We measured the published results ourselves, from the pixels each preview changed. The samples are small and come from one provider version each; they identify differences worth reviewing, without establishing overall fidelity.
Clothes colour: 6 of 7 plain tops are within ΔE 10 of the product's reviewed colour (ΔE 1.5 to 14.8). The 3 patterned tops are not measured.
Beauty shades: 6 of 8 readable results are within ΔE 5 of the requested shade (ΔE 2.3 to 7.4); 1 changed too little to measure.
Earring size: 1 of 2 earrings render in proportion to their catalog length, compared with the pair bought. The 38 mm Blue Crystal Drops renders 51% longer than its length implies. The size sent to YouCam does not control drop length, so Rechoice resizes an earring that renders out of proportion, from where it hangs, and measures it again: resized, 2 of 2 are in proportion. The shopper is told, and can see it as YouCam drew it.
ΔE is the CIEDE2000 colour difference: about 2 is just noticeable, under 5 is small and over 10 is clearly different. Colours are re-measured from the stored results, which the app saves as JPEG; for shades this differs from the live check, made on YouCam's original file, by up to ΔE 1.8, which is the measurement's own noise. Earring proportions are compared with the pair bought because the photo has no ruler. The correction uses catalog length ratios; it does not establish physical length or fit on a person.
Earring size, compared with the 58 mm pair bought
Earring
Catalog length
Expected
Rendered
Result
Resized by Rechoice
Baroque Pearl Drops
58 mm
1.00
1.00
reference
–
Blue Crystal Drops
38 mm
0.66
0.99
longer (+51%)
0.65 (-1%)
Blue Topaz Studs
12 mm
0.21
0.20
in proportion (-3%)
–
Beauty shades: requested and measured
Shade
Requested
Measured
ΔE
Live check
230 Terracotta
#b5532f
#aa4f37
4.1
4.6
220 Rosewood
#a04a55
#99474c
2.7
2.6
240 Mauve
#9c5a6e
#a25e69
3.8
4.6
220W Light Warm
#d9ae7f
#c49979
7.4
7.4
210C Light Cool
#dcb2a3
#caa18a
6.0
6.3
310C Tan Cool
#b8877a
too subtle
–
–
01 Coral Crush
#ef6f5a
#eb6655
2.3
1.9
02 Cherry
#b3203a
#a62f42
3.3
2.7
04 Lilac
#b9a3d6
#aa94ca
4.5
2.7
Clothes: product colour and preview colour
Top
Product
Preview
ΔE
Sage Knit Top
#607e77
#587470
4.0
Midnight Floral Blouse
#343a53
patterned, not measured
–
Ribbed Turtleneck
#69111d
#83091d
5.5
Indigo Denim Shirt
#25365e
#294072
4.1
Fair Isle Knit Sweater
#5a4438
patterned, not measured
–
Burgundy Knit Top
#533a3e
#57373a
2.7
Sunset Stripe Sweater
#6c4d5e
patterned, not measured
–
Heather Mauve Knit
#866b6c
#836563
2.7
Khaki Utility Shirt
#b39152
#b8914f
1.5
Sky Blue Tee
#3381ac
#48aee4
14.8
Measured 2026-10-28 by scripts/measure_pilot_accuracy.py from the full-size results.
Clothes pilot
Before building the shopper flow we ran 10 model-and-top pairs through YouCam AI Clothes (cloth-v4) on 2026-10-19. Every model is an AI-generated image from Pixabay; no real person appears. These are the real results, scaled down, not retouched.
Measured
Pairs tried
10
Succeeded / failed / unclear
10 / 0 / 0
Generation time
median 21.0 s, range 20.2–22.4 s
Units charged
20 (reconciled from unit history)
Gate
Proceed only if at least 8 of 10 pairs are useful enough for a style comparison, none materially misrepresents the garment, and delay and cost fit the owner-approved budget.
Owner decision
Passed the gate on 20 October 2026: the owner judged the pilot useful enough to continue.
Per-pair ratings
10 of 10 rated usable for a style comparison (the gate needs 8 of 10). Rated by Claude (AI assistant) at the owner's request, 2026-10-27.
Rating scale
Garment detail per aspect: 0 absent or misleading, 1 partly kept, 2 broadly recognizable. Person: any gross face, pose or body drift. Usefulness: usable, uncertain or misleading, with a reason.
Small sample, one provider version, our own catalog photos. It shows the integration works and where it struggles; it is not an accuracy benchmark.
Pair 1 · Plain knit; long sleeves drawn over bare arms
Model A (AI-generated, source)Catalog photo: Sage Knit TopYouCam result, 20.2 s, 2 units
Colour check: close to the product photo. (ΔE 4.0)
Rated usable: pattern 2/2, sleeves 2/2, neckline 2/2, person stable. Plain green knit, crew neck and long sleeves all kept; the heather texture survives.
Pair 2 · Busy floral print; does the pattern survive
Model A (AI-generated, source)Catalog photo: Midnight Floral BlouseYouCam result, 20.9 s, 2 units
Patterned top: the preview may simplify the pattern. Check the product photo.
Rated usable: pattern 2/2, sleeves 2/2, neckline 1/2, person stable. The floral on navy is broadly kept with smaller flowers; the shirt collar is drawn as a softer open neck.
Pair 3 · Turtleneck detail; source is on a dress form
Model A (AI-generated, source)Catalog photo: Ribbed TurtleneckYouCam result, 20.5 s, 2 units
Colour check: close to the product photo. (ΔE 5.5)
Rated usable: pattern 2/2, sleeves 1/2, neckline 2/2, person stable. Ribbed red turtleneck kept; the cuffs gain a small ruffle that the source does not have.
Pair 4 · Collar, buttons and chest pockets
Model B (AI-generated, source)Catalog photo: Indigo Denim ShirtYouCam result, 20.8 s, 2 units
Colour check: close to the product photo. (ΔE 4.1)
Rated usable: pattern 2/2, sleeves 1/2, neckline 2/2, person stable. Collar, buttons, both flap pockets and the yoke are kept; the sleeves are drawn rolled up.
Pair 5 · Fair Isle yoke; source includes mannequin legs
Model B (AI-generated, source)Catalog photo: Fair Isle Knit SweaterYouCam result, 22.0 s, 2 units
Patterned top: the preview may simplify the pattern. Check the product photo.
Rated usable: pattern 2/2, sleeves 2/2, neckline 1/2, person stable. Fair Isle yoke and fuzzy body kept; a grey layer shows at the neck and cuffs, likely from the mannequin.
Pair 6 · Plain dark knit on a dark-shirted model
Model B (AI-generated, source)Catalog photo: Burgundy Knit TopYouCam result, 21.0 s, 2 units
Colour check: close to the product photo. (ΔE 2.7)
Rated usable: pattern 2/2, sleeves 2/2, neckline 2/2, person stable. Plain burgundy knit kept closely despite the dark shirt underneath.
Pair 7 · Bold horizontal stripes
Model C (AI-generated, source)Catalog photo: Sunset Stripe SweaterYouCam result, 22.0 s, 2 units
Patterned top: the preview may simplify the pattern. Check the product photo.
Rated usable: pattern 2/2, sleeves 2/2, neckline 2/2, person stable. Navy body, red, yellow and teal bands, striped sleeves and ribbed cuffs all kept; the bottom band runs out of frame with the photo.
Pair 8 · Plain light knit
Model C (AI-generated, source)Catalog photo: Heather Mauve KnitYouCam result, 21.3 s, 2 units
Colour check: close to the product photo. (ΔE 2.7)
Rated usable: pattern 2/2, sleeves 2/2, neckline 2/2, person stable. Plain mauve knit, crew neck and long sleeves kept.
Pair 9 · Utility shirt with pockets; arm crosses the torso
Model C (AI-generated, source)Catalog photo: Khaki Utility ShirtYouCam result, 22.4 s, 2 units
Colour check: close to the product photo. (ΔE 1.5)
Rated usable: pattern 2/2, sleeves 1/2, neckline 1/2, person stable. Collar, pockets and buttons kept; it is styled open at the neck with rolled sleeves, which the product may not suit.
Pair 10 · Short-sleeve control case
Model C (AI-generated, source)Catalog photo: Sky Blue TeeYouCam result, 21.0 s, 2 units
Colour check: this preview looks lighter than the product. Trust the product photo for colour. (ΔE 14.8)
Rated usable: pattern 2/2, sleeves 2/2, neckline 2/2, person stable. Short sleeves and crew neck kept; the blue comes out noticeably lighter, which the colour check flags.
Beauty pilot
On 2026-10-27 we ran 13 shade-and-person pairs through YouCam Makeup VTO and Nail VTO, on AI-generated demo faces and a demo hand from Pixabay. Below are the three comparisons the shopper page makes when a shade looks too orange: the shade bought, then the two cooler shades it suggests. They replay as recorded results. The other 4 results stay private.
Measured
Pairs tried
13
Succeeded / failed / unclear
13 / 0 / 0
Generation time
16.9–18.0 s each
Units charged
13 (reconciled from unit history)
First probes
The first lipstick and nail probes were refused with InvalidParameters before any task existed, and were not charged. The adapter was fixed to send the fields YouCam's API definitions require, and both were run again.
Gate
Switch a feature on only if each of its results puts the colour on the right area alone (lips, skin or nails), reads as the requested shade, leaves the face or hand otherwise unchanged, and delay and cost fit the owner-approved budget.
Owner decision
On 27 October 2026 the owner chose to publish the lipstick, foundation and nail comparisons as recorded results. Per-result ratings are not yet recorded.
First reading by Claude (AI assistant), 27 October 2026, before the owner's ratings:
Lipstick: each shade sits on the lips only and keeps the lip outline; the three shades are clearly distinct.
Nail polish: every nail is painted with a gloss highlight, and the hand is otherwise unchanged.
Foundation: the colour stays on the skin and moves the right way, cooler for 210C and darker for 310C, but the three results are only 3 to 5 ΔE apart on the cheek, so the difference is subtle.
Makeup VTO brightens the whole image by about 5 L*, background included; showing the purchased shade on the same face cancels this out.
The shade check compares the colour that changed with the shade. It is not calibrated yet, so the shopper page shows it to staff only.
Lipstick on Face A
Face A as sent (AI-generated, source)230 Terracotta, the shade bought 17.7 s, 1 unit · shade check ΔE 4.6220 Rosewood 17.7 s, 1 unit · shade check ΔE 2.6240 Mauve 18.0 s, 1 unit · shade check ΔE 4.6
Foundation on Face A
Face A as sent (AI-generated, source)220W Light Warm, the shade bought 17.4 s, 1 unit · shade check ΔE 7.4210C Light Cool 18.0 s, 1 unit · shade check ΔE 6.3310C Tan Cool 17.9 s, 1 unit · too subtle for the shade check
Nail polish on Hand A
Hand A as sent (AI-generated, source)01 Coral Crush, the shade bought 16.9 s, 1 unit · shade check ΔE 1.902 Cherry 16.9 s, 1 unit · shade check ΔE 2.704 Lilac 16.9 s, 1 unit · shade check ΔE 2.7
Earrings pilot
On 2026-10-28 we ran 4 earrings through YouCam Earrings VTO on an AI-generated demo face, each from a product photo cropped to a single earring, with its catalog length sent as its size. Below is the comparison the shopper page makes for "too long, and I wanted silver": the pair bought, then the two silver pairs it suggests. They replay as recorded results. The other 1 result stays private.
Measured
Pairs tried
4
Succeeded / failed / unclear
4 / 0 / 0
Generation time
20.0–20.6 s each
Units charged
4 (reconciled from unit history)
Gate
Switch earrings on only if each result shows the earring on the ear at a believable size, keeps its metal and design, leaves the face otherwise unchanged, and delay and cost fit the owner-approved budget.
Owner decision
On 28 October 2026 the owner chose to publish the earrings comparison as recorded results. Per-result ratings are not yet recorded.
First reading by Claude (AI assistant), 28 October 2026, before the owner's ratings:
Each earring keeps the design and metal of its product photo.
Each hangs from the earlobe on both ears, from a photo of a single earring, and replaces Face A's own studs.
Size does not always follow the catalog length: measured against the 58 mm pearl drop, the 12 mm stud renders in proportion, but the 38 mm crystal drop renders about as long as the pearl drop. The previews show style and metal; the page says their size is approximate.
The face is otherwise unchanged.
Earrings on Face A
Face A as sent (AI-generated, source)Baroque Pearl Drops, the pair bought gold drop, 58 mm · 20.6 s, 1 unitBlue Crystal Drops silver drop, 38 mm · 20.0 s, 1 unitBlue Topaz Studs silver stud, 12 mm · 20.2 s, 1 unit
Hair colour pilot
On 2026-10-28 we ran 4 hair colour shades through YouCam AI Hair Color on AI-generated demo people. Below is the comparison the shopper page makes for "too brassy": the shade bought, then the two cooler shades. They replay as recorded results. The other 1 result stays private: The fourth pair, 9.1 Very Light Ash Blonde on Model C's dark curly hair, barely changed it: the engine tints hair rather than lifting it, so a shade far lighter than the hair in the photo does not show. It stays private.
Measured
Shades tried
4
Succeeded / failed / unclear
4 / 0 / 0
Generation time
17.2–17.5 s each
Units charged
4 (reconciled from unit history)
Gate
Switch hair colour on only if each result colours the hair alone, reads as the requested shade, leaves the face and clothes unchanged, and delay and cost fit the owner-approved budget.
Owner decision
On 28 October 2026 the owner chose to publish the hair colour comparison as recorded results. Per-result ratings are not yet recorded.
First reading by Claude (AI assistant), 28 October 2026, before the owner's ratings:
Each shade colours the hair alone; the face, skin and tee are unchanged.
7.3 Golden Blonde comes out a brighter, more saturated gold than its swatch; 7.1 Ash Blonde and 6.1 Dark Ash Blonde both come out as a cool grey-blonde and are close to each other, so the step between them is smaller on screen than in the catalog.
The shade check is not calibrated for hair: it reads the dominant changed colour, and shaded strands dominate, so it called the ash blonde dark. Its raw ΔE is recorded here but not shown to shoppers.
The fourth pair, 9.1 Very Light Ash Blonde on Model C's dark curly hair, barely changed it: the engine tints hair rather than lifting it, so a shade far lighter than the hair in the photo does not show. It stays private.
Hair colour on Model A
Model A as sent (AI-generated, source)7.3 Golden Blonde, the shade bought 17.5 s, 1 unit · raw shade check ΔE 16.1, not calibrated for hair7.1 Ash Blonde 17.2 s, 1 unit · raw shade check ΔE 31.0, not calibrated for hair6.1 Dark Ash Blonde 17.4 s, 1 unit · raw shade check ΔE 10.9, not calibrated for hair
Necklace pilot
On 2026-10-28 we ran 4 necklaces through YouCam Necklace VTO on an AI-generated demo person, each from a product photo of the necklace as worn. Below is the comparison the shopper page makes for "too long": the necklace bought, then the shorter ones. They replay as recorded results. The other 1 result stays private: The Rose Gold Leaf Pendant's task returned success but pasted a beige block over the chest: the engine's background removal failed on that product photo. The size check refused to read it, because the change covered 24% of the photo, and the result stays private.
Measured
Necklaces tried
4
Succeeded / failed / unclear
4 / 0 / 0
Generation time
18.3–19.6 s each
Units charged
4 (reconciled from unit history)
Gate
Switch necklaces on only if each result hangs the necklace at the neck at a believable length, keeps its metal and design, leaves the person otherwise unchanged, and delay and cost fit the owner-approved budget.
Owner decision
On 28 October 2026 the owner chose to publish the three necklaces that rendered properly as recorded results. Per-result ratings are not yet recorded.
First reading by Claude (AI assistant), 28 October 2026, before the owner's ratings:
Each necklace hangs at the neck over the tee, keeping its design and metal; the Y chain's pendant falls down the chest and the two collars sit at the collarbone.
Measured against the 55 cm chain bought, the 40 cm bib and the 45 cm collar both render in proportion.
Model A is otherwise unchanged.
The Rose Gold Leaf Pendant's task returned success but pasted a beige block over the chest: the engine's background removal failed on that product photo. The size check refused to read it, because the change covered 24% of the photo, and the result stays private.
Necklaces on Model A
Model A as sent (AI-generated, source)Amethyst Y Chain, the necklace bought silver drop, 550 mm · 19.2 s, 1 unitBloodstone Bib silver collar, 400 mm · 19.6 s, 1 unit · size check: in proportionLapis Collar silver collar, 450 mm · 18.3 s, 1 unit · size check: in proportion
Shoes and bags pilot
On 2026-10-28 we ran 4 pairs of shoes through YouCam Shoes V3 on a full-length AI-generated demo person, each from a product photo, and probed Bag V2 with one bag. Below is every pair of shoes, starting with the pair bought; they replay as recorded results. The bags probe failed its gate, so bag try-on is off: Bag V2 drew the red shoulder bag well, but it made a new picture around it: a gallery background, a green dress in place of the white tee and jeans, and a new pose. Its documentation describes preset styles, chosen at random by default, and a fixed output size. That cannot show a shopper's own bag on them, and each bag would come with a different outfit, so the comparison would not be fair. The other three bags did not run, and Rechoice compares bags by their catalog details only.
Measured
Shoes tried
4
Succeeded / failed / unclear
4 / 0 / 0
Generation time
18.9–20.6 s each
Units charged, shoes
8 (reconciled from unit history)
Units charged, bag probe
2 (reconciled from unit history)
Gate
Switch shoes and bags on only if each result puts the product where it is worn or carried, keeps its colour and design, leaves the person otherwise unchanged, and delay and cost fit the owner-approved budget.
Owner decision
On 28 October 2026 the owner approved up to 16 units, a probe first for shoes and for bags, and their publication. The shoes passed and are published as recorded results; the bags probe failed its gate, so the other bags did not run and bag try-on is off. Per-result ratings are not yet recorded.
First reading by Claude (AI assistant), 28 October 2026, before the owner's ratings:
Each pair of shoes appears on Model F's feet below the trouser hem, keeping its design and colour: the red pointed boots, the cream peep-toes, the metallic platform brogues with their striped sole and the nude pumps.
Model F, her trousers and the background are unchanged in all four.
The wide trousers cover the ankles, so the boots' shafts and the heel heights are mostly hidden; the previews show style and colour, and the page says to check the heel height in the details.
Bag V2 drew the red shoulder bag well, but it made a new picture around it: a gallery background, a green dress in place of the white tee and jeans, and a new pose. Its documentation describes preset styles, chosen at random by default, and a fixed output size. That cannot show a shopper's own bag on them, and each bag would come with a different outfit, so the comparison would not be fair. The other three bags did not run, and Rechoice compares bags by their catalog details only.
Shoes on Model F
Model F as sent (AI-generated, source)Red Heeled Ankle Boots, the pair bought bright colour, ankle boot, heel 100 mm · 20.6 s, 2 unitsCream Block-Heel Peep-Toes neutral colour, peep-toe, heel 50 mm · 19.0 s, 2 unitsMetallic Platform Brogues bright colour, brogue, heel 30 mm · 18.9 s, 2 unitsNude Suede Pumps neutral colour, pump, heel 110 mm · 19.1 s, 2 units
Why bag try-on is off
Model A as sent (AI-generated)Bag V2 result for the Red Slouch Shoulder Bag: new outfit, pose and background · 19.4 s, 2 units
Looks pilot
On 2026-10-28 we made 3 looks on an AI-generated demo person: each top through YouCam Clothes V4, then the lipstick from the same order through Makeup VTO, then the necklace from the order through Necklace VTO, each added to the result before. They show a choice with the rest of the order, beside the top bought made the same way, and replay as recorded results. The first probe was planned with the earrings as a third step. Earrings VTO refused the chest-up photo: "[PHOTO_CHECK_INVALID] Ear size should fit in range and position should be in the center." YouCam refunded that charge, and look 1 is that probe's first two steps. The necklace was added to each look afterwards, on its lipstick result, with look 2's crew neck as the probe; the top and lipstick were not made again.
Measured
Looks made
3
Succeeded / failed / unclear
3 / 0 / 0
Generation time
56.4–57.0 s per look, all steps
Units charged
12 (reconciled from unit history)
Gate
Show recorded looks only if each result keeps the top's design, puts the lipstick on the lips alone, hangs the necklace at the neck at a believable length, keeps the person recognisably the same from step to step, and delay and cost fit the owner-approved budget.
Owner decision
On 28 October 2026 the owner made looks the top and lipstick, after Earrings VTO refused the probe's photo, and chose to publish the three looks as recorded results. Later that day the owner approved adding the necklace from the order to each look, 1 unit each, and publishing them. Per-look ratings are not yet recorded.
First reading by Claude (AI assistant), 28 October 2026, before the owner's ratings:
Each top keeps its design: the turtleneck's ribs, the sage knit's heathered texture and the blouse's busy floral print.
The 230 Terracotta lipstick sits on the lips alone, added to the result with the top.
Model A stays recognisably the same person across the three looks and all three steps.
The Amethyst Y Chain hangs at the neck on each top, keeping its design and metal, with the top, the lipstick and the face unchanged by the third step.
On the turtleneck the chain sits below the rolled collar and hangs a little lower than on the crew neck, as it would be worn. On the floral blouse it is hard to pick out against the print.
Look 1 has a small pink smudge on the jeans near the hem, made by Clothes V4 in its step.
The shade check could not read the lips: on a chest-up photo they cover too few pixels at the check's working size.
Look 1: Ribbed Turtleneck with Velvet Matte Lipstick, 230 Terracotta and Amethyst Y Chain
Model A as sent (AI-generated, source)Step 1, Clothes V4: Ribbed Turtleneck 19.3 s, 2 units · from the probeStep 2, Makeup VTO: adds Velvet Matte Lipstick, 230 Terracotta 17.5 s, 1 unit · from the probeStep 3, Necklace VTO: adds Amethyst Y Chain 19.6 s, 1 unit
Look 2: Sage Knit Top with Velvet Matte Lipstick, 230 Terracotta and Amethyst Y Chain
Model A as sent (AI-generated, source)Step 1, Clothes V4: Sage Knit Top 19.1 s, 2 unitsStep 2, Makeup VTO: adds Velvet Matte Lipstick, 230 Terracotta 17.6 s, 1 unitStep 3, Necklace VTO: adds Amethyst Y Chain 20.1 s, 1 unit
Look 3: Midnight Floral Blouse with Velvet Matte Lipstick, 230 Terracotta and Amethyst Y Chain, what the shopper has now
Model A as sent (AI-generated, source)Step 1, Clothes V4: Midnight Floral Blouse 19.6 s, 2 unitsStep 2, Makeup VTO: adds Velvet Matte Lipstick, 230 Terracotta 17.8 s, 1 unitStep 3, Necklace VTO: adds Amethyst Y Chain 19.7 s, 1 unit