What's in the picture: how to describe images and distinguish observations from inferences
The 'What's in the picture' tool is designed for one task: to get a text description of visible image details and answers about it. Its main feature is the focus on accurately listing what is actually visible and on separating observations from assumptions. This distinguishes it from universal chatbots, which often guess the context. The tool works via a web form: you enter a text instruction, optionally attach an image, select the response language, and press the Generate button. The result appears in a separate panel, from where it can be copied or downloaded as a plain text file. Below are practical scenarios with request examples and checking criteria. All examples are illustrative and are not results of real tests.
Open tool: Image description ↗How the tool works and what it can do
"What's in the picture" is a web tool for describing images and answering questions about them. You work with a text instruction: describe exactly what you want to find out, and optionally attach an image. The interface is a form with text fields, without style presets, sliders, and change level settings. After pressing the Generate button the result appears in a separate panel, from where it can be copied or downloaded as a plain text file.
The tool does not perform OCR in editing mode, does not generate photorealistic images, and does not create audio or video. It does not support batch processing, automatic saving, integrations, and publication. Vector images are saved only in SVG and PNG formats, speech - WAV via eSpeak, subtitles - SRT. This is important to bear in mind so as not to expect extra from the tool.
Input limits: the first text field - up to 8000 characters, details field - up to 1000 characters, for speech - up to 3000 characters. Image - up to 4 MB, browser audio - up to 20 MB or 5 minutes. Model answers may contain errors, and queue and availability may vary.
Request: «Describe what objects are visible in the image, how they are arranged, what main colors. What is in the foreground, what is in the background?»
Scenario 1: basic description of objects and composition
The first scenario is when you need to get a general idea of the image content. It is suitable for photographs, screenshots, diagrams, and vector illustrations. The request should be specific: list exactly what you want to know. For example, ask to describe main objects, their arrangement, colors, and background. The more precise the question, the more useful the answer.
Request formulation: «Describe what objects are visible in the image, how they are arranged relative to each other, what main colors are used. Separately indicate what is in the foreground, what is in the background.» Such a request sets the answer structure and reduces the risk of getting general words.
Result check: compare the listed objects with what you see yourself. If the tool mentioned an object that is not there, or missed an important one, this is a signal to refine the request. Remember that the model may make mistakes, and vector images are not photographs, so details may be conditional.
Request: «List all objects you see. Indicate their approximate location. Do not guess what is not visible».
Scenario 2: targeted questions about the image
The second scenario is hypothesis checking. For example, you want to know if there is text on the picture, a person in glasses, or a specific color. Instead of a general description, ask a precise question. The tool will answer based on visible details, and you will be able to make a quick decision.
Request formulation: «Is there text in the image? If so, provide it. Is a person in headwear visible? What color is the top clothing?» Break down a complex question into several simple ones - so the answer will be clearer.
Result check: if the answer is affirmative, double-check against the picture. If negative, but you see the object, it may be small or blurry. In this case, ask to describe the area in more detail. Remember: the tool does not perform OCR in editing mode and does not guarantee recognizing all text.
Request: «Is there text in the image? If so, provide it. What color is the background?»
Scenario 3: distinguishing observation and inference
The third scenario is the most valuable for analytics. The tool must clearly separate what it sees from what it infers. This helps avoid interpretation errors. For example, on a picture may be shown a person with closed eyes, but one cannot assert that he is sleeping.
Request formulation: «Describe only what is visible: pose, facial expression, clothing, background. Then separately list possible interpretations, but mark them as assumptions.» Such a request forces the model to follow facts.
Result check: in the answer there should be two parts - observations and inferences. If the model mixes them, ask to rephrase. Remember that even with good separation, inferences may be erroneous, so use them only as hypotheses.
Request: «Describe only visible details: pose, clothing, background. Then infer what may be happening, but mark it as an assumption».
How to formulate requests for exact results
To get a useful description, avoid general phrases like «tell about the picture». Better list specific aspects: objects, their quantity, mutual arrangement, colors, presence of text, background features. The more details you request, the more structured the answer will be.
Use the imperative mood and clear instructions: «List», «Indicate», «Compare». If needed, set a limitation: «Do not guess, write only what is visible». This reduces the risk of hallucinations.
Consider input limits: text up to 8000 characters, image up to 4 MB. If the picture is larger, it needs to be reduced. The tool does not support batch processing and does not save results automatically - copy the answer manually.
Request: «Describe the left part of the image. Is there text? If so, provide it».
Checking and tool limitations
After receiving the answer always verify it against the image. Especially if it concerns small details, text, or colors. The model may make mistakes, and vector images are not photographs, so some elements may be stylized.
The tool does not perform OCR in editing mode, does not generate photorealistic images, and does not create audio or video. It works only with text description and answers to questions. If you need to recognize text, it is better to use specialized programs.
Also remember that queue and availability may vary. Answers do not guarantee accuracy, and results should not be used for making critical decisions without additional verification.
Check: if the tool writes «there is a cat in the picture», but you see a dog, then the description is unreliable. Refine the request or check the image.
Practical recommendations for use
Start with a simple request to list objects, then ask clarifying questions. Thus you will gradually get a complete picture. If the answer is incomplete, ask to describe a specific area or object.
For images with text first ask if there is text, and only then ask to provide it. Do not expect ideal recognition of handwritten or styled fonts.
Use the tool for quick hypothesis checking, but do not rely on it as the only source of truth. Always leave room for human verification.
Request: «Describe the central part of the image. What objects are there?»
The 'What's in the picture' tool is a practical assistant for describing images and answering questions. Three scenarios - basic description, targeted questions, and separating observations and conclusions - cover most everyday tasks. The main thing is to formulate requests specifically and check the results. Remember the limitations: the tool does not replace professional analysis and may make mistakes. Use it as an additional tool, not as the truth in the last instance.
Open tool: Image description ↗Interface translations are machine-assisted and may need refinement.