# What happens when you hand it two pictures > Two references aren't twice one reference. What the pair means depends on what the two images are — two photos combine, a design plus a photo swaps the subject, and the text decides. August 16, 2026 · https://justog.club/stories/two-reference-images Most reference-image advice stops at one picture: hand it a thing, get a version of that thing. The more interesting case is two, because a pair isn't twice as much input — it's a *relationship*, and the relationship changes what should come back. Two photographs of two dogs and a photograph plus an engraving are not the same request, and nothing about "two references" tells you which one you made. ## What can two references actually mean? Four relationships, and they're decided by what the two images *are*: - **Combine** — two photographs, or two images with no text between them. Two subjects, one composition, and the likeness of both has to survive. Two dogs. Two people. A person and their boat. - **Subject swap** — a finished design plus a photograph. Keep the design's whole grammar — its layout, its lettering, its texture — and put the new subject inside it. - **Style donor** — a photograph plus a design, where what you want from the design is only its look. The donor's own subject and words are discarded; you're taking palette, line, treatment. - **Fuse** — two finished designs. Explicitly: this from A, that from B. The pair is read before anything is drawn, which is why the same two files can mean different things depending on what you say about them. ## Who decides which one it is? Your words, always. Text is the highest authority in the read. That matters because the images alone are ambiguous. A photo and a poster could plausibly be *swap the subject into this design* or *give my subject this look* — those produce different results and only you know which you meant. Saying "in this style" or "make this the subject" resolves it in four words. Where you say nothing at all, the safe reading is the faithful one: two photos become two subjects, and a lone image becomes the design itself rather than raw material. ## What survives from a photo of a specific animal or person? Likeness, and it's held deliberately rather than hoped for. When a reference is a specific pet, person or mascot, the identifying features get named and pinned — the markings, the colouring, the silhouette, the shape of the face — and every variation carries them unchanged. That's the mechanical version of a promise most people don't believe you can make: **the dog stays your dog; the rendering is what varies.** With two subjects it applies twice, which is the whole difficulty of the combine case and the reason it's worth doing properly rather than collaging. ## Why do two subjects usually look wrong when people try it? Because they get rendered in two different techniques and read as a collage. The fix is a rule rather than a taste: **one rendering technique across both subjects.** If the terrier is engraved line work and the boxer is flat vector, you have two designs sharing a shirt. Same line weight, same palette, same treatment, and they become one picture with two animals in it. The other half is composition. Two subjects need an arrangement with a reason — side by side and symmetric, facing each other, one behind the other, or the old heraldic answer of two supporters flanking a centre. Chosen deliberately, that reads as designed. Left to chance, it reads as two photos that happened to land near each other. ## Does text inside a reference image get kept? Yes — word for word, and this one was learned the hard way. Any lettering visible in a reference is treated as fixed: the typeface can be restyled, the words cannot. That rule exists because an earlier version of the system helpfully rewrote the words on somebody's design into something that fit the mood better, which is not a thing anyone wants from a tool that was handed an existing piece of work. If you *do* want different words, saying so is what changes them. Silence means keep. ## Can I use one of my own old designs as the reference? That's the most useful version of this, and it's what the style-donor path is for. A back catalogue is a style: your palette, your line weight, your way of lighting a subject. Point that at a new subject and you get the thing this shirt is about, wearing the look you spent a year building. It's the same operation as remixing someone else's design, aimed at your own. For a small brand this is the difference between a folder of unrelated designs and a body of work, because [consistency of treatment is what makes a set read as a set](/stories/start-a-shirt-brand-with-ai). ## What if I have five photos, not two? Work in rounds rather than trying to submit them all at once. The draw takes a pair, so five images is a sequence: combine two, take the result forward as the new reference, bring in the third. That's a normal way to work rather than a workaround — each round is a decision point where you can look at what you've got and change your mind. The thing to avoid is treating a limit on one draw as a limit on the idea. Nothing stops a five-subject design; it just isn't a single button. ## How do I get one made? Bring the images to [JustOG](/) — the pair, plus a sentence about what they are to each other. Pick a direction, drag the crop frame, see it composited on the real garment, and it's made to order and shipped. Designs other people have published are in the [shop](/shop). Two pictures is a sentence, not a pile. The sentence is the part you write.