Service «What's in the picture»: how to describe images and check outputs
The service «What's in the picture» helps describe visible details of an image and answer questions about it. The interface has text fields: main (up to 8000 characters) and for details (up to 1000 characters). You choose the output language, press «Generate», and get one result in a separate panel. You can copy it or download it as a plain text TXT file. Vectors are available only in SVG and PNG, speech via eSpeak in WAV, subtitles in SRT. No DOCX or PDF export, batch processing, auto-saving, integrations, settings panels, highlighting changes, special comparison modes, direct publishing, email delivery, document upload, premium plan, registration, speech voice cloning, photorealistic image generation, or audio/video generation.
Ouvrir l'outil: Description d'image ↗What exactly does the service «What's in the picture» do
The service «What's in the picture» solves one task: based on an image and text instruction, return text — description of visible details or answer to a question. You fill the main field (up to 8000 characters) and optionally the details field (up to 1000 characters), choose the output language and press «Generate». The result appears in a separate panel, you can copy it or download it as TXT. This is the entire cycle: there are no additional modes, comparisons, or change highlighting in the interface.
The key requirement for working with the service is distinguishing observation from inference. Observation records what is visible: objects, colors, location, presence of text. Inference is interpretation: a person's mood, purpose of an object, cause of an event. Formulate the request so that the service clearly separates these layers. For example: «List visible objects and their colors. Separately indicate which statements are assumptions». You will immediately see where fact ends and where guess begins.
Formats are limited: vectors only SVG and PNG, speech WAV via eSpeak, subtitles SRT. Image accepted up to 4 MB, audio up to 20 MB or 5 minutes, text up to 8000 characters in the first field and up to 1000 in the details field. The service does not recognize text as a document, does not save history, and does not publish results. Therefore, it is reasonable to use it where a quick textual draft about a picture is needed, not ready documentation.
Example request: «Describe visible objects, their colors and their arrangement. Separately list what is an observation and what is an assumption».
Which parts of the task can be delegated to the service
Primary listing of details — the safest part of the work for delegation. The service can name objects, their colors and order of arrangement: «laptop, cup, notebook, pen, lamp; laptop silver, white cup, notebook with green cover». This is routine work that is easy to verify, by comparing the text with the image. If details are small or image quality is low, the probability of error is higher, so verification is mandatory.
Answers to clearly posed questions — the second suitable task. Formulate them specifically: «How many people are in the image?», «What color is the car on the left?», «Is there text on the picture?». The service will return a text answer, which you verify against the image. Do not ask for legal, medical or financial assessment based on the image: this goes beyond describing visible details and increases the risk of error.
Third task — help formulating an inference, when you already understand what you see but cannot find the words. For example: «Describe what the person is doing and guess what their goal is». The service will propose a draft text, which you will edit to your context. This speeds up work, but does not replace your judgment: the final formulation remains with you.
Example: «On the image three people at a table. Two are looking at a laptop, one is holding a pen. Colors: blue chairs, white table, gray walls. Text on the board is illegible».
What must be checked manually
Check the accuracy of observations. The service may make a mistake in color, number of objects or their arrangement: call a stool a chair, confuse shades, swap left and right. Always compare the text with the original image, especially small details — text, logos, facial expressions. If the image is low quality or overloaded with elements, the risk of error increases, and verification takes more time.
Separate observation from inference. The service may mix them in one sentence: «A person looks tired» — this is an inference, not an observation. Decide whether such an inference is appropriate in your context. For official documents, leave only observations. For artistic text, inferences are useful, but they need to be correlated with the intent and possibly rephrased if necessary.
Consider format and volume limitations. Image up to 4 MB, audio up to 20 MB or 5 minutes, text up to 8000 characters in the main field and up to 1000 in the details field. There is no batch processing: each file is loaded and processed separately. There is no auto-saving: the result must be copied or downloaded as TXT manually. This affects how you integrate the service into your process.
Verification: the service wrote «A red book lies on the table». You open the image and see that the book is blue. You correct the error manually.
When the service is not suitable
Do not use the service where guaranteed accuracy is required: medical images, legal documents, financial reports. An error in these areas can have serious consequences, the service does not provide guarantees and may make mistakes. It is also not intended for generating photorealistic images, audio or video — this is not its task.
The service is not suitable for large volumes of images. The absence of batch processing and integrations means that each file must be loaded and processed separately, and the result copied manually. If you have hundreds of images, the time for manual operations becomes unacceptable, and you should consider other tools.
The service is not suitable if you need to extract text from an image and edit it as a document. There is no OCR in proofreading mode, no export to DOCX or PDF. You will get only a textual description that can be downloaded as TXT. If the task is to preserve the formatting of the original text, the tool will not cope.
Situation: you need to process 500 product photos for a catalog. The service is not suitable due to the absence of batch processing.
Practical example: describing an image for an article
Suppose you are writing an article about handicrafts and AI and want to illustrate it with a photo of a workplace: table, laptop, coffee cup, notebook with pen, lamp. You upload the image to the service «What's in the picture» and enter the request: «Describe main objects, their colors and arrangement. Separately indicate what is visible exactly, what is an assumption». The service returns text: «On the table stands a silver laptop, to the right of it — a white cup with coffee, to the left — a notebook with a green cover and a pen. Behind — a lamp with warm light. Everything is located on a wooden surface. Presumably, this is a workplace».
You verify the description: everything is correct, except for the arrangement — the cup is actually on the left, the notebook on the right. You correct it. Then you ask: «Formulate a conclusion about what the person at this workplace is doing». The service answers: «Probably, a person works at a laptop, possibly writes text or programs. The presence of a notebook and a pen suggests that he is taking notes». You use this conclusion as a draft and edit it to the style of the article.
In the end you get a ready description that saves time. But you do not trust the service completely: you manually verified the arrangement of objects and corrected the conclusion. This example shows a working scheme: the service gives a draft, you give the final check and editing.
Request: «Describe main objects, colors and arrangement. What is visible exactly, what is an assumption?» Result: text description, which you edit.
How to integrate the service into the workflow
Determine the stage at which the service is useful. It works best at the information gathering stage: you upload an image and get a draft description. Then at the editing stage you verify and supplement it. Do not use the service at the final stage without verification — this increases the risk of error in the finished text.
Keep a log of requests and results. Since history is not saved, record successful formulations and typical errors. If the service often confuses left and right, add a clarification to the request: «Indicate the arrangement of objects relative to each other». If you need text in another language, use the language selection feature.
Remember the limitations: file size, absence of batch processing, manual copying. For large volumes of images, combine tools or choose another solution. For small tasks, where speed and flexibility are important, «What's in the picture» suits as a quick assistant for draft description.
Tip: save successful requests in a separate file to reuse them.
The service «What's in the picture» is a tool for describing visible details of an image and answering questions about it. It speeds up routine work but requires manual verification. Delegate it primary description and rough conclusion drafting, always verify the result with the original image. Consider limitations: no batch processing, no DOCX/PDF export, no auto-saving, no integrations; image up to 4 MB, text up to 8000 characters in the main field and up to 1000 in the details field. Do not apply the service where guaranteed accuracy is required. All examples in the article are fictional and provided for illustration. The best result comes from combining manual work and AI, when you control the process and are responsible for the final text.
Ouvrir l'outil: Description d'image ↗Les traductions de l'interface sont assistées par machine et peuvent nécessiter des améliorations.