# A Jev-like vision wrapper turns image questions into typed choices > An independent Python example uses vision-model token probabilities to return typed decisions from images. Its webcam rates are author-reported, and accuracy remains untested. By BIG CHANGE Editorial Published: 2026-09-26T10:09:47.074Z Updated: 2026-09-26T10:09:47.074Z Canonical: https://bigchange.ai/blog/jev-like-llm-vision-wrapper-logprobs ![Conceptual charcoal illustration of a webcam clipped to a monitor and facing a leafy plant on a shelf.](https://bigchange.ai/api/media/file/jev-vision-webcam-plant-hero-v1.png) AI-generated conceptual illustration by BIG CHANGE. A [Python example published September 25 by developer Allan Riordan Boll](https://allanrbo.blogspot.com/2026/09/a-jev-like-wrapper-for-llms-including.html) extends a Jev-style decision request with image attachments. It sends each webcam frame to a vision model alongside short questions, then converts the model's next-token probabilities into a yes/no value, a choice or a score. The example is an independent wrapper, not a Jev vision feature or a test of Jev's model. The useful part for developers is the boundary of the technique. A vision model can answer a constrained question without writing a description, but the returned option probabilities depend on the prompt, available token alternatives and the model. The post provides a working pattern to inspect, not an accuracy study. ## The big change - **What changed:** A developer has applied the small-decision format associated with Jev to images using general vision models. The image is attached to each request, while a one-letter answer turns a visual question into a value software can handle. - **Why it matters:** Developers can change the visual criterion in text and receive a typed result without building a separate image classifier for each question. This example covers visible people, plants, scene setting and brightness; it does not establish how reliably those judgments transfer to other cameras or scenes. - **What to watch:** The practical decision is whether a chosen model and endpoint return the required alternative-token scores consistently enough for the task. Frame throughput and decision quality need to be measured together on representative images before a webcam result drives an action. ## How the image decision works [TypeSafe AI's Jev quick start](https://docs.typesafe.ai/introduction/quickstart) documents a `state` and a set of typed `questions`: `noul` for a yes/no value, `choice` for named alternatives and `score` for ordered levels. Boll's script uses those names and adds an `attachments` array of image paths or base64 data URLs. That field is his extension to the request object; the cited Jev quick start describes text state and does not document it as a Jev API input. For each question, the script builds a prompt with lettered options such as `[A] true` and `[B] false`. It asks the model to answer with the best letter and reads the first output token's `top_logprobs`. It exponentiates the returned log probabilities, normalizes the weights across the listed letters and maps them back to the question's type. A `choice` returns the highest-weight option and its distribution. A `noul` returns the weight for `true`. A `score` returns a weighted average of the ordered levels. The script rejects a response when omitted option tokens could still carry material weight. The image is supplied with each question. The example sends separate requests rather than obtaining all answers in one model call. Its OpenAI path uses the Responses API with `input_image`, `top_logprobs` and `message.output_text.logprobs`; its local llama.cpp path uses Chat Completions with an `image_url` content item and log probabilities. [OpenAI's image guide](https://developers.openai.com/api/docs/guides/images-vision) documents base64 image data URLs, and its [Responses reference](https://developers.openai.com/api/reference/resources/responses/methods/create) documents the log-probability output and a maximum of 20 returned alternatives per token position. [llama.cpp's server documentation](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md) documents image URLs in its chat interface. These sources support the request pattern; we have not executed the example against either endpoint. ## What the webcam example measures The script captures a frame with OpenCV, encodes it as JPEG and asks four questions: whether a person or plant is visible, whether the setting is indoors or outdoors, and how bright it is. A background worker evaluates one frame at a time while the preview continues. Its camera setup uses Linux V4L2, so the posted file is not a portable webcam setup without changes. The article text says three questions per frame, but the published code contains four; the code is the basis for that count here. Boll reports about **one evaluated frame per second** with a locally served Gemma 4 12B QAT model on an RTX 3090, and about **0.2 frames per second** using hosted GPT-6 Luna. He suggests repeated connections may contribute to the hosted result. The post does not provide a controlled comparison of hardware, network, image size, caching, accuracy or request timing. Those numbers describe this author's setup and code, not a general speed ranking for the models. [OpenAI lists GPT-6 Luna as accepting image input](https://developers.openai.com/api/docs/models/gpt-6-luna), and its [model guidance](https://developers.openai.com/api/docs/guides/latest-model) says Luna supports the `none` reasoning setting used by the example. For a developer adapting the script, the first checks are concrete: confirm the model accepts image input and exposes the required first-token alternatives; check whether all option letters appear; then evaluate the returned decisions on labeled images from the intended camera or dataset. The normalized weights are relative to the listed letter tokens. They are not, by themselves, measured probabilities that a visual judgment is correct. OpenAI's [older logprobs cookbook](https://developers.openai.com/cookbook/examples/using_logprobs) explains the token-probability idea but is marked archived and may contain outdated API examples. This wrapper also clarifies what a typed result leaves to application code. The model judges the supplied image against the written criterion. The application chooses frames, handles missing scores and decides whether any result is safe enough to act on. Boll's example prints a table; it does not report an automated action or a measured deployment. ## Sources & further reading - [Allan Riordan Boll, "A Jev-like wrapper for LLMs, including vision models," September 25, 2026](https://allanrbo.blogspot.com/2026/09/a-jev-like-wrapper-for-llms-including.html): original Python example, webcam workflow and author-reported frame rates. The prose says three questions per frame; the code defines four. Its timings are not an independent or controlled benchmark. - [TypeSafe AI, Jev quick start](https://docs.typesafe.ai/introduction/quickstart): documents text state and the `noul`, `choice` and `score` question types. It does not document the author's custom `attachments` field as a Jev API feature. - [OpenAI, Images and vision](https://developers.openai.com/api/docs/guides/images-vision): documents `input_image`, image URLs and base64 data URLs for vision input. [The Responses API reference](https://developers.openai.com/api/reference/resources/responses/methods/create) describes `message.output_text.logprobs` and the limit on returned alternatives. - [OpenAI, GPT-6 Luna model and model guidance](https://developers.openai.com/api/docs/models/gpt-6-luna): confirms image input and the model's `none` reasoning setting; the [model guidance](https://developers.openai.com/api/docs/guides/latest-model) details parameter compatibility. These documents do not verify the blog author's throughput. - [llama.cpp server documentation](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md): describes its OpenAI-compatible chat endpoint and image URL input. Backend support and returned token lists still need checking with the exact version and model in use. - [OpenAI cookbook, Using logprobs](https://developers.openai.com/cookbook/examples/using_logprobs): background explanation of token probabilities. OpenAI labels this recipe archived and warns that some models or APIs may be outdated. ## Sources - [Allan Riordan Boll: A Jev-like wrapper for LLMs, including vision models](https://allanrbo.blogspot.com/2026/09/a-jev-like-wrapper-for-llms-including.html) — Original independent Python code and author-reported webcam observations. Code defines four questions although surrounding prose says three; no controlled benchmark or independent accuracy study. - [TypeSafe AI: Jev quick start](https://docs.typesafe.ai/introduction/quickstart) — Documents text state, question types and example request body. The blog author adds attachments to his own Jev-like request object; this quick start does not document that field. - [OpenAI: Images and vision](https://developers.openai.com/api/docs/guides/images-vision) — Documents image input and base64 data URLs. Supports the request pattern, not the author's observed throughput. - [OpenAI: Create a model response](https://developers.openai.com/api/reference/resources/responses/methods/create) — Documents image inputs, message.output_text.logprobs and up to 20 top log probabilities per token position, sometimes fewer. - [OpenAI: GPT-6 Luna and model guidance](https://developers.openai.com/api/docs/models/gpt-6-luna) — Lists image input and none reasoning effort; OpenAI's model guidance explains logprob parameter limits by effort. - [llama.cpp server README](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md) — Documents OpenAI-compatible chat endpoint and image_url input. Exact backend/model version was not independently exercised here. - [OpenAI Cookbook: Using logprobs](https://developers.openai.com/cookbook/examples/using_logprobs) — Background on token probabilities. OpenAI marks this cookbook example archived and possibly outdated for current models or APIs.