What are ElevenLabs Agents & Flows?
ElevenLabs Agents are conversational AI interfaces. Instead of typing text for the AI to read later, you provide the AI with a "brain" (an LLM like Gemini or Claude), a set of instructions, and a knowledge base. The Agent can then have real-time, spoken conversations with humans over the phone, on the web, or via WhatsApp. This is how you can completely automate customer service.
ElevenLabs Flows is a visual workspace introduced to streamline multimedia generation. Instead of using five different websites to generate an image, turn it into a video, and add audio, Flows provides a node-based canvas where you can connect industry-leading models (like Nano Banana 2 for images and Veo 3.1 for video) directly to your ElevenLabs audio tracks to render complete AI video generation pipelines.
Building REALISTIC AI Voice Agents
The "Press 1 for Support" era is over. Learning how to deploy an omnichannel AI agent allows you to qualify leads and run outbound campaigns simultaneously.
Step 1: Configuring the Agent's Brain
Navigate to the Agents tab.
Select the LLM: Choose the underlying logic model. Gemini 2.5 Flash and Claude are excellent for balancing low latency (speed) with high reasoning capabilities.
Choose the Voice: Select a conversational voice model. Ensure you are using models optimized for ultra-low-latency turn-taking so the bot responds as quickly as a human.
For maximum security, we recommend that you set Zero Retention Mode. To do this, you need to navigate to 'Advanced Settings', then the bottom section labelled as 'Personal', and flip the switch for 'Zero Retention Mode'. When this is on, ElevenLabs will not store and use your data.
Step 2: The System Prompt Framework
A generic prompt results in a confused bot. Treat your system prompt like an employee handbook.
Role: "You are Sarah, the front desk receptionist for Peak Fitness."
Tone: "You are upbeat, patient, and use concise sentences."
Rules/Guardrails: "Never invent pricing. If a user asks for a discount, state that you do not have authorization to alter fees."
Steps/Flow: Give it a conversation map (e.g., 1. Greet, 2. Check membership status, 3. Resolve query).
Step 3: Uploading the Knowledge Base (RAG)
To stop your AI voice agents from hallucinating, use the Knowledge Base feature. Upload your PDFs, pricing sheets, or scrape your company's website. This uses Retrieval-Augmented Generation (RAG) to ensure the AI answers based only on your approved data.
Step 4: Connecting Tools and Telephony
External Tools: Connect your agent via Webhooks to platforms like Make.com or Zapier. This allows the AI to trigger a calendar booking or send a follow-up email autonomously during the call.
Telephony: ElevenLabs allows you to instantly connect your agent to a Twilio phone number. The proprietary turn-taking model ensures the AI handles interruptions naturally, pausing when the human speaks over it.
Create AI Ads with ElevenLabs Flows
If you want to create AI ads at scale, Flows is the ultimate cheat code. It transitions the platform from an audio tool to a full creative pipeline.
Understanding the Node Canvas
Instead of a linear timeline, Flows gives you an infinite canvas. You add "Nodes" (representing specific tasks) and draw lines between them to pass data. This allows for non-destructive editing - if you want to change the voiceover, you only re-run the audio node, saving time and compute credits.
Step-by-Step AI Ad Creation Workflow
The Image Generation Node:
Add an image generation node using models like Nano Banana 2 (powered by Gemini 3.1 Flash Image). Nano Banana 2 is exceptional for product placement and maintaining character consistency.
Prompt Example: "Wide shot of an elegant glass perfume bottle resting on black volcanic rock, neon pink studio lighting."
Add an Image Generation Node in ElevenLabs Flows The Video Generation Node:
Connect the output of your image node to a video generation node like Veo 3.1. Veo 3.1 excels at taking your Nano Banana 2 image and animating it with flawless physics. Just click on the 'Image' button, then on 'Video Generation'. This will add a new video generation node, already connected to your first node - the beauty of node-based AI tools. Alternatively, you can connect two nodes with a simple drag of the mouse from one action button to the other.
Action Example: "Slow camera pan to the right, water droplets subtly running down the glass."
Connect a Video Generation Node in ElevenLabs Flows The Voiceover Node:
Drop in a Text-to-Speech node and write your ad copy. Route this alongside your video output.Music and Sound Effects (SFX):
Add an ElevenMusic node for a custom background track and an SFX node for specific foley sounds (e.g., "gentle ocean waves crashing").Export to Studio 3.0:
Export the entire Flow into ElevenLabs Studio. This is the traditional timeline editor where you can fine-tune the timing, adjust the audio mixing, and generate dynamic captions.
Best Practices for Professional Automation
Category | Best Practice | Why it Matters |
Agent Testing | Use the "Agent Testing Framework" before going live. | Run LLM evaluations to ensure the bot handles angry customers and complex logic without breaking character. |
Agent Guardrails | Explicitly tell the Agent what it does not know. | Prevents the AI from confidently giving out false information to your customers. |
Flows Templates | Build reusable "Templates" for your campaigns. | Once you design a Flow for an ad, you can just swap the text prompt and hit "Run" to generate a new ad in minutes without rebuilding the pipeline. |
Conclusion
Mastering AI voice agents and ElevenLabs Flows allows you to operate a full-scale creative agency and a 24/7 customer service department from a single browser tab. Start by building an internal test agent, map out a simple visual node flow for a social media post, and gradually scale your automation.
