Exam objective 5.2 asks you to know how to use AWS AI services for vision and content tasks, so you must understand the key tools: Rekognition, Textract, Comprehend, Polly, and Lex. These services allow you to add powerful AI capabilities to your applications without needing to build or train any machine learning models yourself. For the AIF-C01 exam, you only need to know what each service does and when to use it.
Jump to a section
A simple way to picture AWS AI Services for Content and Vision
A busy restaurant manager is the central figure in this scene. They have to handle a flood of incoming tasks every day: reading handwritten reservation notes, checking food quality in photos sent by suppliers, typing up menu changes, listening to voicemails from upset customers, and answering the same questions from diners again and again.
Instead of doing all of this manually, the manager hires a team of specialist assistants. One assistant can read any handwritten note and turn it into typed text. Another can look at a photo of a dish and tell you exactly what ingredients are visible. A third can listen to a voicemail, understand the customer\'s frustration, and summarise the complaint. A fourth can take a typed menu and read it aloud in a natural voice. A fifth can handle the most common questions from diners without bothering the manager.
Each assistant has a single, narrow talent. None of them can cook or take payments. But together, they free up the manager to focus on the bigger picture. Now, imagine the manager is your application, and the assistants are AWS AI services. Each service is a specialist. You don\'t build the intelligence yourself. You just call on the right assistant for the task. That is exactly how AWS AI Services for Content and Vision work for an IT professional.
AWS offers a set of pre-built AI services that solve common problems with images, text, speech, and conversation. These are called \"AI Services\" because they are fully managed — AWS handles all the complexity of training and running the underlying machine learning models. You simply send data to the service and get back a result. This is a core concept for AIF-C01 because the exam tests whether you can match the right service to a given business scenario.
Let us break down each service in plain language.
Amazon Rekognition is the service for analysing images and videos. You give it a picture, and it can tell you what objects are in it (a dog, a car, a person), whether there are any faces, and even how people might be feeling (happy, sad). It can also compare two faces to see if they belong to the same person. For the exam, remember that Rekognition is for vision tasks: detecting objects, recognising celebrities, moderating content (finding inappropriate images), and identifying faces.
Amazon Textract is the service for extracting text from scanned documents. Think of it as supercharged OCR (Optical Character Recognition). While basic OCR just finds the letters on a page, Textract understands the structure. It can read a table in a PDF and give you the data in rows and columns. It can read a form and pair up each label (like \"Name\") with its value (like \"Jane\"). Textract is for document analysis. You use it when you need to turn a scanned contract, invoice, or form into structured, machine-readable data.
Amazon Comprehend is the service for understanding the meaning of text. You give it a sentence or a whole document, and it can detect the language, extract key phrases (like \"new York office\"), determine the overall sentiment (positive, negative, neutral, mixed), and identify named entities (people, places, dates, organisations). It can also do topic modelling on a large collection of documents. Comprehend is for natural language processing (NLP) tasks. You use it when you need to analyse customer feedback, sort emails, or find patterns in text.
Amazon Polly is a text-to-speech service. It takes written text and turns it into lifelike speech. You can choose from many different voices and languages. Polly uses advanced deep learning to sound natural, with proper intonation and emphasis. It is for speech generation. You use it when you want your application to read content aloud, such as in a voice-enabled app or for accessibility tools.
Amazon Lex is the service for building conversational interfaces. It powers the chatbots you interact with on websites. Lex can understand natural language (using automatic speech recognition, or ASR) and then determine the user\'s intent (using natural language understanding, or NLU). It also can respond back in spoken form using Polly. Lex is for conversational AI. You use it to create bots that can answer questions, book appointments, or process orders through text or voice.
The reason all these services exist is that building your own AI models is incredibly difficult, expensive, and time-consuming. It requires specialised data scientists, huge amounts of training data, and powerful computers. By using AWS AI services, you bypass all of that. You pay only for what you use, and you can add intelligence to your application with just a few lines of code. This is the fundamental value of managed AI services in the cloud.
Identify the Task Type
Read the business scenario and decide whether it involves an image, a document, a text block, speech output, or a conversation. This first step determines which AWS AI Service you will consider. If the task involves a photo or video, you lean toward Rekognition. If it involves a scanned form, you lean toward Textract.
Match the Task to the Correct Service
Once the data type is clear, select the exact AWS service. For image moderation, choose Rekognition. For extracting data from an invoice, choose Textract. For analysing customer sentiment, choose Comprehend. For reading text aloud, choose Polly. For building a chat interface, choose Lex. This step is the core skill tested in section 5.2.
Prepare Your Data
Ensure your data is in a format the service accepts. For Rekognition, images can be in JPEG or PNG format, uploaded to S3 or sent as bytes. For Textract, documents can be PDF, TIFF, JPEG, or PNG. For Comprehend, you send plain UTF-8 text. For Polly, you send plain text or SSML. For Lex, you define intents and sample utterances in the Lex console.
Call the Service Using an SDK or Console
You interact with these services via the AWS SDK (for programming languages like Python or JavaScript) or directly through the AWS Management Console for testing. The SDK sends an API request with your data and returns the result as JSON. For example, you call the detect_labels method of Rekognition and get back a list of labels with confidence scores.
Process the Response and Integrate
The service returns a structured response. You then write code to extract the information you need and act on it. For example, you take the sentiment score from Comprehend and store it in a database, or you take the audio stream from Polly and play it in your user interface. This step turns the AI output into a business action.
Imagine you work for a company that runs a large online marketplace similar to Etsy. Sellers list handcrafted items, and buyers leave reviews. Your job is to improve the platform using AWS AI services. You do not have a background in machine learning, but you know these services exist. Let us walk through a typical day.
First, sellers upload photos of their products. You need to make sure these images do not contain inappropriate content. You send each new image to Amazon Rekognition. The service returns a label such as \"SafeSearch\" with a confidence score. If the score is too low, you flag the image for manual review. This is content moderation.
Second, sellers also upload scanned copies of their business licences to prove they are legitimate. You need to extract the licence number and expiry date from these scans. You send each PDF to Amazon Textract. Textract analyses the document and returns a structured JSON object with the text and its location. You write a small script to find the fields you need and store them in a database. This replaces a manual data entry task.
Third, you want to understand how customers feel about recent changes to the shipping policy. You collect the last 10,000 customer reviews and send them to Amazon Comprehend. The service analyses each review and gives you a sentiment score. You find that 40% of reviews are negative. Drilling down, Comprehend can also extract key phrases like \"shipping cost\" and \"slow delivery\". You now have concrete data to present to management.
Fourth, you want to build a simple customer service bot to answer common questions like \"What is my order status?\" and \"What is your return policy?\" You use Amazon Lex to define the intents (what the user wants) and the slots (pieces of information you need, like an order ID). Lex handles the understanding and the conversation flow. It even uses Polly to speak the answers aloud if the customer calls in.
Finally, you decide to add an accessibility feature. For visually impaired users, you offer an option to hear product descriptions read aloud. You write a function that takes the product description text and sends it to Amazon Polly. Polly returns an audio stream of the description spoken in a natural female voice. The user hears the product details without needing to read the screen.
In this scenario, you did not write any machine learning algorithms. You used five AWS services, each for a specialised task. The key takeaway for your daily role is that you orchestrate these services, connecting them to your existing applications via the AWS SDK (software development kit) or via the AWS Management Console for one-off tasks. You focus on the business logic, and AWS handles the AI.
The AIF-C01 exam will ask you which AWS AI service to use for a specific content or vision task. This is the single most important concept to master. The questions are scenario-based. They give you a short description of a business problem and ask you to select the correct service from a list that includes Rekognition, Textract, Comprehend, Polly, Lex, and sometimes confusingly similar services like Amazon Transcribe (speech to text) or Amazon Translate.
Here are the exact patterns the exam uses and the traps they set.
- Question Type: Image Analysis The scenario will mention a photo, video feed, or security camera. Correct answer: Amazon Rekognition Trap: They may mention \"reading text from an image.\" That is a Textract task, not Rekognition. Rekognition detects objects, faces, and scenes. Textract extracts text. They are different.
- Question Type: Document Processing The scenario will talk about a scanned PDF, a form, or an invoice. Correct answer: Amazon Textract Trap: They might say \"using OCR,\" which is technically what Textract does, but they will offer Rekognition as a distractor. Rekognition does form text in images? No. Textract is the correct choice for any document with structured data.
- Question Type: Text Analysis The scenario involves analysing customer feedback, detecting sentiment, or extracting entities like dates and names from a paragraph. Correct answer: Amazon Comprehend Trap: They might mention \"searching for a keyword.\" That is a database search, not NLP. Comprehend is about understanding meaning, not finding exact keyword matches.
- Question Type: Speech Output The scenario wants to convert text to speech for an application. Correct answer: Amazon Polly Trap: They might offer Amazon Transcribe, which does the opposite (speech to text). Always read whether the scenario says \"converts text to speech\" (Polly) or \"converts speech to text\" (Transcribe).
- Question Type: Conversational Bot The scenario describes a chatbot or voice assistant that can understand and respond to questions. Correct answer: Amazon Lex Trap: They might say \"uses natural language understanding (NLU)\" which is part of Lex, but they could also offer Comprehend. Comprehend only analyses text — it does not conduct a conversation. Lex manages the back-and-forth dialogue.
- Key Concept to Memorise: Mis Fits The exam loves to test if you know which service does NOT fit a scenario. For example: \"Which service would NOT be used to analyse a video?\" Answer: Textract or Comprehend, because those work on still text or documents. Rekognition handles video.
- Trap: Services That Sound Similar Amazon Transcribe (speech to text) vs Amazon Polly (text to speech). Amazon Comprehend (text analysis) vs Amazon Lex (conversations). Amazon Rekognition (image/video) vs Amazon Textract (documents). You must know these pairs cold.
The exam does not ask you how to code an API call. It asks you which service to use. Practise by reading a short scenario and immediately naming the right service without looking at any options. Train yourself to spot the key phrases: \"video\" -> Rekognition, \"scanned form\" -> Textract, \"sentiment\" -> Comprehend, \"speech output\" -> Polly, \"chatbot\" -> Lex.
Amazon Rekognition is the service for analysing images and videos to detect objects, faces, scenes, and inappropriate content.
Amazon Textract is the service for extracting text and structured data (tables, forms) from scanned documents.
Amazon Comprehend is the service for natural language processing tasks like sentiment analysis, entity extraction, and language detection.
Amazon Polly is the text-to-speech service that converts written text into lifelike audio using deep learning.
Amazon Lex is the service for building conversational chatbots and voice assistants that can understand natural language using ASR and NLU.
On the AIF-C01 exam, you must match a business scenario to the correct AWS AI service, never mix Rekognition with Textract, or Pollys with Transcribe.
These come up on the exam all the time. Here's how to tell them apart.
Amazon Rekognition
Analyses images and videos for objects, faces, and scenes.
Does not understand text structure like tables or forms.
Can detect text in images but not extract it in structured form.
Amazon Textract
Extracts text and data from scanned documents, forms, and tables.
Understands document layout, returning data in a structured format.
Does not analyse the content for objects or faces; only text extraction.
Amazon Polly
Converts text into spoken audio (text-to-speech).
Used for generating speech for applications like audiobooks or voice assistants.
Output is an audio file (e.g., MP3) from text input.
Amazon Transcribe
Converts spoken audio into written text (speech-to-text).
Used for transcribing meetings, customer calls, or podcasts.
Output is a text transcript from an audio input.
Amazon Lex
Builds conversational interfaces (chatbots) with dialogue management.
Understands user intent and can manage back-and-forth conversation context.
Uses both ASR (speech-to-text) and NLU (natural language understanding).
Amazon Comprehend
Analyses static text for sentiment, entities, and key phrases.
Does not manage conversations or remember past interactions.
Purely a text analysis tool, not a dialogue system.
Mistake
Rekognition can read and extract text from any image, like a scanned receipt.
Correct
Rekognition can detect text in images, but its primary strength is object and face detection. For extracting structured data from documents (tables, forms), you should use Textract.
Both services can handle images, so beginners assume they are interchangeable. The exam forces you to distinguish them by the type of data being extracted.
Mistake
Comprehend can build a chatbot that answers customer questions.
Correct
Comprehend analyses text but does not manage conversations. It has no memory of past interactions and cannot maintain a dialogue. Lex is the service for building chatbots.
Beginners see 'natural language processing' on Comprehend's description and think it can do everything NLP-related, including conversations. The exam tests the specific boundaries of each service.
Mistake
Polly can understand what it's reading and customise the speech based on the meaning of the text.
Correct
Polly converts text to speech using advanced techniques, but it does not understand the content. It bases intonation on punctuation and SSML tags (Speech Synthesis Markup Language), not on the semantics of the words.
The voices sound so lifelike that people assume there is understanding behind them. The exam will not test this nuance deeply, but it can appear in a distractor option.
Mistake
Textract can analyse videos to detect moving objects.
Correct
Textract works on static documents (images, PDFs, forms). Analysing video content is the job of Rekognition, which can process video files frame by frame for object and activity detection.
Both services deal with visual data, so beginners blur the line between 'document' and 'video'. The exam explicitly tests the difference between these two service scopes.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Rekognition analyses images and videos to find objects, faces, and scenes. Textract extracts text and structured data from scanned documents like forms and tables. Use Rekognition for visual content analysis, and Textract for document data extraction.
No, Polly does not understand the content. It converts text to speech using rules about pronunciation and intonation, which can be further refined using SSML tags. The speech sounds natural, but there is no comprehension of the underlying meaning.
Use Lex when you need a two-way conversation with a user, like a chatbot or voice assistant. Lex manages dialogue flow and understands intents. Use Comprehend when you need to analyse a static block of text to extract entities, sentiment, or key phrases, without any conversation.
No, they do opposite tasks. Transcribe converts speech (audio) into text (transcription). Polly converts text into speech (audio). They are exact inverses. The exam often asks you to pick the correct one based on the direction of conversion.
No, AWS AI Services like Rekognition, Textract, Comprehend, Polly, and Lex come pre-trained on massive datasets. You do not need to train them. You can optionally customise some of them (like Rekognition Custom Labels or Comprehend Custom), but the basic services work out of the box.
A confidence score is a number between 0 and 100 that tells you how sure the AI is about its prediction. For example, if Rekognition detects a cat with a confidence score of 98, it is very confident the image contains a cat. A low score means the detection is uncertain.
You've finished AWS AI Services for Content and Vision. Continue through the AIF-C01 study guide to build a complete picture of the exam.
Done with this chapter?