
Introducing Jumper’s New Model Picker: Your Visual Search, Your Way
Big news, Jumper fans! We’re thrilled to launch a new version of our app that puts you in the driver’s seat. Now, you can choose the exact machine learning model that powers your visual search—tailoring performance, accuracy, and even language support to your needs.
Why This Matters
Our visual search models are built on cutting-edge research from Google DeepMind and OpenAI, blending innovations like sigmoid loss pre-training and compute-optimal scaling laws (shoutout to the brilliant minds behind these papers). Whether you’re analyzing footage for reverse image searches, text detection, or non-English queries, you’ve got options:
Model | Resolution | Parameters | Size on Disk | Notes |
---|---|---|---|---|
Medium | 256×256 | ~203M | ~812MB | Base model bundled with the application. |
Medium x-high-res | 512×512 | ~203M | ~812MB | Higher resolution for improved detail (e.g., OCR, reverse search). |
Large multilingual | 256×256 | ~370M | ~1.48GB | Boosts accuracy for non-English searches. |
Large | 256×256 | ~652M | ~2.61GB | Larger version of Medium for sharper results. |
Large high-res | 384×384 | ~652M | ~2.61GB | Higher resolution version of Large. |
XLarge | 384×384 | ~877M | ~3.51GB | Even larger model with top-tier precision. |
XLarge multilingual | 256×256 | ~1.13B | ~4.51GB | Highest accuracy for non-English searches. |
Recommendation: For most use cases, go with at least the Large variant. On a MacBook Pro (M1 Pro, 16GB RAM), anything up to Large runs smoothly. XLarge might push your hardware, but the results? Worth it. High-res models won’t slow you down—they just need more RAM and storage for those beefier analysis files.
Transcription Gets Smarter Too
For audio, we’re rocking OpenAI’s Whisper tech (large-v3-turbo, to be exact). Windows and Intel Mac users get the full package, while M-series Mac folks enjoy a lightning-fast, quantized version optimized for Apple’s MLX framework—same power, smaller footprint.
Platform | Model Variant | Parameters | Size on Disk |
---|---|---|---|
Windows / Intel Mac | whisper-large-v3-turbo | ~809M | ~1.62GB |
Apple M-series Macs | whisper-large-v3-turbo | ~809M | ~467MB |
Notes:
- Windows / Intel Mac: Bundled with Windows and Intel Mac installers
- Apple M-series Macs: Uses a quantized version converted to Apple’s MLX framework for hardware acceleration (faster analysis, smaller model size). Bundled with Apple M-series installer.
How It Works
Head over to the Settings tab and under “Visual Model Selection” you can now choose your model. Pick your model, re-analyze your footage, and go. Already processed a file at the same resolution? You’re halfway done—Jumper skips the frame extraction. Models live in ~/.jumper/models
(e.g., /Users/max/.jumper/models
on macOS), so they’re ready when you are.
Try It Now
Download the update, play with the options, and let us know what you think! Whether you’re a casual user or a power analyst, this is Jumper, customized for you.
Happy searching,
The Jumper Team