How AI Reads Your Pet's Photo to Reveal Their Past Life

When you upload your pet's photo to Paw Labo, something interesting happens before any story is generated. A large AI model — specifically Google's Gemma 4 31B, a multimodal model capable of analyzing both images and text — examines the photograph in detail. The story it builds is not random. It is anchored to what the model actually observes.

What the Model Looks For

The model is instructed to read the photo carefully before generating anything. Specifically, it looks at:

• Breed and physical type — A Greyhound carries different historical associations than a Saint Bernard. A Siamese has a different presence than a tabby. The model uses breed as one anchor point for the era and geography of the past life.
• Posture and bearing — Is the animal relaxed or alert? Sprawled or upright? Posture carries personality, and personality carries history. An animal photographed mid-stride, head raised, suggests a different past life than one curled in a tight ball in a sunbeam.
• Gaze and expression — The direction of the gaze, the openness or half-closure of the eyes, the presence of what photographers call "the decisive moment" in the animal's expression — all of this feeds into the model's interpretation.
• Surroundings — A pet photographed against a wood cabin background versus marble floors versus a garden will receive subtly different readings, because the model incorporates the visible environment.
• Lighting and color — Warm golden light versus cool grey; high contrast versus soft — these influence mood, and mood influences the era and character the model reaches for.

Why the Echoes Feel Personal

Each reading includes a section called "Echoes" — two specific connections between something visible in your pet's photo and a parallel from their past life. The model is explicitly instructed to tie these to details it actually observed in the image, not generic observations. If your dog has a distinctive white patch on its chest, the model may connect that to the crest on a knight's tunic. If your cat has unusually large ears, the model might connect those to the acute listening required by a spy or a scout.

This is why two photos of the same breed will often produce different readings. The model is not applying a breed template. It is reading the specific animal in the specific photograph. Eyes are particularly rich with information — what your pet's gaze reveals about their past life goes deeper than breed alone.

The Limits

This is entertainment, not prophecy. The model is a language model trained on human-generated text — it has absorbed a vast amount of historical knowledge, narrative pattern, and cultural reference, and it applies those creatively. The connections it draws feel meaningful because meaningful storytelling is what it has learned to do.

Whether any of it reflects something genuinely true about your pet's soul is not a question AI can answer. That part is yours to decide.