AI eyewear product photos: lens glare and frame edges
Glasses are hard for image models: thin frames and clear lenses. Send a sharp packshot and ask for the lenses and edges to stay unchanged, then inspect them.

For AI eyewear product photos, start from a sharp packshot of the real frame, ask the model to keep the frame shape, hinge, nose pads and lens tint identical, and change only the scene. Thin frames and clear lenses are where edits go wrong, so inspect the temple tips, the bridge and the lens reflections at full size.
What goes wrong first
Thin geometry has little room for error. Common faults are a bent temple, a bridge that moves, lenses that turn opaque, and a reflection that shows a scene that is not in the picture. Treat each as a checkpoint when you review.
A prompt that protects the frame
Name the parts. Say the frame silhouette, the bridge width, the temple length, the lens tint and the logo on the temple must not change. Then describe the new scene: a stone ledge, soft daylight, a neutral reflection in the lenses.
curl -X POST https://api.sume.com/v1/images \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"openai/gpt-image-2.5","prompt":"Keep the eyeglass frame identical: silhouette, bridge, temples, hinge, pads, lens tint and temple logo. Place it on a pale stone ledge in soft daylight, gentle neutral reflection in the lenses.","quality":"high","aspect_ratio":"4:5","input_references":[{"type":"image_url","image_url":{"url":"https://example.com/frame-packshot.png"}}]}'Choose resolution for review
Nano Banana 2 and Pro list 0.5K, 1K, 2K and 4K. A 4K frame lets you check the hinge and temple tip, and a 0.5K draft is enough to test the scene. Ask for a draft first and spend on the larger size only for the composition you keep.
| Model id | Resolution tiers | References |
|---|---|---|
| google/nano-banana-2 | 512, 1K, 2K, 4K | up to 10 |
| google/nano-banana-pro | 512, 1K, 2K, 4K | up to 10 |
| openai/gpt-image-2.5 | size presets or custom pixels | up to 16 |
Compare with the source
Place the output and the packshot side by side at the same scale and flip between them. If the frame changed, do not keep editing the output. Return to the packshot and tighten the preserve list. Chaining edits tends to stack small errors, as covered in the edit-drift guide.
Try-on is a different job
To show the frame on a face, use a try-on workflow with a face photo as a second reference. Keep consent for any real person's photo in mind, and do not use a customer's face without permission.
Related posts
More in Use cases
- AI backdrops for TikTok Shop: staging must match what you sell
TikTok Shop's page says LIVE backgrounds and staging items must match the products sold. How to review a generated backdrop and use video filter dim or crop.
- AI kitchen counter product photo: scale cues for the right size
Image models often guess size. Give a real dimension in the prompt and include familiar objects of known size next to the product to keep scale believable.
- AI music as the main focus of a Short: YouTube says to disclose it
YouTube lists music that is the main focus of a video as altered or synthetic content to disclose. How that applies to a Short built on a Sume music track.
- AI music video: a Lyria 3.5 track as the Timeline spine
Generate a song on the Music Router, then cut generated clips to it on Timeline 1.0. Track length is set in the prompt, not a duration field.
Written by Sume