Macgi
AI fashion photography prompts are only useful for an online clothing store if the output still looks like the product you will ship. A prompt that says “elegant blue dress in a sunny loft” may give you a lovely image, but the shade of blue, neckline, hem length, and fabric may all drift. This guide gives a repeatable structure for writing prompts for product scenes and model images, with camera and lighting terms, negative prompts, and a review process that compares the result against the physical garment.
Why free-form prompts create wrong garments
AI image tools do not see the actual product. They predict pixels from a text description. If the prompt says “a white linen shirt”, the tool will draw a generic white shirt: maybe with a chest pocket, maybe button-down, maybe curved hem. It will not know your shirt has no pocket, a straight hem, or a specific recycled-linen texture.
Free-form prompts cause the most damage when they mix product and scene language. For example:
“A stylish woman wearing a red summer dress in a Paris street, golden hour, photorealistic”
That tells the AI to make a flattering lifestyle image, not to preserve the dress. The model may be beautiful and the dress may be the wrong red, with added belt loops and different sleeve length.
A safer version:
“Product photo of a woman wearing the exact garment. Garment: midi-length A-line dress in matte terracotta red, round neckline, short sleeves, side-seam pockets, invisible back zipper, no waist seam, no belt, fabric is light cotton poplin with no print. Scene: quiet Paris side street. Camera: full body, eye level, 85mm lens. Lighting: soft overcast daylight. Style: catalog-ready, crisp fabric detail, do not alter garment.”
Notice the garment block comes first and the scene cannot change it.
Build a prompt in four blocks: garment, scene, camera, lighting
Garment block: describe the product, not the mood
Use details from the SKU name, size chart, material label or physical inspection. Include color as specific as possible — “matte crimson”, not just “red” — plus silhouette, length, neckline, sleeve type, closure, pockets, seams, print or pattern, fabric weight and drape. Avoid adjectives that describe how it should feel — “elegant”, “cozy” — unless you also lock the visible features.
Scene block: keep the scene replaceable
If you need a lifestyle scene, write the location, surface, furniture or backdrop. Then add “keep garment identical to the product” or “do not alter garment”. If color accuracy is critical, use a neutral background at first. Scenes with strong warm or cool light can shift the garment color, so control the lighting block.
Camera block
Use terms that match the product view: full body, three-quarter, chest-up, flat lay, ghost mannequin, detail shot. Add lens terms like 85mm for full-body model shots, 50mm for product ghost mannequin, 35mm for a wider scene but watch distortion. Specify eye level or slightly low angle. Avoid extreme closeups with a wide lens; they can distort proportions and construction details.
Lighting block
State the light quality and direction: soft diffused daylight, overcast, studio softbox at 45°, ring light, no harsh shadows. Add color temperature if needed: neutral daylight, 5500K, cool white. Avoid “moody” or “golden hour” when the goal is product color accuracy; they can add a yellow or orange cast.
A reusable template:
“Product image of [garment]. Garment: [specific color, material, silhouette, neckline, sleeve, closure, details]. Scene: [background or context]. Camera: [shot type], [lens], [angle]. Lighting: [quality, direction, color temperature]. Style: photorealistic catalog photo, sharp fabric texture, no alteration to garment.”
Photo: Antoni Shkraba / Pexels
Model image prompts that keep the product accurate
When you want a model image, tell the AI where the model can move and where the garment cannot.
- “Full-body model wearing the exact product. Model standing straight, arms slightly away from the sides. Front view showing full hem and sleeve length. Do not crop the garment. Do not change the neckline, seam position, button placement, pocket shape, or hem length.”
- “Model photo with the product unchanged from reference image. Scene: light gray studio backdrop. Camera: full body, 85mm, eye level. Lighting: soft diffused daylight. Negative prompt: altered garment, missing buttons, extra pockets, changed fabric, color shift.”
If your AI tool supports a reference image, upload the flat lay or ghost mannequin photo and write “keep the garment exactly as the reference image; change only the model, pose, background and lighting.” That is often the strongest way to preserve details. Some tools call this “image prompt”, “style reference” or “product reference”.
For poses, avoid hands or arms covering zippers, pockets, seams, or waistline. Specify “hands at sides” or “one hand not covering the closure”. If you need back views, generate a separate image with “back view, show back neckline, zipper, and hem” rather than expecting a single prompt to capture everything.
Negative prompts that prevent common clothing errors
Negative prompts are separate instructions that tell the AI what not to add. They work differently by tool, but a short focused list is usually more effective than a long list of every possible mistake.
A strong clothing negative prompt includes:
- “altered garment, changed silhouette, wrong neckline, wrong hem length, wrong sleeve length”
- “extra buttons, missing buttons, added pockets, added belt, added drawstrings, added logo, added text”
- “fabric texture changed, fabric weight changed, wrong material, changed opacity”
- “color shift, oversaturated color, underexposed garment, wrong colorway”
- “invented print, distorted pattern, broken seams, double stitching”
- “cropped garment, blurred fabric, low detail, plastic skin, warped hands, extra limbs”
Use only the negatives that match your product. If your item has no buttons, “extra buttons, button placement changed, missing buttons” is useful. If your item has a complex print, “distorted pattern, wrong scale, invented print” is essential. Include “no altered garment” as a broad negative if your tool supports it.
Photo: PNW Production / Pexels
Lighting and camera terms that help or hurt accuracy
Some terms make an image look good but harm product accuracy. Use this table as a quick reference.
| Term | When it helps | Risk |
|---|---|---|
| “soft diffused daylight” | True color, even fabric detail | May look slightly flat |
| “overcast sky” | Neutral white balance | Low contrast if exposure missed |
| “studio softbox at 45 degrees” | Controlled shadows, seam detail | Can make fabric look heavier |
| “neutral daylight 5500K” | Color-correct product shots | Less mood |
| “golden hour” | Lifestyle, emotional scenes | Can shift reds, blues, whites |
| “moody low-key” | Editorial campaigns | Risks losing fabric detail |
| “85mm lens” | Full-body model, realistic proportions | Less environment |
| “35mm lens” | Environmental story | May distort edges |
| “eye level” | Accurate garment proportions | Can feel plain |
| “low angle” | Heroic, fashion editorial | Can shorten or tilt proportions |
For product pages, prioritize color and construction. For social or lookbook content, you can trade some accuracy for mood, but keep the garment details consistent.
Check AI results against the real product
Do not approve an AI image because it looks beautiful. Approve it because it matches the physical garment. Use a side-by-side workflow.
- Open the original product photo and the AI image on the same screen.
- Zoom in on seams, hem, neckline, buttons, zippers and pockets.
- Compare the color against the source photo or the physical item under daylight. If your tool allows, use a color picker and compare hex values; small shifts are expected, large shifts are not.
- Check the print or pattern direction and scale. Stripes should run the same way. Floral placement should not be mirrored or invented.
- Look at fabric weight and drape. If a stiff denim jacket becomes soft and slouchy, reject it.
- Check proportions: sleeve length, dress length, shoulder drop, waist position.
- Record errors before batch generating. For example, if you generate 50 images and 6 have incorrect buttons, that is a 12% error rate in that prompt. Fix the garment description or negative prompt and regenerate.
A useful rule: if a customer service agent would need to explain that the image is “just a render”, the image should not be on the product page.
Put the prompt into your fashion workflow
Save a master prompt template in a shared document or spreadsheet. For each product, fill in the garment block from the size chart, material label and physical inspection. Generate product scene and model images in batches, but approve only after the side-by-side check.
If your source photos are weak, fix them first. A prompt cannot reliably recover a missing button or an inaccurate color if the reference image is poor. See AI Product Photography for Clothing: Keep Product Shots Accurate and Clothing Product Photography Setup: Small Home Studio on a Budget for the base image quality.
The same rule applies whether you upload the result to Shopify, WooCommerce, Haravan or Sapo: the product page image should document the garment, not invent it. Keep the AI scene for lifestyle galleries or lookbooks, and keep the main product shot as close to the physical item as possible.
When you have product images ready, consider letting shoppers judge the look on their own body before they buy. Rather than relying only on a static AI-model scene, a virtual try-on button next to “Add to cart” can let a shopper see themselves wearing the product in about 15–40 seconds. Macgi works with your existing product photos, does not store shoppers’ photos, and is free in beta for 1,000 try-ons. You can sign up for the Macgi beta if you want to test it on your store.
FAQ
Can I use the same prompt for different AI image tools?
You can reuse the garment, scene, camera and lighting blocks, but expect to adjust negative prompts and reference image features. Some tools handle “keep garment identical” well, others need stricter wording like “same neckline as reference”.
How many negative prompt terms are too many?
A focused list of 10–25 terms usually works better than 100. Too many can contradict the main prompt or make the tool ignore important instructions. Prioritize the features that cause returns for your category: closures, color, print, and hem shape.
Should AI model images replace real product photos?
No. Use AI model images for lifestyle content, lookbooks, or social ads if they pass a product accuracy check. The main product page image should usually remain a real photo or a carefully checked AI edit of the real product.
What should I do if the AI keeps changing the color?
Add a color reference: “keep the exact color of the reference image” or name the color more specifically, such as “matte terracotta #B85C38”. Reduce mood lighting terms like “golden hour” or “warm sunset” that can shift the white balance. Then compare against the physical product under neutral daylight.
