FLUX.2 for Product Photos: Multi-Reference Control, Exact Colours and 4MP Output

Most AI image models can make an attractive product scene. The harder ecommerce problem is keeping the product recognisable while changing the context around it. FLUX.2 from Black Forest Labs is worth looking at because its current feature set is unusually focused on production controls rather than generation alone.

FLUX.2 supports image generation and editing in the same model family, multiple reference images, output up to 4 megapixels, precise hex colour control and improved text rendering. Black Forest Labs explicitly lists ecommerce product photography, contextual lifestyle shots and variant generation among its use cases.

Those capabilities do not guarantee that every generated bracelet, watch or packaged product will be exact. But they give sellers several useful ways to reduce how much information the model has to guess.

Multi-reference editing is the most useful product feature

FLUX.2 can combine several source images in one edit. The exact limit depends on the variant and interface. Black Forest Labs currently documents up to eight reference images for the FLUX.2 Pro API and up to ten in some other FLUX.2 workflows.

For product photography, the point is not to upload ten nearly identical photos. It is to give the model complementary evidence. A jewellery seller could provide a clean hero photograph, a side view that shows thickness and a close-up of the clasp. A handbag seller might provide front, rear and hardware details. A packaged-product seller could add a separate sharp image of the label.

FLUX.2 lets prompts refer to those inputs naturally or by image number. That makes a brief such as “use the bracelet from image 1, preserve the clasp shown in image 2 and use image 3 only to understand the side profile” much clearer than expecting the model to infer hidden geometry from one photograph.

Exact hex colours solve a different problem

Black Forest Labs also advertises exact colour matching through hex codes across the current FLUX.2 family. This is useful when the generated scene needs to follow a brand palette. Instead of describing a background as “dark forest green” and accepting whatever shade appears, a seller can specify the brand colour directly.

That is particularly useful for campaign backgrounds, packaging mockups, graphic elements and catalogue templates. If your website uses the same cream and green palette across every collection, colour control can help generated assets fit the existing brand rather than creating a slightly different palette each time.

It should not be confused with accurate material reproduction. Asking for a specific gold-like colour is not proof that an AI edit accurately represents the physical plating or finish of a real product. For jewellery colour variants, compare the result with a genuine sample whenever colour accuracy affects the buying decision.

4MP output gives sellers more room to work

FLUX.2 supports output up to 4 megapixels, such as a 2048 by 2048 square image. That is useful for ecommerce because product images are often reused across several destinations. A larger master gives you more room to crop a square marketplace image, a 4:5 advert or a banner without immediately running out of detail.

Resolution still does not fix incorrect geometry. A 4MP image with an altered clasp is simply a sharper inaccurate image. Treat resolution as a delivery feature, not a substitute for product verification.

The FLUX.2 variants are built for different jobs

FLUX.2 is a family rather than one model. Pro is positioned as the production-grade balance of quality and speed. Max is the highest-quality option and adds real-time grounding. Flex is aimed more heavily at typography and detail control. Klein is the fast, compact option and can run on consumer hardware in supported configurations.

For a seller manually creating a few campaign images, the distinction may not matter much at first. For a business generating hundreds of catalogue variations through an API, speed, reference limits and per-megapixel pricing become part of the workflow design.

Black Forest Labs currently prices FLUX.2 by megapixel, with reference images contributing to the calculation. That means uploading more references has a real processing cost as well as a potential fidelity benefit. Use the smallest reference set that actually adds useful product information.

A practical FLUX.2 workflow for ecommerce

Start with your clearest real product image and make it the primary reference. Add only the views that reveal information the first image cannot show. Then describe the new scene separately from the product-preservation instructions.

For example: “Image 1 is the primary bracelet reference. Image 2 shows the exact clasp and image 3 shows the side thickness. Preserve those product details. Place the bracelet on a cream stone pedestal in a premium studio scene with soft window light. Use #F4F0E8 for the background and leave clear negative space above the product for advertising copy.”

Generate the composition, then compare it with every reference at full size. For jewellery, inspect the clasp, chain path, engraving, stone count, edges, thickness and reflections. For other products, check labels, ports, stitching, hardware and proportions. If the model changes something customers would reasonably expect to receive, the image needs another pass.

Where FLUX.2 fits beside specialist tools

FLUX.2 is interesting because it exposes several controls that developers and production teams can build into repeatable systems. A general model still leaves the seller responsible for deciding which references to supply, how to phrase the brief and how to organise repeated generations.

A specialist workflow such as Lustra Studio handles a narrower problem: repeatable jewellery references, model images and colour variations without rebuilding the setup each time. The underlying lesson is the same. Product imagery improves when the model receives reliable visual evidence and the workflow limits what it is allowed to invent.

The takeaway

FLUX.2 is not notable simply because it can generate photorealistic images. For ecommerce, its more useful features are multi-reference editing, explicit colour control, flexible high-resolution output and a model family designed for different production needs.

Those controls can make product-image generation more systematic, but they do not remove the need for inspection. Use real product references as the source of truth, add extra views only when they reveal missing information, and judge success by whether the product stayed accurate before judging whether the scene looks impressive.

Sources: Black Forest Labs, FLUX.2 model and product pages; Black Forest Labs Help Center, FLUX.2 and multi-reference editing documentation, updated June 12, 2026; Black Forest Labs pricing page, accessed August 17, 2026.