Prompt-driven guide for building multimodal search using Gemini Embedding 2 + Pinecone + Claude Code. Includes example data (NASA public domain), step-by-step prompts, concepts explainer, cost breakdown, and troubleshooting guide.
4.3 KiB
Image Descriptions
Good descriptions are the most important part of making images searchable. The AI uses these descriptions to understand what each image shows. Better descriptions lead to better search results.
Below are descriptions for each image in this folder. We include both a "bad" and a "good" version so you can see the difference.
earthrise.jpg
Bad description: "Photo of Earth from space."
Good description: "Earthrise, photographed by astronaut William Anders during the Apollo 8 mission on December 24, 1968. Shows planet Earth rising above the lunar horizon, with the grey, cratered surface of the Moon in the foreground and the blackness of space behind. Earth appears as a blue and white marble, partly in shadow, with visible cloud patterns and ocean. This was the first photograph of Earth taken by a human from lunar orbit. It became one of the most influential environmental photographs ever taken."
Why the good version works: It includes the mission name (Apollo 8), the date, the photographer, what is visible in the image, and why the photo matters historically. A search for "first photo of Earth from the Moon" or "Apollo 8" or "William Anders" will all find this image.
aldrin-moon.jpg
Bad description: "Astronaut on the Moon."
Good description: "Buzz Aldrin working beside the Apollo 11 Lunar Module Eagle on the surface of the Moon, July 20, 1969. Aldrin is wearing a white spacesuit (A7L Extravehicular Mobility Unit) and is positioned near scientific equipment deployed on the lunar surface. The Lunar Module is visible behind him with its gold and silver thermal protection layers. The grey lunar soil shows footprints and equipment tracks. Photographed by mission commander Neil Armstrong. This was the first crewed Moon landing in history."
Why the good version works: It names both astronauts, the mission, the spacecraft, and the equipment visible. A search for "first Moon landing equipment" or "Apollo 11 Lunar Module" or "Buzz Aldrin" will find this image.
jupiter-great-red-spot.jpg
Bad description: "Planet Jupiter."
Good description: "Full-disk color portrait of Jupiter captured by the Voyager 1 spacecraft in 1979 during its flyby of the planet. Shows Jupiter's distinctive horizontal cloud bands in shades of orange, brown, and cream. The Great Red Spot, a massive storm larger than Earth that has been raging for hundreds of years, is visible in the southern hemisphere. Jupiter is the largest planet in our solar system, with a mass 318 times that of Earth. It is a gas giant composed primarily of hydrogen and helium."
Why the good version works: It mentions the spacecraft (Voyager 1), the Great Red Spot, the cloud bands, and key facts about Jupiter. A search for "largest storm in the solar system" or "gas giant cloud bands" or "Voyager Jupiter photos" will all find this image.
iss-over-earth.jpg
Bad description: "Moon and Earth from space."
Good description: "The Moon photographed from the International Space Station (ISS), showing a gibbous Moon suspended above Earth's atmosphere. Earth's surface is visible in the lower portion of the image, covered with clouds and showing the thin blue line of the atmosphere at the horizon. The photo demonstrates the perspective astronauts have from the ISS, orbiting approximately 400 kilometers (250 miles) above Earth's surface. The ISS has been continuously occupied since November 2000 and serves as a microgravity research laboratory."
Why the good version works: It describes what is actually in the frame (the Moon seen from ISS, not the ISS itself), includes the orbital altitude, and mentions the ISS as a research laboratory. A search for "view from the space station" or "Moon from orbit" or "Earth's atmosphere from space" will find this image.
Tips for writing your own descriptions
- Name what you see. People, places, objects, colors, positions.
- Add context. When was it taken? By whom? Why does it matter?
- Include facts. Numbers, dates, names. These make searches precise.
- Think about how someone would search. What question would lead to this image? Make sure your description contains those words.
- Be specific. "A 12-mile-high cliff on Miranda, a moon of Uranus" beats "a cliff on a moon" every time.