Why Your Flat Lay Determines AI Output Quality
AI on-model photography is not a magic filter applied on top of a bad photo. The AI needs to understand the garment: its edges, fabric weight, texture, drape potential, and colour, before it can accurately render it onto a model. If your flat lay obscures any of that information, the AI either interpolates (introduces guesswork) or fails to render it at all.
The two most common causes of poor AI output are not lighting direction or camera model, they are incorrect background choice and garment bunching. Both are completely preventable.
The input-output relationship: In our production pipeline, flat lays submitted to our own brief outperform unguided flat lays on final image quality scores by an average of 40%. The brief takes under 10 minutes to follow. The improvement in output is consistent across garment categories.
Background Setup
Background choice is the single most impactful variable in flat lay preparation for AI photography, and the most frequently misunderstood.
Use mid-tone grey, not white
The instinct for most brands is to shoot on white: it looks clean, it is the marketplace standard for final delivery, and it photographs consistently. For AI input, white is the wrong choice. Here is why:
AI models identify garment boundaries by detecting the contrast edge between the garment and the background. Pure white backgrounds (or backgrounds exposed to white during shoot) cause the bright areas of light-coloured garments, pale pink, cream, white, ivory, to blend into the background. The AI sees a garment with undefined edges and must guess where the fabric ends. This introduces distortion in the final rendered image, particularly on shoulders, collar lines, and hems.
A mid-tone grey (around 40–60% brightness, or roughly RGB 100–160,100–160,100–160) provides sufficient contrast for both light and dark garments without creating the compression problems of a black background on dark navy, charcoal, or black fabrics.
Matte surface, not glossy
Glossy backgrounds (laminated boards, reflective card) cause specular reflections that appear as false bright spots in the image. The AI interprets these reflections as part of the garment and introduces artefacts accordingly. Matte surfaces: cartridge paper, foam board, fabric, produce a consistent tone that reads as pure background throughout the frame.
Background size
The background should extend at least 15–20 cm beyond the garment on all sides. Garments shot tight to the edge of the background create hard vignettes or colour bleed from the shooting surface underneath, both of which confuse AI edge detection on hem and sleeve ends.
Lighting
The goal in lighting a flat lay for AI is simple: even, shadowless illumination that reveals fabric texture without creating high-contrast shadows that the AI would interpret as structural garment detail.
Diffused overhead lighting
Position your light source directly above the garment and diffuse it, through a softbox, a shoot-through umbrella, or a large piece of diffusion material. The light should fall evenly across the garment from edge to edge. Check by shooting a flat grey card: the card should appear one consistent tone across the entire frame.
A single overhead diffused light source is sufficient. Two lights positioned symmetrically either side of the shooting axis (45 degrees left and right) also works well and eliminates the slight shadow that can appear under fabric edges with a single overhead source.
What to avoid
- Side lighting: Creates directional shadows across fabric texture that the AI reads as structural wrinkles and renders accordingly on the model
- Window light: Changes throughout the day and produces a warm/cool colour gradient across the frame, inconsistent between shots in the same batch
- On-camera flash: Creates a central hot spot and falls off at the edges, producing an unevenly lit garment
- Strong shadows under collar or sleeves: These read as depth cues and distort the AI's interpretation of garment construction
Colour temperature
Set your camera white balance to match your light source, typically 5500K for daylight-balanced strobes or LED panels. Mixed colour temperatures (one warm light, one cool) create a colour cast gradient that makes fabric colours appear to shift across the garment. The AI calibrates colour accuracy from your input file; any colour drift in the flat lay becomes colour drift on the model.
Garment Positioning
Positioning determines whether the AI understands the garment's intended silhouette or has to reconstruct it from context. Follow these rules by garment category:
Tops, shirts, and blouses
- Lay completely flat: no bunching, folding, or creasing in the body
- Fold sleeves in neatly at the sides, parallel to the body of the garment
- Straighten and open the collar or neckline: it should be clearly visible and symmetrical
- Smooth any manufacturing fold lines with a steamer before shooting
- Buttons: fasten the top two buttons, leave the rest open to show the placket cleanly
Trousers and skirts
- Lay both legs or the skirt flat, no crossing or folding
- Smooth the waistband flat: it should be horizontal and clearly defined
- For wide-leg or flared silhouettes, open the leg fully to show the intended shape
- Zip or fasten the closure and smooth the front panel flat
Outerwear and jackets
- Lay fully open: do not fasten the zip or buttons
- Fold cuffs back one turn to show the lining and sleeve construction
- If the garment has a hood, lay it flat behind the collar, folded down, symmetrical
- Shoulder pads or structured shoulders: shape these correctly before shooting, the AI renders the silhouette it sees, not an idealised version
Knitwear
- Steam thoroughly before shooting: any wrinkles become permanent in the AI rendering
- Lay on a flat surface with no stretch distortion; knitwear draped over edges reads as a draped garment to the AI
- For chunky knits, ensure the cable or rib pattern is fully visible and not compressed against the background
Camera Setup and Shooting Angle
Shoot from directly overhead: the camera axis should be perpendicular to the garment surface (90 degrees). Any angle deviation from true overhead introduces perspective distortion: one shoulder appears larger than the other, the hem appears closer than it is, and the overall proportions are skewed. The AI corrects for some distortion but cannot fully reverse a significantly angled shot.
Recommended settings
| Setting | Recommended | Avoid |
|---|---|---|
| Aperture | f/8–f/11 | f/2 or below (shallow depth of field blurs fabric edges) |
| ISO | 100–400 | ISO 1600+ (noise degrades fabric texture) |
| White balance | Fixed to light source | Auto WB (shifts between shots) |
| File format | RAW + JPEG, or PNG | Compressed JPEG at quality below 80% |
| Lens | 50mm–85mm (or equivalent) | Wide angle (distorts garment proportions) |
File Format and Resolution
Send your flat lays at the highest resolution your camera captures. The minimum useful resolution for AI on-model processing is 2000px on the longest edge, but more is always better. A 24MP camera file (typically 6000×4000px) gives the AI substantially more fabric texture data to work from than a compressed 2MP image.
Format preference
- PNG: Lossless: preferred when you have the option. No compression artefacts in fabric texture.
- JPEG: Acceptable at quality 90 or above. Avoid quality settings below 80: compression blocks appear in fine texture like linen, jersey, or silk weave, and the AI reads these as fabric pattern rather than compression noise.
- TIFF: Accepted and lossless: larger file sizes but no quality compromise.
- RAW: Accepted directly. Our pipeline converts internally. Preferred if you want maximum control over colour accuracy.
The Most Common Mistakes
These are the flat lay problems that most frequently produce suboptimal AI output, in order of impact:
- White background: Light garments lose edge definition against white, the most common cause of blurred or incorrect garment outlines in AI output
- Garment bunching or unsteamed creases: The AI renders what it sees; creases in your flat lay become creases on the model
- Angled camera shot: Even 10–15 degrees off-vertical creates proportional distortion the AI cannot fully correct
- Mixed light sources: Results in colour cast gradients across the garment that become incorrect colour rendering on the model
- Tight crop: Garment edges clipped by the frame give the AI no boundary context, always leave clear space around all garment edges
If your existing flat lays already have some of these issues, our fashion photo retouching service can correct background values, colour casts, and exposure before AI processing, saving you a reshoot.
Frequently Asked Questions
A neutral mid-tone grey (around 50% brightness) is the best background for flat lays destined for AI on-model processing. Pure white causes blown highlights that remove the fabric edge detail the AI needs to understand garment boundaries. Pure black compresses dark garments into the background. Mid-grey preserves clear edges across all fabric colours and tones.
No. A modern smartphone camera shooting at its highest resolution produces adequate flat lay images for AI on-model processing. What matters most is consistent lighting, a flat surface, a neutral background, and correct overhead positioning, not camera sensor size. A DSLR or mirrorless with a 50mm or 85mm lens will produce sharper results, but smartphone flat lays with good lighting regularly produce excellent AI output.
Send JPEG, PNG, TIFF, or RAW files at the highest resolution your camera captures. Minimum useful resolution is 2000px on the longest edge, more is better. Avoid heavy JPEG compression (quality below 80%) as it degrades the fabric texture detail the AI uses to understand material type. PNG is preferred when available as it is lossless.
Lay the garment completely flat with no bunching, creasing, or folding. For tops and jackets, fold the sleeves neatly at the sides, parallel to the body. For trousers, lay both legs flat and parallel. The garment should fill 70–80% of the frame. Shoot from directly overhead (90 degrees above). Any tilt in the camera angle creates distorted proportions the AI must interpret rather than model accurately.
Yes: existing flat lays work as AI input. Output quality depends on input quality. A flat lay with even lighting, a neutral background, and visible fabric texture will produce a better AI result than a heavily styled or composite flat lay. If existing images have strong colour casts, harsh shadows, or are low resolution, reshooting to the recommended specification will meaningfully improve AI output quality.