A Dark Product Photo Does Not Always Need AI Relighting

A dark product photo does not automatically need generative AI. Sometimes the light direction, reflections and shadows are already correct, but the capture is simply too dim. In that case, conventional exposure and tonal controls may recover the photograph without asking a model to reinterpret the product.

AI relighting solves a different problem. It can make the subject appear lit from a new direction, rebuild highlights and shadows, or help a cutout match a generated background. That extra power can be useful, but it creates more ways for colour, material and shape to change.

The practical comparison is therefore not old editing versus new editing. It is pixel adjustment versus light reconstruction. Sellers should choose the smaller intervention that fixes the actual fault.

What conventional exposure correction does

Adobe describes Exposure as a control for overall image brightness. Its Highlights and Shadows controls target lighter and darker areas, while Whites and Blacks set the end points of the tonal range. These adjustments change the recorded tones without inventing a new light source.

That makes tonal correction the safer first option when the source lighting is believable. If a silver ring has the right highlight shape and shadow direction but the whole frame is half a stop too dark, raising Exposure and carefully opening Shadows can solve the presentation problem while leaving the product geometry intact.

The limits are equally important. Brightening cannot move a reflection from the right side to the left, create a missing rim light or make a front-lit object look convincingly side-lit. Lifting deep shadows can also reveal noise, colour blotches or insufficient captured detail.

What AI relighting changes

AI relighting analyses the subject and generates a new interpretation of how light should fall across it. Photoroom’s API, for example, includes an AI Relight feature and offers an updated mode intended to preserve hue and saturation more accurately.

Photoroom’s own documentation warns that AI Relight can alter colours and recommends human validation when product accuracy matters. That warning is especially relevant for jewellery, cosmetics, fabrics and packaging where colour or surface finish is part of the product promise.

OpenAI’s current image prompting guide also treats lighting transformation as a generative edit. It recommends specifying that only environmental conditions should change while identity, geometry, camera angle and object placement remain fixed. Those preserve instructions are useful, but they are not a guarantee that every pixel outside the lighting will remain unchanged.

Run a fair two-route test

Choose one difficult photograph rather than an easy matte product. A polished pendant, glass bottle, watch or metallic cosmetic tube will reveal whether an edit is changing material behaviour as well as brightness.

Keep the original RAW file or highest-quality master. Route A should use conventional controls only: Exposure, Highlights, Shadows, Whites, Blacks and, if needed, a local mask. Do not add a new background or use generative fill.

Route B should use the AI relighting tool on the same starting image. Describe a specific target such as “soft window light from the upper left with a gentle contact shadow.” Add a preserve list covering exact product shape, colour, logo, engraving, texture, camera angle and scale.

Match the final crop, pixel dimensions and colour space. Try to bring both outputs to a similar overall brightness so the comparison is not simply bright versus dark. If the AI tool produces variable results, generate several candidates and record how many attempts were needed. The conventional edit is usually deterministic, so repeatability is part of the comparison.

Judge the product before the mood

First compare silhouette, edges and small factual details with the master. Check prongs, chain links, clasps, labels, stitching, seams and control positions. A relit image that looks more dramatic but changes one sale-relevant detail has failed the product test.

Next inspect colour. Compare neutral metal, gold tone, gemstones and brand colours. A warmer light can legitimately make a scene feel warmer, but it should not make a silver item appear rose gold or change a red package into orange.

Then assess material behaviour. Polished metal should have coherent highlight bands and dark reflections. Glass should retain believable transmission and edge highlights. Matte products should not suddenly acquire a plastic shine. AI can produce attractive light that describes the wrong material.

Finally check shadow direction, contact and scene fit. If the product has been placed into a new lifestyle background, its highlights and cast shadow should agree with visible windows, lamps or sunlight. Conventional brightening cannot repair a fundamental direction mismatch, which is where relighting earns its extra risk.

Use the result to choose a workflow

Choose conventional correction when the captured lighting already explains the product correctly and the problem is exposure, contrast or recoverable shadow detail. It is faster, cheaper and less likely to change logos, engravings or geometry.

Choose AI relighting when the lighting direction or quality genuinely needs rebuilding, particularly after background replacement or compositing. Use the least dramatic setting that achieves the match, compare against the real item and reject outputs that alter colour or surface finish.

A hybrid workflow can work, but order matters. Correct obvious capture problems from the RAW file first, then use a restrained relight only if the product still conflicts with the target scene. Do not repeatedly relight an already generated result, because each pass creates another opportunity for drift.

For reference-led tools such as Lustra Studio, keep the approved real photograph as the product master even when a lifestyle scene needs new lighting. The generated scene can guide the campaign look, while the master remains the evidence used to check metal colour, chain structure, engraving and proportions.

The decision rule

If the light is correct but the pixels are dark, correct the pixels. If the light itself is wrong for the intended scene, test a relight.

That distinction prevents sellers from using a generative tool for a problem that a reversible tonal adjustment can solve. AI relighting is most valuable when it changes the relationship between product and environment, not when it merely makes a dim photograph brighter.

Sources

Adobe Lightroom, “How to adjust Light in an image,” reviewed 4 October 2026: https://helpx.adobe.com/ca/lightroom/web/edit-photos/apply-effects/adjust-light.html

Photoroom API Documentation, “Introduction,” reviewed 4 October 2026: https://docs.photoroom.com/

Photoroom API Documentation, “Changelog,” including the hue-and-saturation-preserving AI Relight mode, reviewed 4 October 2026: https://docs.photoroom.com/getting-started/changelog

OpenAI, “GPT Image Generation Models Prompting Guide,” reviewed 4 October 2026: https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide