AI photoshoot luxury: what works, what doesn't yet | Mira Studio
Leather, hands, text: an honest review of AI imagery limits in 2026 and how we work around them in production.
· Insight · 6 min · By Estée Jaillon, Mira Studio
Generative AI now produces images that many viewers can no longer distinguish from traditional photography. But in the studio, we know exactly where the technology still struggles. Here's what works in 2026 for an AI photoshoot luxury production, what remains difficult, and the workarounds we apply at Mira Studio to deliver regardless.
Materials: uneven but reliable under certain conditions
What works well
Velvet, silk, satin, fluid fabrics: AI renders them with a level of detail and texture that works perfectly for AI photography luxury brands. Reflections, folds, cast shadows come through without issue. Brushed metal, gold, silver: smooth or lightly textured finishes are handled well, especially in wide or medium shots.
What remains tricky
Grained leather in extreme close-up, transparent glass with multiple layers of overlapping reflections, multi-faceted gemstones (diamond, emerald). AI tends to smooth out micro-details, invent reflections that don't obey physics, or merge focal planes. For complex jewelry, we still see inconsistencies in the geometry of prongs or pavé settings.
Our workaround
We compose the shot to minimize depth of field or frame wider. We use multiple generation passes with depth control, then merge in post-production. For Jacob & Co, Holiday 2025 campaign delivered in 14 days, we combined separate renders of case, bracelet and background, then assembled to guarantee reflection coherence. The final result is pixel-perfect, but it requires real art direction.
Hands: the truth test of any AI image
What works well
Hands at rest, fingers slightly apart, three-quarter views. As long as the pose is natural and fingers don't cross, AI produces credible hands with nails, knuckles and veins correctly placed.
What remains tricky
Intertwined fingers, hands in motion, fine gestures (holding a champagne flute by the stem, slipping on a ring). AI still confuses perspective, adds or removes a knuckle, distorts proportions as soon as the pose becomes complex. This is one of the most persistent limits of AI imagery in 2026.
Our workaround
We avoid acrobatic poses. We favor framing where the hand is suggested rather than detailed (artistic blur, half out-of-frame). When the hand must be perfect, we generate several variants, select the best and retouch locally. For certain productions, we integrate a real hand photo as a composite: an assumed hybrid, flawless result.
Text in image: almost there, but not quite
What works well
Simple logos, monograms, short typography (three to five letters) in large point size. AI can now reproduce a brand name on packaging, a label, a ribbon, if we provide a clear reference and constrain the generation.
What remains tricky
Paragraphs, legal mentions, serial numbers, complex typographic compositions. AI still invents characters, mixes cases, curves baselines randomly. For a luxury brand, wonky text destroys credibility in a fraction of a second.
Our workaround
We generate the image without text, then overlay the typography in post-production with the correct fonts, proper letter spacing, accurate perspective. This is exactly what we do for Borsalino Parfums: the bottle, ambiance and lighting are AI-generated, the logo and mentions are added in traditional graphic editing. Result: no visible difference from a traditionally retouched photo.
Multi-image consistency: the real campaign challenge
What works well
Maintaining a color palette, light direction, general ambiance across a series of visuals. With a solid prompt and locked style parameters, we achieve sufficient visual unity for a lookbook or Instagram series.
What remains tricky
Having exactly the same model appear from multiple angles, in multiple settings, with the same hair, same makeup, same clothing. AI generates subtle variations (face shape, skin tone, hair texture) that break continuity as soon as you look at three images side by side.
Our workaround
We use face reference tools (face reference, character consistency) that anchor the visual identity of a virtual model. We generate all images in a series with the same reference set, we validate consistency before delivery. For Sporty & Rich × Juicy Couture, we produced several dozen visuals with three recurring virtual "models": each maintains their identity from one image to the next. This is production work, not just generation.
Movement and video: the next frontier
What works well
Static shots, slow camera movements (tracking, zoom), short animations (3 to 5 seconds) with a stationary subject. Current AI video (early 2026) produces usable sequences for teasers, motion design, social media loops.
What remains tricky
Complex human movements (walking, dancing, rapid gestures), physical interactions (picking up an object, pouring liquid), narrative continuity beyond a few seconds. AI invents distortions, jumps, involuntary morphing. For a long video campaign, we're not there yet.
Our workaround
We cut sequences into short shots, control each shot individually, edit in post-production with transitions that mask inconsistencies. We hybridize: an AI-generated background, a product movement in traditional 3D, a final composite. For certain brands, we embrace the short format (5 to 10 seconds per shot) and build the narrative through editing rather than continuous footage. This is what we explore with Jacob & Co: creating the impossible in 14 days, where each visual is a piece of a narrative puzzle, not a single scene.
What we don't work around: art direction
The limits of AI imagery in 2026 are not a barrier if you know them and build production around them. What cannot be replaced is the eye: knowing how to frame to avoid the hand trap, choosing the material that will render well, composing the image so text overlays naturally, orchestrating a series so it tells a story.
AI is a rendering tool, not an art director. We produce for Assouline, Jacob & Co, Borsalino, Chow Tai Fook: every campaign starts from a creative brief, a moodboard, an intention. AI executes, we direct. And when it stalls, we work around it. That's what an AI photoshoot luxury production means in 2026.
FAQ
Can AI generate jewelry with realistic gemstones?
Yes, but with limits. Complex faceted stones (diamond, emerald) still show reflection and geometry inconsistencies in extreme close-up. We work around this by framing wider, working multiple generation passes, or hybridizing with traditional 3D for critical details.
Can you produce an entire campaign with the same virtual model?
Yes, provided you use face reference tools and lock generation parameters. We achieve sufficient consistency for a complete series, but it requires real production rigor and image-by-image validation.
How do you handle text (logo, mentions) in an AI-generated image?
The simplest approach remains adding text in post-production with the correct fonts and accurate perspective. AI can reproduce short, simple logos, but as soon as you want precise typographic rendering, traditional graphic overlay is more reliable and faster.