Can Jev read images?
Yes, in preview. Since 14 October 2026, Jev 1.13 and 1.14 accept still PNG and JPEG images when the request sends one header, and Jev 1.14 adds a Box answer that points at one region of the picture. TypeSafe says neither is ready for production, and there is still no audio or video. This page is how to send an image, the limits, and the older routes — OCR, a vision model, a detector — that still carry production work, with the 12 builds in this directory that give Jev eyes.
Updated · by Jon Kraayenbrink

Updated 15 October 2026: Jev 1.14 added image input. Before this release, the answer on this page was no. TypeSafe’s release post.
In short
- Jev reads PNG and JPEG images in preview, on 1.13 and 1.14, with the header TypeSafe-Danger-Zone-Images: true. No audio, no video.
- Images cost what text costs: input tokens at $0.042 per million, with no fee per image. Box, one bounding box per question, is 1.14 only.
- OCR is the smallest useful first step for anything with writing in it, and it reads visible text and nothing else.
- A “free API” is a capped allowance: 1,000 Cloud Vision units a month, 25,000 OCR.space requests, 5,000 Azure transactions. None of them is unlimited throughput.
- A vision-language model writes a richer record, but it can also answer the question directly. That is the baseline Jev has to beat, and sometimes it does not.
- Scores do not stack. An OCR score, a detector score and Jev’s confidence are different numbers, so a confidently misread image can be routed just as confidently.
How to send Jev an image: the header, formats and limits
TypeSafe turned image input on at its JevDay release on 14 October 2026, and says in the launch post that Jev 1.13 “was multimodal all along”. It is off by default. Each request that carries an image sends TypeSafe-Danger-Zone-Images: true to the same endpoint as before, and names jev-1.13.0, jev-1.14.0 or jev-preview. A request without the header gets a 400. Images go in a top-level assets map and are referenced from the state, the instructions or the criteria, and every question type works with them.
| Limit | Value |
|---|---|
| Formats | PNG or JPEG, still images only. Animated PNG is rejected |
| File size | 16 MiB per image |
| Width and height | 256 to 1568 pixels each |
| Total pixels | 78,400 to 2,458,624 |
| Request body | 32 MiB, including the JSON, base64 text and multipart framing |
| Price | Input tokens at the standard rate; no separate charge per image |
| Audio and video | Not supported |
The full reference is TypeSafe’s image documentation. One inconsistency to know about: on 15 October the models table in the same docs still listed the input as text for both versions. The image page is the one to follow.
TypeSafe is plain about the status. It labels images “dreamer-ready” and says they “aren’t yet up to our standards to call ready for production”. So label a set of your own images and measure before anything acts on the answer.
The Jev Box type: one bounding box per question
Jev 1.14 adds a fourth answer type next to Choice, Score and Noul. A Box question returns one bounding box on the image — the top, bottom, left and right edges as fractions of its size — and a confidence score. It needs jev-1.14.0 or jev-preview and the images header, and it is not available on 1.13. The reference is TypeSafe’s Box documentation.
The first builds use it to crop before a second model looks: a doorbell camera crops to the person it found, and a demo moves a crosshair to whatever you name on the screen. One builder’s caution is worth keeping: a valid box does not prove the thing is in the picture, so ask a Noul question first when it might not be.
TypeSafe AI
@typesafeai
we announced Jev 1.14 on 10.14 at 1:14pm PT 🔥 a new and improved model, along with experimental features for our dreamers: • images • a new box type • longer context surprise, Jev was already able to see you! 👁️ BUT our bar for production-grade intelligence is quite high, so builders beware, these are preview features. enter the ~*~ danger zone ~*~ at your own risk. 🚨
XDocuments and OCR
PickJev 1.14: images, a Box type and longer context
- Long context, opt-in
- 262k tokens
- Price, per M input tokens
- $0.042
Todd Dailey
@twid
Tried @typesafeai’s jev 1.14 preview on my doorbell cam’s alert-to-Discord pipeline. My old setup: Gemma 4 26B describes the frame, then DeepSeek V4.1 Flash writes the caption. It caught only 14 of 23 people in recent motion events and missed the mail carrier. Asking jev “is a person in this frame?” across 4 frames of each event got all 23 in a test, with zero false alarms, in about 200 ms. (!) The new jev Box API then crops to the person, so Gemma can describe them far more reliably (21/23, up from 16). Concept to live in less than an hour. This replaces a LOT of AI calls and bounding box tools. The jev call costs about $0.00008 per event. (!) Things are changing so fast. 🚀🚀🚀
XRobotics and devices
PickA doorbell camera on Jev 1.14
- People found, Jev
- 23 of 23
- People found, old pipeline
- 14 of 23
- Answer time
- about 200 ms
- Cost per event
- about $0.00008
Malay
@MalayVasa
text-to-screenshot just say what you want and the crosshair goes and gets it. built with jev 1.14 from @typesafeai
kyle_AI
@kylefu07571
A model release with a literal "danger zone." Jev 1.14 can locate a signature on a scanned form. API: jev-1.14.0 + TypeSafe-Danger-Zone-Images: true Prototype only: a valid box doesn't prove the target exists. typesafe.ai/blog/jev-1-14
XDocuments and OCR
Finding a signature on a scanned form
When to turn an image into text before Jev
Native input does not retire the older routes. Use them when the work is in production and the preview label rules out the native feature, when the file is not a still PNG or JPEG, when it falls outside the size limits, or when the input is audio or video. Customer fine-tuning and LoRA adaptation are not offered either: you can build an adapter around Jev, but you cannot upload a vision adapter into its hosted weights.
In those cases the division is clean: the visual stage reads the image, and Jev decides how the evidence affects the workflow. The routes below follow keno’s walk-through, which measured the limits on each one; the first call itself is on how to use Jev and the price on the pricing page.
If you need audio or video, or a native image route that is out of preview, Cloudflare’s Clef reads text, JSON, images or video, and Clef-omni adds audio, at $0.15 per million input tokens with open weights. It returns a probability for each option and has an API compatible with Jev. How it compares is on Jev vs Cloudflare Clef, and who should pick which is on the Clef alternatives page.
Jev OCR: reading the text in an image
For screenshots, scans, receipts and photographed notices, OCR is often the smallest useful first step. Several of the options are local and free to run.
| Option | What it contributes | Where it fits |
|---|---|---|
| Tesseract | Recognised text, plus TSV and hOCR | A straightforward local baseline |
| EasyOCR | Text, bounding boxes and scores; CPU mode | A multilingual prototype |
| PaddleOCR | OCR and document-processing tools | Documents where layout and structure matter |
| Apple Vision | Text observations from images | Native apps on supported Apple platforms |
| Google ML Kit | On-device text recognition | Android and iOS apps, offline |
Keep the context around the extracted text, because a bare list of words loses it: which attachment it came from, whether recognition was incomplete, and which field or region held it. Without that, a price and a discount read the same, and so do a warning and a button label.
OCR has a clear boundary. It reads visible writing, so it will not explain a text-free product photograph or tell you what is happening in a street scene. Those need the stages further down.
Hosted OCR and vision APIs: what a free tier really means
“Free API” usually means a limited allowance, and each provider counts something different — requests, pages, features or tokens.
| Service | What can become Jev input | Published free access |
|---|---|---|
| Google Cloud Vision | OCR text, image labels, detected objects | First 1,000 units of each feature a month; billing required |
| Gemini Developer API | Descriptions or structured visual observations | Selected models have a free tier; limits vary by model and project |
| OCR.space | Recognised text, optional word positions | 25,000 requests/month, 500/day/IP, 1 MB/file, 3 PDF pages |
| Azure Vision F0 | OCR and supported image analysis | 5,000 transactions/month, 20/minute; not a long-term default |
| Hugging Face Inference | Outputs from supported image models | $0.10 monthly credits; an experiment allowance, not throughput |
Two footnotes worth keeping. Azure’s Image Analysis 4.0 is deprecated and scheduled to retire on 25 September 2028, so it is not a version to build new work on. And the data rules differ: Cloud Vision says it does not use submitted content to train its models, while Gemini’s unpaid-service terms generally allow product-improvement use and human review, with regional exceptions. Check the applicable terms before sending private images.
Gemini is the interesting one for this page, because image understanding includes captioning and visual question answering. It can supply richer evidence than OCR alone — and it might also solve the final task directly, which is exactly why the last section here tests whether Jev adds anything.
Local vision-language models
When the question is about objects, relationships or layout, a vision-language model can produce an evidence record from the image. One downloadable option is Qwen3-VL, including a 2B Instruct checkpoint. Hardware needs depend on the checkpoint, the quantization, the image resolution, the context and the runtime, so a model-size label is not a memory guarantee.
Ask the visual stage for specific observations — readable text, visible damage, occlusion, uncertain details — rather than a general caption, because a caption can omit the exact detail a later decision needs. A richer record costs more to produce and transmit, so build its fields around the downstream question and keep uncertainty in rather than replacing missing evidence with a confident guess.
One caveat that catches people out: local processing does not make the whole workflow local. Anything you send onward to Jev leaves that local stage.
Detectors and embeddings: objects, positions and similar images
A detector returns a bounded vocabulary of objects and their locations. YOLOX is one open-source example with small model variants. For an inventory workflow it might identify boxes and equipment, if those classes are supported or it was trained for them, and code can count the detections and compute the spatial relationships. Jev then evaluates those observations against a written handling policy — which avoids asking it to do geometry from a list of coordinates. If the final task is simply “did the detector find a box?”, a rule may be enough and Jev is not needed.
An image encoder is the other local route. Google’s SigLIP 2 supports image-text retrieval, zero-shot classification and image embeddings, and the linked checkpoint is Apache-licensed; CLIP is the established reference for the family. Two useful patterns: compare a photo with candidate descriptions and pass the names and scores onward, or retrieve similar human-labelled examples and give Jev their labels and context. In both, do the vector comparison outside Jev and pass it the labels and scores, not the raw vector.
And read the similarity straight: a cosine similarity of 0.83 is not an 83% chance of being correct.
Colour, histograms and the limit of a few points
A few colour samples might flag a dark photo or sort packaging by colour. They will not reliably tell you which object is in the picture. A local image-processing stage can compute brightness, colour proportions, edge density or a coarse spatial grid — OpenCV’s histogram tools describe distributions of intensity or colour — and those are useful for detecting nearly blank scans, sorting standard packaging and flagging dark photos.
The boundaries are structural. A global histogram throws away spatial arrangement, different objects can share similar colours, and sampling a few points can miss small objects entirely. Keeping regions preserves some layout but still supplies no object semantics.
TypeSafe documents one more limitation here, and it is specific: Jev performs better with semantic colour names than with hex values, and cannot reliably judge closeness between RGB or hex values. So do the conversion in code and give Jev meaningful fields or buckets — an experimental input might read “image quality: dark, dominant colour: red, text detection: incomplete” rather than a hex code.
Check what your app already knows before you look at the picture
Sometimes the image is only a presentation of data your application already has. A document may carry a text layer, a webpage may expose useful DOM or accessibility text, a product asset may have verified catalogue metadata, and a barcode may encode an identifier that resolves to a structured record. Passing that data straight to Jev avoids unnecessary visual inference.
Filenames and unverified alt text are especially weak evidence, because they may be stale or simply wrong. For repeated, unchanged assets, cache an extraction keyed to the image and the extractor version, and invalidate it when the image changes. If you substitute a matching reference image’s label, say that it is retrieval from known material rather than recognition of new content.
The cascade: OCR, detector, vision model, human
The methods above can form a cascade — existing text, then OCR, then a detector or encoder, then a vision-language model, then human review. That is a design pattern, not a required sequence: a scene with no writing should not have to pass through OCR first. For a support screenshot, start with the text and escalate when it is unreadable or insufficient; for a product photo, start with the visual attributes.
A conceptual evidence record, the kind of state you would hand to Jev, might look like this. It is an example, not a measured model response or an SDK schema.
{
"attachment_type": "support_screenshot",
"extraction_method": "local_ocr",
"observed_text": ["Payment failed", "Update payment method"],
"customer_message": "I cannot renew my subscription",
"limitations": ["small footer text unreadable"],
"available_routes": ["billing", "account_access", "other", "manual_review"]
}Ask one narrow question per decision, keep observations separate from instructions, and retain an explicit fallback for missing evidence. Jev’s state documentation supports objects with descriptive fields, so the record above can carry the uncertainty through to the decision instead of hiding it.
Why a confident score does not repair missing evidence
There are several different numbers in a pipeline like this: an OCR recognition score, a detector score, an embedding similarity, Jev’s answer probability and Jev’s choice confidence. They are not interchangeable. TypeSafe describes confidence as a statistic derived from the answer distribution, not an independent visual verification.
So if the first stage confidently misreads an error message, the next stage can confidently route the wrong issue, and multiplying two scores together does not produce an end-to-end correctness probability. The fix is the same one every other page here gives for confidence: label a set of cases yourself, see where the confidence falls when the pipeline is right against when it is wrong, and set the threshold there. For a binary event, cases assigned a probability near 0.8 should contain that event about 80% of the time. That is what a calibration curve checks, and it is the only way to know whether a threshold means anything.
Give Jev eyes with Obsidian, Claude and MCP
The second half of this idea is not about pixels at all: it is about retrieval. Kev Builds Apps shows the setup in a full walkthrough. Install the Model Context Protocol server plugin in Obsidian, exit restricted mode, enable the plugin and copy the API key it generates. Then, from Claude Code, tell the agent to connect to your vault, name the graph, and pass the key. Jev reaches the vault through the MCP server.
With the connection up, the vault becomes the visual memory. Put media and data in it — images, logos, videos, PDFs, CSV — and have the agent store the media metadata and structured information in the graph. After that, ask Jev to retrieve a specific asset, like a logo file or a clip, or to answer a correlation question across the stored metadata, and it searches the graph and returns the referenced file. The directory build for this is MikeLembo’s jev-graph-search, which adds a Jev-powered search to Obsidian.
Albiona Hoti
@albicodes
I built a visual reference finder with Jev One single prompt → 100 images from Cosmos, NASA, and The Met my new rabbit hole for creative work 🌻

GitHubSearch
Jev-powered Obsidian search
Search an Obsidian vault with Jev deciding relevance.
MikeLembo
12What the image stage costs to run
The price people quote is the price of the decision stage, not the image workflow. Jev is $0.042 per million input tokens with no output-token charge, and an image sent natively is billed the same way, as input tokens, with no fee per image. TypeSafe says guidance on estimating image tokens is coming. Everything in front of Jev is a separate bill. For a sequential pipeline:
- Total latency = preparation + visual extraction + Jev + validation + network and queue time.
- Total cost = visual extraction + Jev + infrastructure + retries and review.
Parallel processing can raise throughput without lowering every item’s latency, and local inference spends hardware and time even when there is no per-call invoice. Report a cached run and a first-time extraction separately, because they cost different amounts and only one of them is honest about steady state.
When a vision model alone is enough
The result that argues against this whole page: an image-capable model like Gemini can answer the bounded question itself, so there is no guarantee that OCR plus Jev beats it on accuracy — only on cost and latency at volume. The proposal in the source article is a first experiment on consented or synthetic support screenshots with known routing labels, comparing four setups.
| Baseline | What it tests |
|---|---|
| OCR + rules | Whether semantic classification is needed at all |
| OCR + Jev | Whether Jev improves routing from extracted text |
| VLM alone | Whether one image-capable model is sufficient |
| VLM + Jev | Whether a separate decision stage improves the result |
Include blurry images, unfamiliar layouts, missing text, irrelevant attachments and cases with no valid automatic route, and split by document or template family so near-duplicates do not leak between development and evaluation. Report accuracy, false automatic actions, the manual-review rate, p50 and p95 latency, and cost per correctly completed task. Freeze the prompts and thresholds before the final test.
The builds that already do this
Each of these turns an image, a screen or a vault into the text Jev decides over, and each one publishes what it measured. The builds that send the image to Jev directly are in the Box section above.
keno
@kenonews
JEV can’t see images. So how do you give it eyes? What’s possible? What can run locally? How far can you get for free? The answers (and the catches) are in the full breakdown below:

XDocuments and OCR
How to Give JEV Eyes
Fayaz Ahmed
@fayazara
Made myself a little image classifier with OCR + Jev It was able to categorise ~900 images in 40 seconds Pretty cool
Trinay Hari
@hari_trinay
Built a construction plan-set classifier with Jev. Proq turns civil and building plan sets into bills of materials using an LLM pipeline we built on GPT-4.1. Jev classified an entire 26-sheet plan set in 2.9 seconds for $0.0052. It matched GPT-4.1 and GPT-6 Astra on 100% of sheet-level classifications while running 17–21x cheaper and 5x faster than our production pipeline.
Milind S
@milindlabs
Okay so Jev can actually do computer use really well Without any screenshots, or LLMs and no Pixels leave my mac I dont even read the Dom elements A local CoreML model segments every button and UI element on screen. On-device OCR reads the labels. That text is all Jev gets. It returns a probability across those elements and tells me the best one to click. Then it clicks, re-runs detection, and decides again. In a loop until the goal is done. ~90ms per decision. Faster than any LLM computer use I've tried. Blazing fast computer use, without any latency @typesafeai is building something really interesting
XAgents and browsers
Computer use without screenshots
Richard
@richard_meng_01
Nitpicky, an AI generated photo detector powered by jev AI generated photos can be told from nits. That's why we build something to zoom into every detail: faces, fingers, characters, numbers, poses, where common senses fall apart, judged by jev
XDocuments and OCR
Nitpicky, zooming into AI images

Live siteOpen source
Laya Vision
Typed decisions about an image plus text, in one forward pass.
r33drichards
201MWhere these numbers come from
Every figure on this page is its author’s or the provider’s own: TypeSafe’s launch post and image docs for the native feature and its limits, keno’s walk-through for the routes and their limits, Kev Builds Apps for the Obsidian setup, and the quickstart, the reranking guide and the tracing setup for the decision stage they feed.
Guideyoutube.com/@RoboNuggets
Pick“Jev for Images” Just Changed Visual Automations Forever
Jay E of RoboNuggets on Jev’s new image input: what changed, setting it up in one prompt, the limits and the cost, and eight visual use cases, in about twelve minutes.
Guidetypesafe.ai
Jev 1.14.0 + experimental features for dreamers
TypeSafe’s JevDay post: jev-1.14.0 as jev-preview, image input on 1.13 and 1.14, the Box type and opt-in long context, all behind headers and none of them production-ready yet. jev-latest moves to 1.14 a week after 14 October.
Guidedocs.typesafe.ai
Images
TypeSafe’s reference for image input: the TypeSafe-Danger-Zone-Images header, PNG and JPEG only, 16 MiB per file, 256 to 1568 pixels per side, 32 MiB per request, and images billed as input tokens with no separate charge.
Diogo Almeida
@CompleteSkeptic
Types Beyond Decisions: box We’re also releasing a new type that is somewhere in the gray area of decisions: a single bounding box. We warned you there’s a reason we call them System One Models! (only on jev-preview)
Guidex.com
Types beyond decisions: box
Diogo Almeida of TypeSafe on the new Box type: one bounding box per question, somewhere in the grey area of decisions, and only on jev-preview.
Guideyoutube.com/@KevBuildsApps
Jev + Claude + Obsidian Just Changed AI Forever (Full Setup)
Kev Builds Apps wires Jev into an Obsidian vault over MCP: store the media and its metadata in the graph, then ask Jev to retrieve the logo, video or file a task needs instead of guessing.

Guidedocs.typesafe.ai
Quickstart: state, questions, typed answers
The three question types, Choice, Score and Noul, in one support-ticket example.
Guideyoutube.com/@engineerprompt
Steerable Reranking: How JEV Solves RAG
Prompt Engineering puts Jev in the reranking step of a RAG pipeline: why vector search and cosine similarity fail, how a steerable reranker works, and how it compares with LLM rerankers and cross-encoders. A Colab notebook comes with it.
Akshay 🚀
@akshay_pachaar

Guidex.com
Build a Jev Judge
Akshay Pachaar’s worked example: evaluate a refund-support agent with Jev instead of a generative judge, and record the verdicts in Opik. Separates the judge from the evaluation system, and is explicit that a 0.98 is a probability about one proposition, not a percentage of the answer.
Guidearize.com
Trace every judgment with Phoenix
Arize’s instrumentation for Jev: one line of code to trace each decision, with the integration docs.
Common questions
- Can Jev read images?
- Yes, in preview since 14 October 2026. Jev 1.13 and Jev 1.14 both accept still PNG and JPEG images when the request sends the header TypeSafe-Danger-Zone-Images: true. Images are off by default, and TypeSafe says the feature is not ready for production yet.
- Does Jev support images, audio or video?
- Images yes, in preview. Audio and video no. Images must be still PNG or JPEG: animated PNG is rejected, and GIF, WebP and HEIC have to be converted first. For audio or video, turn it into text or frames before the call, or use a decision model that takes them, such as Cloudflare's Clef-omni.
- How do I turn on image input in Jev?
- Send the header TypeSafe-Danger-Zone-Images: true with each request that has an image or a Box question, to the same endpoint as before. Name the model jev-1.13.0, jev-1.14.0 or jev-preview. The latest Python and JavaScript SDKs build the image references for you.
- Is Jev image input production-ready?
- No, by TypeSafe's own account. It calls images, Box and long context preview features for “dreamers” and says there is still too much jaggedness to vouch for them in production. Test on your own labelled images before anything acts on the answer.
- Do images cost more on Jev?
- No. Images are billed as input tokens at the normal rate, $0.042 per million, and there is no separate charge per image. Image tokens count toward the context limit. Output is still free.
- What is the Jev Box type?
- A new answer type in Jev 1.14 that returns one bounding box on an image — the top, bottom, left and right edges as fractions of the image — plus a confidence score. It needs jev-1.14.0 or jev-preview and the images header. A valid box does not prove the thing is in the picture.
- Should I still use OCR before Jev?
- Often, yes. OCR, a vision-language model or a detector in front of Jev is still the route for production work, for formats other than PNG and JPEG, for images outside the size limits, and for audio and video. Whatever that stage produces as text is what Jev receives.
- What is the cheapest way to read text from an image?
- Local OCR, because there is no per-call bill: Tesseract, EasyOCR and PaddleOCR all publish under Apache 2.0 and run on your own machine. A hosted free tier is capped — 1,000 Cloud Vision units a month, 25,000 OCR.space requests — and none of them is unlimited.
- Can a vision model replace Jev?
- Sometimes. If one image-capable model can answer your bounded question directly, Jev may add nothing, and the honest way to find out is to test that baseline against OCR plus Jev before you build anything. Jev earns its place when the visual stage has to feed many narrow decisions, or when the decision runs at a volume that makes the larger model too slow or too dear.
- Why is a confident answer still wrong?
- Because the numbers are not interchangeable. An OCR score, a detector score, an embedding similarity and Jev's confidence measure different things, and confidence is a statistic derived from the answer distribution, not an independent check of the image. If the first stage misreads the picture, Jev can route the mistake with equal certainty.
Made with Jev is independent and not affiliated with TypeSafe AI. Every figure on this page is the one its author published, linked to where it can be checked.