Most image AI generates a picture and hopes it is right. The Carnela system also checks it against the product data and the material library before anyone sees it, and learns from every check.

1. No product photos, only 3D data and material sheets

Carnela had no samples and no photos. Nine concept rounds narrowed the collection to 16 pieces, each fixed by its dimensions, four product views and a material sheet. The chair in the dining scene is the same chair as in the office scene. From 708 reference photos we kept 224 real photographs of the fabrics, stone and wood as the standard for every image. Brands with CAD or 3D files hand those over directly.

2. Checks every image, learns from every check

16 checks of two kinds run before anyone sees an image. Accuracy checks compare each product with its reference: shape, proportions, colors, the number and position of compartments and doors. Material checks use two vision models to compare every textile and surface with the real material photos. A textile fault blocks the image and only the marked area is regenerated. A fault on a hard material becomes a note for retouching. Every image goes through all 16 checks, whatever the volume.

Every review comment became a rule for all following images: no rug as the default floor, travertine in cool ivory with visible banding, lived-in rooms instead of catalogue polish, soft daylight for close-ups.

3. High-end quality at 100 %

The base model gets shapes right and guesses materials. Chenille turns into bouclé, velvet goes flat, patterned fur turns into a print. At 100 % crop that decides whether an image works for a premium brand.

The same image at 100 %, every pixel as it came out of the system:

4. A custom system teams operate themselves

Any format goes in, every format comes out. A designer starts from a sketch, a product manager from a CAD file, a marketer from a campaign photo. The system turns each into the same fixed product and delivers what each team needs: 2D drawings and material sheets for production, clay and 3D renders for sales and configurators, material close-ups for e-commerce, campaign images for social and print.

Briefs, shot lists and feedback are plain sentences. No prompt engineering, and every review improves the next image.

The result. A system that knows all 16 products and produces new imagery at the same high-end standard at any volume, with every product identical across all scenes.

Why it matters. A premium brand cannot afford a chair that changes between pictures. A system that knows the products, checks every surface and learns from each review holds that standard as volume grows.