Overview & result
Wes Bos demonstrates a Jev Score question over a clothing image. The question defines shirt, jacket and pants as its criteria, and the shared result favors jacket. Jev reads the supplied image and returns category scores; it does not generate or edit an image. The screenshots show one example, not a dataset evaluation.
From the original post
Jev images is here
One question about an image that is 850x850 cost one hundredth of a cent - $0.000096936
about 10,000 images for a dollar.
referencing the image more than once duplicates the token usage - so asking two questions in a single request is equivalent to uploading two images
How Jev fits in the loop
- Supply a clothing image as a reference inside the question instructions.
- Ask what type of clothing it is, using a Score question with shirt, jacket and pants as the criteria.
- Inspect the returned category distribution and confidence. The author's result shows jacket at 98% and confidence at 97%; these describe one prediction, not measured accuracy.
How to reproduce
- Open the original post and inspect both the Questions screenshot and the result screenshot.
- With image input enabled in the Jev Playground, attach your own clothing image and reproduce the shown Score question: what type of clothing is this? Set its criteria to shirt, jacket and pants.
- Compare the returned category scores with the image yourself. For an API implementation, follow the official image-input documentation; this post supplies no repository or executable project.
Why this build matters
Provides a concrete question-and-result example for developers exploring category tagging of product imagery. The next step is to test their own categories and images before using the scores in a workflow.
Limitations
- The demonstration has not been independently reproduced by JevForAgents; no test dataset or measured accuracy is provided.
- The author reports $0.000096936 for one question about an 850 × 850 image and says repeated image references duplicate token usage. These are author-reported billing observations, not our measurements or a guaranteed per-image price.
- Native image input is experimental. Consult the current official model and image documentation before implementation.
- The original post contains two screenshots and no video, application link or GitHub repository. Its exact UTC time is not returned by the visible page.