What's in the picture: how to describe images and answer questions
The "What's in the picture" tool helps you get a textual description of an image: list objects, their colors and location, and answer questions about the picture. It does not create photorealistic images or generate audio or video. Work is done in a web form with text fields: the main field up to 8000 characters, details field up to 1000 characters, image up to 4 MB. You choose the output language, press Generate and get one result in a separate panel, which you can copy or download as plain text TXT. In this article we will look at how to prepare a request, perform the task and check the answer, in order to distinguish observation from inference.
Open tool: Image description ↗What exactly does the tool do
"What's in the picture" analyzes the uploaded image and outputs a textual description of visible details. Typical questions: what color is the car, how many people are in the photo, what is to the left of the table. The tool responds in text, not picture: it does not create photorealistic images and does not generate audio or video. Vector formats SVG and PNG, as well as speech in WAV via eSpeak and subtitles SRT, belong to other service capabilities and are not the result of image description.
Input limitations are important in practice: the main field accepts up to 8000 characters, the details field - up to 1000, the image - up to 4 MB. Model answers may contain errors, and queue and availability may change, so the result should always be checked against the actual image. Export is available only as plain text TXT.
The tool does not perform text recognition on the image with guaranteed accuracy: it may notice that text exists, but is not required to read it without errors. It should not be used for medical, financial or legal conclusions - it is an auxiliary means for describing the visible.
Example request: "List main objects, their colors and mutual location. Answer in Russian."
How to prepare image and request
Before loading check the file size: up to 4 MB. If the image is larger, compress it or crop unnecessary fields. The clearer the picture, the more accurate the description: heavily blurred or detail-overloaded shots give blurry answers. For vector images SVG or PNG are suitable, but remember the tool describes what is visible, not creates photographs.
Formulate the request specifically. Instead of "What's here?" write: "List main objects, their colors and location. Indicate what is in the foreground, what is in the background." If you are interested in a detail, ask directly: "What color is the left person's jacket?" Several questions in one request are acceptable if they do not contradict each other. Avoid formulations like "Describe everything" - they lead to generic phrases.
The details field is for context: "This is an interior photo, I am interested in furniture and lighting." Do not overload it: 1000 characters are enough for several sentences. The output language is selected in the interface, so you do not need to write "answer in Russian" in the request itself.
Example: "Photo of a room. Describe what furniture items are visible, what colors they are and where they are located. Separately indicate if there are paintings or posters on the walls."
Step-by-step order of operation
Step 1. Open the web form. You will see the main text field (up to 8000 characters) and the details field (up to 1000 characters). Step 2. Upload an image up to 4 MB. Step 3. Enter a request, for example: "Describe main objects, colors and location. Answer questions: how many people in the photo? What is in the center?" Step 4. Select output language. Step 5. Press "Generate". Step 6. Copy the result from a separate panel or download it as TXT.
During generation a queue is possible, and service availability may change. If the result does not appear, try later, without pressing the button repeatedly if not necessary. After receiving the answer verify it against the image: the tool may confuse colors, miss small details or make incorrect inferences.
The main habit when working with description - separate observation and inference. Observation: "A cup is on the table". Inference: "The person is drinking tea". The model may mix them, so mark which statements are directly visible in the picture, which are interpretations. History is not saved automatically, so copy the result immediately.
Sequence: upload photo, enter "List objects, their colors and location. Answer in Russian", press "Generate", copy text.
How to check the result
Check the description against five criteria. Completeness: are main objects listed? Accuracy: do colors and location match what you see? Distinguishing observation and inference: are there statements that cannot be seen directly, e.g. "The person is sad". Logic: are there contradictions? Language: does it match the selected output language. If the answer is unsatisfactory, refine the request and generate again.
It is convenient to verify the description by zones: foreground, background, left and right sides. Check if objects are mentioned in each zone. If something is missed, ask a clarifying question: "What is in the lower left corner?" If there is a color error, ask: "What is the color of the object in the center?" Each request is a separate generation, so you can ask again, but do not expect the model to remember the previous answer.
Consider limitations: small text may not be recognized, shades may be transmitted inaccurately, complex scenes may be described simplistically. Model answers may contain errors, so critically evaluate the information and do not make important decisions based solely on the description.
Checklist: 1) All objects listed? 2) Colors correct? 3) Location indicated? 4) Are there inferences without observation? 5) Language correct?
Example: describing an interior
Suppose you have a photo of a room and need a description for a presentation. Upload an image up to 4 MB. In the main field write: "Describe main objects, colors and location. Indicate what is in the foreground, what is in the background. Answer questions: how many windows in the room? What color are the walls?" Add to the details field: "This is a living room interior, I am interested in furniture and decor." Select Russian output language and press "Generate".
Fragment of possible answer: "In the image is visible a room. In the center stands a gray-colored sofa. To the left of the sofa - a lamp with a white shade. In the background - a window with white curtains. Walls are light-beige. On the wall on the right hangs a picture in a dark frame. On the floor - a carpet with a geometric pattern. Only one window." Here objects, colors and location are listed, but there is no inference about the interior style, unless you ask separately.
Check the answer against the image. If there is a plant in the photo but the tool did not mention it, ask a clarifying question. If the sofa color is named incorrectly, ask to recheck. Do not ask to generate a similar picture: the tool describes what is visible, not creates images. This example is fictional and illustrated only.
Request: "Describe main objects, colors and location. Indicate what is in the foreground, what is in the background. How many windows? What color are the walls?"
Typical errors
Too general request like "What's in the picture?" gives a blurry answer - specify exactly what you are interested in. A request with ten questions right away may confuse the model - break it into parts or set priorities. Expecting photorealistic generation is meaningless: the tool only describes the image. Ignoring limitations leads to errors when loading: a file larger than 4 MB will not pass, so compress it.
Not distinguishing observation and inference is a common problem. If the answer says "The person is angry", check if this is visible in the image, or it is an interpretation. Do not use the tool for medical, financial or legal conclusions: the article is educational and does not give personalized advice. Do not expect batch processing - images are processed one by one, and export is available only as TXT.
To avoid errors, check the result against the checklist, ask clarifying questions and rephrase the request if necessary. The tool can make mistakes, so use it as an auxiliary, not the only source of information.
Instead of "Describe everything" write "List objects, their colors and location. Separately indicate if there is text on the image."
Limitations and responsible use
The tool does not replace professional image analysis. If the picture contains confidential data, do not upload it unnecessarily and observe the privacy policy. Do not use descriptions to create false information or mislead and do not attribute to the tool capabilities it does not have: it does not generate photorealistic images, does not create audio and video, does not perform text recognition with guaranteed accuracy.
In medicine, finance and law descriptions may be inaccurate or incomplete. The tool is not intended for diagnosis, investment decisions or legal consultations - consult a specialist. Model answers may contain errors, queue and availability may change.
All scenarios in examples are fictional and illustrated for illustration. Use the tool responsibly: check the result, distinguish observation and inference and do not rely on it as the only source of information.
Do not upload images with personal data without consent. Do not use descriptions as the sole basis for important decisions.
"What's in the picture" - a simple assistant for describing visible details of an image and answering questions about it. It works in a web form, accepts images up to 4 MB and text up to 8000 characters, allows you to choose the output language and get one result in a separate panel, which can be copied or downloaded as TXT. Remember the limitations: the tool does not create photorealistic images, does not generate audio or video, does not perform batch processing. Always check the answer against the image and distinguish observation from inference. Following the steps in this article will help you get more accurate and useful descriptions.
Open tool: Image description ↗Interface translations are machine-assisted and may need refinement.