting for sellers. Fast 1K drafts, native 2K output, multiple references and regional edits can shorten the route from packshot to lifestyle concept.

Its best role is narrow and useful: change the scene while the approved product remains the source of truth. Keep fine text editable, budget for rejected outputs and use a stricter inspection process for jewellery, packaging and any product whose identity depends on small details.

Krea 3AI Product PhotographyProduct Image EditingE-commerce ImagesImage Generation