An Honest Answer

Are AI cat portraits realistic?

Sometimes. It depends almost entirely on whether the portrait was painted from your cat or merely painted about a cat — and that is a difference you can see. Here is what separates the two, and what we do about it.

The real question is likeness, not realism

Nobody ordering a Renaissance oil painting of their cat wants a photograph. They want a painting that is unmistakably their cat. So the useful question is not “does this look real?” but “would someone who knew this cat recognise them instantly?”

That is a much harder standard, and it is the one most AI pet portraits fail. A model working from a text description will happily produce a beautiful, competent, entirely fictional tabby. It will be realistic. It will not be your cat.

What makes fur, whiskers, and eyes read as authentic

Fur. Coat pattern is identity, not texture. The exact break of a tabby’s stripes, where a bicolour’s white ends on the chest, the one off-colour patch above the eye — these are what a person recognises before they consciously look at anything else. Fur length and texture matter too: a plush longhair rendered with short-coat shading stops being the same animal, even at the same colour.

Whiskers. Whiskers are the most common casualty, because they are thin, light-coloured, and cross the background. They should read as a small number of distinct, gently curved, individually placed hairs coming from real whisker pads, catching the light of the scene. When a portrait renders them as a fuzzy halo, or drops them altogether, the face goes flat and slightly wrong in a way most people feel before they can name it.

Eyes. Eye colour is not decorative — it is one of the fastest identity cues there is, and a green-eyed cat given amber eyes is somebody else’s cat. Beyond colour, the eyes need the right shape and set for that individual, a pupil sized to the light in the scene, and a catchlight that agrees with where the light is coming from. A catchlight in the wrong place is the single most common tell of a face that was invented rather than observed.

The face underneath. Skull shape, muzzle width, ear set and proportion. The strong temptation for any generative model is to average toward a prettier, more symmetrical cat. Resisting that is most of the work.

What we do about it

We paint from the photo, not from a description. Your portrait is rendered image-to-image: your photograph is the reference the painting is constructed on. The style brief governs medium, light, palette, and setting — never who the cat is.

Identity is written into every style. Every one of our styles inherits the same non-negotiable instruction before any aesthetic is applied: preserve the coat pattern and markings, the fur length and texture, the eye colour, the face shape and proportions. No warped anatomy, no extra limbs, no invented features.

Every render is checked before you see it. Each portrait is scored 1–10 against your original photograph by a separate vision model asking one question: is this the same cat? A render below our threshold is automatically regenerated once and the better result is kept. If it still falls short, we show you the best we have and flag the session for a person to review.

You judge it, free, before any money moves. Fifteen watermarked previews per visit and no checkout until you have approved one. We would rather you reject three rounds than receive a canvas of a cat that is nearly right.

Where AI portraits genuinely fall short

Anything the photograph did not record cannot be recovered. A photo where one side of the face is in deep shadow gives the engine nothing to work from on that side. A distant or blurred photo cannot yield crisp whiskers. Cats with very subtle, high-frequency markings — dense torbie and tortoiseshell coats especially — are the hardest to reproduce exactly, because there is more identity information packed into the coat than into the face.

This is also why we do not sell a single render. See how it works for the whole process, step by step.