I Tested Jev 1.14 on 32 Thumbnails (And It Corrected Me)!

Jev 1.14 image analysis and the new Box question tested on 32 of my own YouTube thumbnails. TypeSafe's Jev can now find things in pictures, and it corrected my answer key twice. Bounding box accuracy, speed, cost, confidence and the docs' recrop fix, measured against the exact boxes read from the thumbnail HTML. ---- Higgsfield AI — Create AI Videos & Visuals Get 10% OFF here: https://higgsfield.ai?fpr=thomas-dccf7e 🚀 DYNAMOUS AI COMMUNITY Want to learn agentic coding with live daily events and workshops? Check out Dynamous AI: https://dynamous.ai/?code=646a60 Get 10% off here 👉 https://shorturl.smartcode.diy/dynamous_ai_10_percent_discount ⚡ HOSTINGER — RELIABLE HOSTING FOR YOUR PROJECTS (10% OFF) Whether you're shipping a portfolio, a side project, n8n flows, or AI agents — I use Hostinger for fast, affordable VPS + web hosting. Get 10% off here 👉 https://hostinger.com/DIYSMARTCODE n8n: https://n8n.partnerlinks.io/diysmartcode (Affiliate links — costs you nothing, supports the channel.) ---- Chapters 0:00 Jev 1.14 Image Test: My Answer Key Was Wrong 0:33 Jev 1.14 Release: Images, Box Question, Danger Zone 1:11 Jev Box Question API Request: Four Edges, Confidence, 280 to 1568 px Limits 1:55 Testing Jev on 32 Thumbnails: Ground Truth Read From the HTML 2:39 Jev Box Accuracy: 48 of 48 Cut-Outs, 96 of 96 Badges 3:06 Jev API Speed and Cost: 431 ms, $0.116 per 1,000 Images 4:09 Jev Boxed My 40 px Avatar: The Answer Key Was Wrong 4:39 Errata: 16 Percent vs 92 Percent and a Better Yes/No Question 5:32 Jev Confidence Is Not Position: Python Logo and Resolution Test 6:30 Crop and Ask Again: Fixing Small Targets, 51 Percent to 96 of 96 7:15 Choice Question With Pictures: 48 of 48 Pose Match 7:52 Verdict: Is Jev Vision Ready for Production? What you will see in this 8-minute test: - jev-1.14.0 with the TypeSafe-Danger-Zone-Images: true header, images, the Box question type and jev-preview vs jev-latest - The request format: references, four edges as fractions, confidence, the 280 to 1568 px limit and the 400 error - 32 thumbnails, 96 calls, three runs each, ground truth from getBoundingClientRect - 431 ms median per call, about 2,764 input tokens per image, $0.116 per 1,000 images - Why a Box answer is always a box, and why confidence tracks presence but not position - The docs' crop-and-ask-again fix on 640 px and 500 px images - A Choice question where the options are pictures Not a benchmark: one dataset of my own layouts, one model. Resources: TypeSafe blog, Jev 1.14: https://typesafe.ai/blog/jev-1-14 Box question docs: https://docs.typesafe.ai/go/box Images docs: https://docs.typesafe.ai/go/images Models and pricing: https://docs.typesafe.ai/models Is Jev's vision ready for production, or a dreamer's toy? Ready or toy? #Jev #Jev114 #TypeSafe #AIVision #ImageAnalysis #BoundingBox #ObjectDetection #ComputerVision #VisionModel #LLMVision #APITest #AITools #AIDevelopers #Developers #BuildInPublic #AIAgents #ThumbnailTest #ImageRecognition #AIModels