The duck survived the move to a garden. The framing didn’t.New Scene, Same Subject?A yellow rubber duck went from a sheet of yellow paper to a stone tabletop, with green foliage behind it and warm sunlight across its head. Its orange bill, black eye and raised wing were still recognizable. But it…
Annons
Annons
The duck survived the move to a garden. The framing didn’t.New Scene, Same Subject?A yellow rubber duck went from a sheet of yellow paper to a stone tabletop, with green foliage behind it and warm sunlight across its head. Its orange bill, black eye and raised wing were still recognizable. But it occupied less of the frame, despite an instruction to preserve the framing.If you’re building a carousel or a set of product images, that difference matters. A pleasing new scene can still leave you adjusting crops and checking whether the product sits in the same place across the set.This small test used one real photograph for three edits: background only, lighting only, and a new scene with new lighting. A fourth output addressed the lighting result. AI kept the duck recognizable throughout; composition and material appearance needed closer review. Here’s what each change would mean for a creator trying to reuse an approved image.Start with details you can actually compareThe reference was a photograph by RandomPerson238: one yellow duck on yellow paper, facing right. Its bill contour, oval eye with a white highlight, wing outline and upturned tail gave the comparison specific landmarks. The duck had no printed label, so these results say nothing about preserving fine text or logos.Each edit started from that same original. The settings were Krea 3 Beta, Fast and Precise Edit enabled, with Auto aspect ratio and one output per request. No region was manually masked. The downloaded results measured 2368 × 1776 pixels.The shared prompt named the features to preserve and explicitly asked to keep the pose, camera angle and framing. That follows the approach in Krea’s prompting guide: specify both the change and the details that should stay fixed.There was one output for each condition, followed by one targeted lighting correction. That makes this a worked example you can inspect, rather than a reliability score for the model.A plain background swap was the closest fitThe first request replaced the yellow paper and supporting surface with a seamless pale blue studio backdrop. It also asked Krea to retain the original lighting and contact shadow.The result delivered the blue background while keeping the duck in a closely matching position. The eye, bill, wing and tail remained easy to compare with the source. For a simple color variation in a visual series, this was the most straightforward starting point of the three initial outputs.Left: original photograph by RandomPerson238. Right: AI-edited in Krea 3 on Krea.ai. Background, lighting and texture changed. CC BY-SA 4.0.Even here, “the same subject” needs a practical definition. The surface texture and illumination were not identical to the photograph. Removing a yellow surrounding also changed the visual impression of the duck’s yellow material. A background edit still deserves a check of the object itself, particularly when an approved product color matters.The garden scene worked; the layout changedThe third request gave Krea more to rebuild: a matte grey stone tabletop, a blurred green garden and warm, low sunlight from the upper left.The output met that scene brief. The tabletop had texture, the foliage stayed soft, and the warm light made the duck feel part of the setting. As an individual lifestyle image, it was the most appealing result in this set.It also changed the composition. Compared at the same canvas size, the duck was smaller, with more space above its head. The instruction to preserve framing had not held.Left: original photograph. Right: Krea 3 scene-and-lighting edit on Krea.ai. Both complete canvases shown at equal height. Original by RandomPerson238; edited adaptation, CC BY-SA 4.0.For a standalone post, you might welcome that extra breathing room. For a series with a fixed product position, it creates another decision: accept the new layout, adjust the crop, or request another edit. Cropping might help the placement, but it would also change how much garden and tabletop remain visible. That adjustment was not tested here.Keep two review questions separate: “Can I recognize the product?” and “Does this image fit the layout?” This output answered the first more convincingly than the second.“Cool daylight” produced an unexpected color castThe lighting-only request kept the yellow paper and asked for soft, cool daylight from the upper left. It also asked to retain the yellow material and physical details.The output introduced strong blue highlights on the head, bill and breast. The duck was recognizable, but those patches gave its surface a glossier, more stylized appearance than the intended soft daylight. The word “cool” had left too much room for a result that didn’t match this brief.The correction returned to the original photograph and replaced that lighting instruction with a more specific one:Keep the original yellow paper background and supporting surface. Replace the lighting with one large diffused window light from the upper left, neutral white daylight, with soft shadows. Keep the duck yellow and the bill orange. No blue light, no coloured rim light, no spotlights, no glossy blue reflections. Preserve the original matte rubber surface and camera framing.The same feature-preservation instructions remained in place. The new output removed the conspicuous blue patches and gave the duck a softer, more natural-looking illumination.This correction also changed the brief from cool daylight to neutral white daylight. It demonstrates a useful repair for this particular unwanted cast, without proving that one negative phrase will fix every lighting problem. The paper’s shadows and texture still differed from the source. Blocky tonal patches also remained in the yellow background, so fixing the blue cast did not make this a finished delivery image.Left: requested cool daylight. Right: one correction specifying a diffused neutral window light. Both AI-edited in Krea 3 from the original photograph by RandomPerson238; CC BY-SA 4.0.For your own image, name the light source and the material appearance you need. If a render introduces a specific problem, describe that problem explicitly in the next instruction. “Better lighting” leaves the model with another broad interpretation to make.Build the series around the referenceFor a set like this, keep the original photograph available beside every result. Match the canvas size when comparing them; enlarging each subject to fill its own preview can hide a framing change.A useful first pass is to check the outline, distinctive details and placement. Then review the material color, reflections and contact shadow. The garden result shows why appearance and instruction-following deserve separate judgments: it was attractive while still missing a requested constraint.Starting each condition from the original made these comparisons easier. A sequence that feeds every edited result into the next could carry an unwanted change forward. That chained workflow was not tested here.These four outputs required no new purchase on the existing Free workspace. That is an account-specific observation, not a general price promise. There was no clean timing record for a speed comparison, so the practical result here is about the files and their remaining review work.Completed Lighting Edit in the Image ViewerFor this duck, the pale blue edit is the closest candidate for a consistent studio series. The repaired lighting image offers a better direction for neutral illumination, but its background still needs attention. The garden image needs its own layout check before joining the set. Krea 3 produced usable directions from one photograph; choosing which direction belongs in the finished series still depends on the details that your layout cannot afford to change.Reference photograph: Yellow Rubber Duck by RandomPerson238, licensed under CC BY-SA 4.0. The comparison outputs are AI-edited adaptations made with Krea 3; changes include backgrounds, lighting, texture and framing. Image adaptations are shared under the same license.This story is published under the Generative AI publication. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!Can You Change the Scene and Keep the Subject? was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.