Brand, packaging, web and product design
packaging, sites, and the work that got cut
Four projects. Two are brands for products I make and sell, one is a set of sites I designed and built, and one is a writing app I specified and never built. The rejected work is here too, because the reasons it got rejected are the part worth reading.
Two of these are products I also had to make and sell, so none of the constraints were hypothetical. A lid has a die line and a minimum print run. A nutrition panel has a format I don’t control and numbers I have to stand behind. The sites came with clients, deadlines, and somebody else’s audience to keep. The last is a writing app I only specified, and the constraint there was a keyboard that already means something to everyone who would use it. Most of what these pages show is what those constraints ruled out.
The method
353 generations, and the ones i threw out
Kova’s lifestyle and product imagery is generated. I swept 353 favorites out of Higgsfield, kept 352 of them in an approved folder, and cut a 59-variant campaign set down from there. Everything in this section is generated, and I am saying so because a page about design should be clear about which pictures a camera took.
Eight things it made
All eight came out of the same pipeline: the mark in felt and in clay, a booth, two packaging renders, a courtyard set, a studio frame, and one frame from the campaign below.
The contact sheets are the rejected work. I am leaving them up because a set of four finished pictures says nothing about how they were chosen.
Two of the four proposed locks hold up
The other two do not, and this is the part worth reading
Both of the failed frames above were written down as clean in my own notes. So I went back and checked all 59 frames of that campaign set at full size. Twenty-two of them carry something I did not put there. Six are a legible Nike swoosh on the tank. The rest are an invented logo on the garment or garbled lettering on the cup. My notes on that folder flag three frames. Two of those are real, one is clean, and twenty more are not in my notes at all. I found them by rendering every frame at full size instead of cropping to where I expected a problem to be. A cropped check tells you about the crop.
The set that shipped went through five rounds
The sweep is one wide pass. The product images on the store are the other kind of work, and every round of those was a specific rejection. One cup read too tall. One wall came out kraft tan instead of brown. One set of scoops read tan when the flavor is vanilla. I fixed each one by adding a reference for the thing that was wrong — the approved fudge hero as the composition target, then the vanilla cup as color truth — instead of rewriting the prompt and hoping.
One round reversed a call I had already made. The rustic scraped texture looked better on its own and was wrong for the brand, so the original smooth scoop became the texture master and both images were rendered again against it. That version was thrown out after it was staged and before it shipped.
The second picture is a different answer to the same brief: the brand as painted illustration instead of a photoreal render. I did not ship it. It reads luxurious, and I want Kova to be fun and approachable.
This is also why the bear never comes out of a generator. A mark that has to stay exactly itself does not go through a process whose whole nature is to approximate.