The real product in a generated world

The workflow that does most commercial work

The brief block established it: generate the world, photograph the claim. This lesson is the execution, and it is the single most commercially useful technique in this course.

The result is a product shot that could not have been photographed (a bottle on a mountain road at dawn, a jar in a 1950s kitchen) in which the product is genuinely, exactly, the client’s product.

Step one: reserve the space

Generate the environment with a place for the product. Two ways:

• Prompt for it: an empty wooden table in the foreground, morning light from the left, shallow depth of field. Leave the space empty rather than generating a similar object and covering it, because a generated object’s shadow will show around your real one.
• Blockout it: a cylinder in Blender at the product’s position and size, depth-conditioned. This gives you correct perspective and, usefully, tells you exactly where the product must sit.

Decide the light direction now and write it down. Everything downstream depends on it.

Step two: photograph the product to match

The photograph must match the scene’s light, and this is where people fail. If the generated scene is lit hard from the left, the product must be lit hard from the left.

With a window and two sheets of paper:

1. Product side-on to the window, matching the scene’s direction.
2. White paper on the shadow side if the scene’s shadows are soft; remove it, or use black card, if they are hard.
3. Plain background: white paper, or black cloth.
4. Match the camera height and angle to the scene. If the generated table is seen from slightly above, photograph the product from slightly above. Mismatched viewpoint is unfixable in an editor.
5. Match the focal length roughly. A scene prompted at 35mm wants a product shot from a similar distance, not a phone held at arm’s length in portrait mode.
6. Thirty frames, keep two.

Step three: cut it out

Background removal, then edge work. Free tools handle this well: GIMP’s Foreground Select, Krita, Photopea, or rembg, which is a small free model that does a good first pass in one command.

Then clean the edge by hand. A 1-pixel refinement around a bottle is what separates a composite from an obvious paste. If the product has a transparent or reflective part (glass, chrome) the background will show through it and the cutout must preserve that, which is hand work and takes time.

Step four: place, and then do the four things that matter

Placement is the easy part. These four are what make it believable.

Contact shadow. The single most important element, and the one people forget. Where an object meets a surface there is a dark, tight, soft-edged shadow. Without it the product floats, and every viewer feels it even if they cannot name it. Paint it on a new layer in Multiply mode with a soft black brush at low opacity, hard and dark directly under the base, fading outward within about 10% of the object’s width.

Cast shadow. The longer shadow thrown across the surface, in the scene’s light direction, softening and lightening as it recedes. Match its angle to every other shadow in the frame.

Reflection and bounce. On a glossy surface, a faint flipped copy at low opacity. From a coloured surface, a hint of that colour on the product’s lower edge.

Grain and optics. The generated scene has grain, a particular sharpness and probably some depth-of-field falloff. Your photograph does not. Match them: add the same grain, blur the product slightly if it sits in a plane that should be soft, and match the colour temperature.

The check

Zoom out to 25%, then look at the small version. Composites fail at small size before they fail at large size, because the eye reads overall integration rather than edges.

Then the three questions: does the shadow point the same way as every other shadow; is the product the size it would be next to the objects around it; does it have the same grain and sharpness as its surroundings. Two of those will be right and one will not.

Where this is the honest choice

This is not a trick to make people think a photograph was taken. It is the standard composite workflow of commercial photography, done since long before generative models, with a generated backdrop instead of a painted or photographed one.

Two lines are worth keeping. The product must be the actual product, unaltered in a way that misrepresents it: no slimmed bottle, no enhanced colour that the item does not have. And the environment must not imply a claim that is false: a food product photographed in a generated restaurant that does not exist, presented as the restaurant serving it, is a different matter from a jar on a generated marble surface.

Where exactly that line sits depends on the market and on advertising rules that differ by country. The business block returns to it. The practical version: if the environment would be understood as a stage set in a photograph, it is fine generated. If it would be understood as documentary evidence, it is not.