How We Use AI
Many dinosaurs and prehistoric creatures in our database have no illustrations available. Rather than leave these pages empty, we use AI to generate images for them. We want to be upfront about how this works and what it means.
Every AI image is labeled
All AI-generated images on the site carry a visible "AI Generated" badge. We don't mix them in with real illustrations or photographs without telling you. If you hover over the badge, you can see which model created the image and what scenario was depicted.
How the images are made
We don't just ask an AI to "draw a dinosaur." Each image goes through a multi-step pipeline designed to ground the results in real paleontological data.
- Research. For each creature, we feed an AI its known classification, geological period, diet, geographic range, and any traits from the Encyclopedia of Life. We ask it to compile a research profile: estimated size, distinguishing anatomy, likely skin covering (feathers vs. scales vs. armor, based on fossil evidence), habitat, and behavior.
- Accuracy requirements. The research step also produces a list of accuracy constraints specific to that species — things like "tail held off the ground," "no grass in a Jurassic environment," or "feathered forelimbs based on fossil impressions." These constraints are included directly in the image generation prompt.
- Scenario design. The research step generates five distinct scenarios for each creature: specific moments showing different behaviors, camera angles, and lighting conditions.
- Image generation. The final prompt combines the animal's physical description, the chosen scenario, the environment details, and all accuracy requirements.
What we try to get right
- Correct geological period and habitat (no anachronistic plants or terrain)
- Body proportions and posture informed by current paleontological understanding
- Appropriate integument (feathers, scales, or armor) based on fossil evidence for each species
- Behaviors consistent with what we know about the animal's diet and social structure
What can still go wrong
Despite the research step, AI image generators can and do make mistakes. Common issues include:
- Anatomically incorrect depictions
- Preference for dramatic teeth and claws
- Proportions that drift toward pop-culture depictions (Jurassic Park Velociraptor, etc.)
- Skin textures and color patterns that are plausible but ultimately speculative
- Anachronistic errors such as certain plants or animals that were not contemporaneous.
We think these images are better than having no image at all, but they are not scientific illustrations and they are not peer-reviewed.
Models used
Our primary image generation model is Google Gemini. Research is conducted using Gemini as well.
Open to feedback
If you spot an image that's clearly wrong — wrong number of limbs, feathers on something that definitely didn't have them, wildly incorrect body plan — we want to know. You can report individual images using the feedback tool on each dinosaur's page.