Analyze Images in Vibe

Update November 2, 2026 — Le Chat is now Vibe. As of May 28, 2026, Le Chat has been rebranded as Vibe (still at chat.mistral.ai): a single agent for work and code, carrying over your conversations, settings, and subscriptions. The legacy Le Chat experience remains accessible in Chat mode; in Work mode, agents have become skills (invoked with @), memories have become the knowledge base, and deep search has become the Deep Research skill. The plans are now called Mistral Free, Pro, Team, and Enterprise. This lesson uses the name Vibe; note that some menus described here may have changed names or locations.

Your first image analysis

Mistral Vibe lets you analyze any image in seconds. The process is as simple as sending a message: upload your file, ask your question, and the multimodal model handles the rest. This lesson shows you exactly how to proceed, what types of analysis are possible, and most importantly, how to phrase your requests to get truly useful answers. That distinction is what separates a gimmick from a real work tool.

How to upload an image

You have three options, and you will quickly adopt the one that fits your workflow. The fastest is drag-and-drop: grab the file from your desktop or file explorer and drop it directly into the conversation area. If you prefer using a file selector, click the attachment icon (the paperclip) next to the input field. Finally, copy-paste is unbeatable for screenshots: copy the image to your clipboard and paste it into Vibe with Ctrl+V (or Cmd+V on Mac). No need to save an intermediate file.

Accepted formats cover common image types: JPEG, PNG, WebP, and GIF. The maximum size depends on the model used, but in practice, photos taken with a smartphone or standard screenshots work without issues.

Types of analysis possible

Once your image is uploaded, the model can answer a wide variety of questions. The first instinct is general description: ask “What do you see in this image?” or “Describe this photo,” and the model identifies the main elements—objects, people, landscapes, colors, visible text. This is useful for generating an accessible description of an image or quickly verifying its content before sharing it.

Next comes identification: show a monument photographed on vacation and ask “What monument is this?”, photograph a leaf and ask “What type of plant is this?”, or look for “What brand is visible on this product?”. The model draws on its knowledge to attempt an identification. It is not infallible—we will come back to this—but results are often surprisingly accurate for common objects.

The third category is text reading, which includes basic built-in OCR. If your image contains text—a sign, business card, poster, or screenshot—the model reads and transcribes it. You can ask “Read the text in this photo,” or target specific information directly: “What is the phone number on this business card?”. This saves considerable time when you need to retrieve contact details from someone you met at a trade show without manually re-entering them.

Finally, the model does more than just describe: it interprets. You can ask whether an outfit is appropriate for a job interview, whether a photographed dish seems nutritionally balanced, or what the main message of an advertisement is. This ability to provide contextual interpretation is what distinguishes a modern multimodal model from a simple object recognition tool.

Asking good questions

The quality of the analysis depends heavily on how you phrase your request. A vague question yields a vague answer: “Analyze this image” will get you a generic description, whereas “What are the sales figures shown in this chart?” gives you exactly what you need. Also provide context when the image alone is insufficient: “Here is a photo of a dish I prepared. Estimate the approximate calories” steers the model toward the right type of analysis. Do not hesitate to specify an output format—“List the visible elements as bullet points” or “Summarize in one sentence what this image shows”—to get a response you can use immediately.

Limits to know

The vision model is powerful, but you must keep its limitations in mind. With very small or blurry text—a photo of a contract taken from a distance, for example—it may misread or even invent plausible characters. This is the sneakiest pitfall because the error is not immediately obvious. Dark, overexposed, or highly abstract images will similarly yield less reliable results. Also consider privacy: the model reads all data visible in the image, so be careful with sensitive documents. A bank account number (RIB) left in the corner of a screenshot will be read just like everything else. Finally, each image is analyzed independently unless you send multiple images in the same conversation: there is no visual memory between distinct conversations.

Hands-on practice

To lock in these habits, try three exercises in Vibe right now. First, take a photo of an object on your desk and ask the model to describe it—compare its description with what you see. Next, take a screenshot of a news article and ask for a summary of the headline and lead paragraph. Finally, photograph a receipt or ticket and ask it to extract the total amount: this exercise is closest to real professional use and will immediately show you the reliability level of the built-in OCR.

Key takeaways

• Three upload methods: drag-and-drop, paperclip button, copy-paste
• The model can describe, identify, read text, and interpret image context
• Specific questions yield better results than vague requests
• Watch out for poor-quality images and sensitive documents