Image (drawing) search
Find patents whose drawings look alike from a single image. A product photo, a paper figure, a hand sketch, or another patent's drawing — give it an image and FindIP returns publications whose shapes and structures resemble it. This shines for mechanisms, structures, and circuit layouts that are hard to put into words.
Two ways to use it
① Image alone — pure visual similarity
An image by itself returns drawings that look alike, ranked by visual similarity. Use it to find "something shaped like this" when you don't know the field.
② Image + text — visual and on-topic
Add a sentence and FindIP re-ranks the visually similar hits by your topic. Use it to keep only patents that both look alike and sit in the right field.
A real example
Real results from searching with the cross-section drawing of an all-solid-state battery electrolyte membrane below.

① Image alone — lookalike drawings come back from any field. All three below are cross-section / layered line art visually similar to the input.



It matches on shape alone, so it spans fields. Add text to narrow by topic (②).
② The same image + "all-solid-state battery electrolyte membrane layered structure" — re-ranked to visually similar patents that are also on topic.



Now it narrows to electrochemical, layered-structure battery patents.
Prompt examples
Attach an image (or pass an image URL) to your AI assistant (ChatGPT / Claude) and ask:
Image alone
"(drawing attached) Find patents whose drawings look visually similar to this figure."
Image + topic
"(drawing attached) Find patents that look like this figure AND relate to the layered solid-electrolyte membrane of an all-solid-state battery."
When to reach for it
- Reverse-tracing lookalike-structure patents from a competitor product photo
- Finding prior drawings that resemble a paper or conference figure
- Locating mechanisms and shapes (connectors, hinges, patterns) that resist keywords
Note
Figures that are more symbol- or text-driven than shape-driven — flowcharts and block diagrams, chemical structure formulas, graphs — aren't a good fit for visual-similarity matching. It works best on drawings with distinct shapes such as mechanisms, structures, and circuits.