The background you're looking at isn't part of it
On the tablet the bird sits on cream. On the shirt it sits on teal, because the cream was never the design. That's the most common surprise in this whole process, and it's a good one.
Look at those two carefully. On the tablet, the bird is on a warm cream field. On the shirt, the same bird sits directly on teal cotton — no cream, no panel, no rectangle.
Nothing was edited. The cream was never part of the design; it was the surface the drawing was made on.
Why does the background disappear?
Because a print's edge is the artwork's own outline, not the file's.
Every generated image arrives with something behind the subject — a colour, a gradient, a scene. Separating the subject from it is a step that happens on this side, so what reaches the garment is the bird itself, with cloth continuing where the feathers stop.
That's also why keeping a rectangle is the fastest way to make a shirt look cheap: a printed background announces that the design was a picture before it was a garment.
Do I have to remove it myself?
No. It's the single most common thing people prepare unnecessarily.
There's no need to produce a transparent PNG, run a background remover, or ask the model for one. Handing over the version with its background intact is usually better, because the separation can then be done properly instead of being baked in along soft edges.
Is Gemini a good choice for this?
For flat illustration, unusually so — and it's one of the models you can pick here.
Generation runs on more than one image model and the choice is yours at the point of generating. Gemini's output leans toward clean editorial illustration: flat colour areas, defined shapes, restrained palettes, minimal rendering. That's very close to what a garment print wants anyway — hard edges, few colours, a clear silhouette.
The bird above is a good example of that register: five colours, no gradients, an outline you could cut around.
When would I pick a different one?
When the design needs texture, atmosphere or photographic realism.
Different models have different centres of gravity. Flat and graphic is one; painterly, detailed and photoreal is another. If what you want is a heavily rendered creature or something that looks photographed, a model that leans that way will get there with less fighting.
The useful habit is to treat the model as one more variable rather than a fixed setting — the same description produces meaningfully different results, and trying two is cheap.
Does the garment colour change the design?
Enormously, and this is the second thing the picture is showing.
The bird was drawn against cream and is being worn against teal. Those are different objects: the teal now functions as part of the composition, sitting behind the bird as sky-ish negative space, and the whole palette relationship shifts.
Which is why choosing the garment colour is a design decision rather than a final preference — the ground was always going to be half of it.
What if I want a background?
Then it has to be a shape rather than a rectangle.
A sunburst behind the subject, a rough brush slab, an arched panel — those are backing shapes that belong to the artwork and have their own outline. What doesn't work is the file's own background carried across as a square.
The distinction is whether the shape was designed or just inherited.
Does this apply to photos as well?
Yes, and more carefully.
A photographic subject lifted out of a real scene loses the lighting that made it sit there, which often reads as pasted. For those, redrawing is usually the better route — the subject survives, the scene doesn't.
For flat illustration like the bird, separation is clean because the edges were already defined.
How do I get one made?
Describe the idea at JustOG, pick a model and a direction, choose the garment colour, drag the crop frame, and see it composited on the real thing. It's made to order and shipped.
Designs other people have published are in the shop.
The cream was the paper. The shirt is the paper now.