What's in the picture: how to describe images and separate observation from inference
The «What's in the picture» tool solves a narrow task: from a loaded image get a text description of visible details and answers to questions about it. Its value in learning lies not in the ready-made answer, but in training observability and accurate formulations. The model must separate what is really visible and what is an assumption. In this material — how to use the tool for practice, what data it accepts and how to check the result independently.
Open tool: Image description ↗What exactly the tool does
«What's in the picture» is a visual tool with one input field: you upload an image and, if necessary, add a text instruction. The output is a description of visible objects, colors, element placement and answers to your questions. The output format is plain text that can be copied or downloaded as TXT.
Important limitation: the tool does not generate photorealistic images and does not create audio or video. If you need vector files, only SVG and PNG are available; for speech — WAV via eSpeak, for subtitles — SRT. This is not a document editor: no DOCX or PDF export, no batch processing, auto-saving and integrations.
The tool does not perform OCR as a separate text verification function and has no comparison modes or change highlighting. It gives a text description and answers, and accuracy remains your check. Therefore in learning it is better to use it as a trainer, not as a source of truth.
Example request: «Describe the main objects, colors and element placement in this image. Separately list what is seen exactly and what is an assumption».
Practical input: what to upload and how to formulate
Input is an image up to 4 MB and a text field up to 8000 characters; in one request you can specify up to 1000 details. The more specific the instruction, the more useful the result. Instead of «what's here?» write: «List objects from left to right, indicate their color and mutual position. Do not guess what is not visible».
For educational practice it is convenient to give an image with multiple layers: foreground, background, small details. Ask to describe each layer separately. Then ask a clarifying question: «Which object is closer to the camera and by what signs is this visible?» Thus you train not only perception but also argumentation.
It is useful to limit the volume: «Give no more than five observations and no more than two assumptions». This forces the model to prioritize, and you — to check whether these details are really visible. If the image contains text, ask to list visible inscriptions and separately mark where they are read uncertainly.
Example: «Describe the scene in three paragraphs: general plan, objects, details. At the end add a list of three questions that will help check the description».
Exercise: observation versus inference
The main educational skill is distinguishing observation and inference. Observation: «a white cup stands on the table». Inference: «a person just drank tea». The first is verifiable by pixels, the second is interpretation. Ask the tool to mark each statement as «visible» or «inferred».
Then check the answer yourself. Take the description and for each point ask: can I show this in the image? If not — it is inference, and it needs to be either removed or reformulated as a hypothesis. Such an exercise develops critical thinking and accuracy in reports.
You can complicate the task: ask to describe the image so that a person who cannot see it could draw a scheme. Then compare the scheme with the original. Discrepancies will show which details were missed or imagined. This is an honest check of description quality.
Example: «For each statement indicate the label: observation or inference. If it is an inference, explain which visible signs support it».
How to check the result
Checking starts with re-viewing the image. Compare each description point with what is really visible. Special attention — colors, number of objects and spatial relationships: here errors most often appear. If the tool says an object is on the right, check if it is actually on the left.
Then check logic: do the points contradict each other? If the description says «the background is blurred» and simultaneously «small details on the background are visible», this is a signal of inconsistency. It is useful to ask a clarifying question about the disputed fragment and compare answers.
Remember that model answers may contain errors, and queue and availability may change. Do not use the description as the only evidence. For an educational task it is enough that you can justify which details are confirmed and which remain assumptions.
Example check: «List three statements from the description that cannot be confirmed from the image, and explain why».
Limitations and typical errors
The tool does not guarantee completeness: minor objects, poor lighting and overlaps reduce accuracy. It does not perform OCR as a separate check and does not compare image versions. No batch processing, auto-saving and integrations are available, so each request must be formatted separately.
Do not ask the tool to «improve» the image or create a photorealistic version — that is not its task. Vector formats are limited to SVG and PNG, speech — WAV, subtitles — SRT. Everything else, including DOCX or PDF export, is unavailable.
Another error — trusting inferences as facts. Formulations like «a person is hurrying» or «the weather is cold» are almost always interpretation. In educational practice it is better to mark them as hypotheses and check by visible signs.
Example limitation: «Describe only what is visible. Do not make inferences about mood, weather or causes of events».
20-minute lesson scenario
Take one image and formulate a request for description with separation into observations and inferences. Get the answer and save it as TXT. Then independently write down five visible details that the tool did not mention.
Compare two lists: what matched, what was missed, what was added without basis. Ask a clarifying question about the most disputed point. Evaluate whether the answer changed and became more accurate.
At the end formulate a short conclusion: what types of details the tool describes reliably, and where your check is required. Such a scenario is an example of educational practice, not reproduction of real research. It develops the skill of observation and careful work with text.
Example final question: «Which three image details are unambiguously confirmed, and which require an additional angle?»
«What's in the picture» is useful as an observability trainer: it gives a text description and answers to questions, but does not replace your check. Upload an image up to 4 MB, formulate precise instructions, require separating observation and inference. Check each point by the image, fix assumptions separately and remember the limitations: no OCR check, batch processing and DOCX or PDF export. Such an approach turns the tool into a practical exercise, not a source of ready answers.
Open tool: Image description ↗Interface translations are machine-assisted and may need refinement.