Atlas 4cwembs21w

HeyGen Help Center

Instant Highlights is a beta feature that automatically turns your long videos into engaging highlight reels. Supported by HeyGen's advanced AI technology, this tool aims to help content creators and teams to promoting their content.

Click here for a quick video showcasing our new Instant highlights!

Paste the video URL or Upload the video you want to create a highlight reel from.

Please note that when submitting your video, you will be selecting the 

 language of the video (its original language). This feature does

 the highlights into another language. For that, you can use our Video Translate feature later!

3. Let our AI automatically identify and create the highlights for you.

4. You can also share your instant highlights to your socials! 

Things to know when using Instant Highlights:

For both Uploaded videos and YouTube links:

Available to anyone with a HeyGen user account, Paid or free!

Please let us know how you feel by leaving us a feedback!

New Feature Alert🚨 Our Instant Highlight is now live on Heygen platform!

Quick Start Guide: Instant Highlights

Getting started with Instant Highlights

Connecting a Interactive Avatar to a Large Language Model, such as ChatGPT or Claude, results in a powerful and engaging interactive experience. However, considering that all of the Interactive Avatar’s responses will be spoken, we have a few best practices to recommend when providing your Interactive Avatar-connected LLM with instructions for guiding interactions with your users.

In general, LLMs like organization and clarity: you are likely to get better results by providing an organized system prompt to your LLM, as opposed to a stream-of-consciousness dump of everything you want the Interactive Avatar to do and know and say.

Therefore, we provide a three-part prompt that will help your Interactive Avatar adopt a certain 

Every time that you respond to user input, you must adopt the following persona:

We highly recommend specifying a name for your Interactive Avatar - otherwise, the LLM in the background will adopt its default persona, which is simply “a helpful AI assistant”.

When providing this name in the Persona section, however, be aware that the LLM may take you literally. If you put, “say your name is Bruce Wayne” in the instructions, the LLM may repeat its name in every reply. That means that a user could ask, “What services do you offer?” and your Interactive vatar will reply, “My name is Bruce Wayne, and we offer the following services…” Not ideal.

An alternative to putting “say your name is Bruce Wayne” in your instructions could be, “Whenever appropriate to state your name, remember that your name is Bruce Wayne. DO NOT always include “As Bruce Wayne” in every response you give.”

Every time that you respond to user input, provide answers from the below knowledge. Always prioritize this knowledge when replying to users:

We find that it works well to separate any ‘knowledge’ that you want the Interactive Avatar to have at its disposal from the ‘Instructions’ that are meant to guide the flow of conversation. You can think of this as the set of ‘facts’, like resources in a database. If you are using a vector database or some external source of knowledge, you can make that clear in your Knowledge section.

If you want your Interactive Avatar to act as a Batman trivia expert, for example, you may want to include facts about Batman in the ‘Knowledge’ section in your prompt:


Batman first appeared in “Detective Comics #27”, which was released in May 1939.

You must obey the following instructions when replying to users:

The Instructions that you provide the LLM in order to guide its interaction will have the biggest impact on the kinds of conversations that your Interactive Avatar has with users. We find that a simple and clear bulleted list of instructions is the best way to guide the LLM.

There’s no limit as to what role your Interactive Avatar can perform - it’s up to you and your imagination! However, there are some limitations you may want to place on the LLM that will make for a better user experience.

Because your Interactive Avatar will be speaking out loud, keep in mind that most users will not want to hear several minutes of uninterrupted speech. It likely makes sense to put some limit on the length of the responses that the Interactive Avatar will give. A three-sentence limit is a good place to start, unless your users will expect lengthy replies.

Imposing a word limit on the sentences that your Interactive Avatar speaks helps it give concise replies. The HeyGen Interactive Avatar API accepts ‘tasks’ of up to 1,000 characters, but sentences of up to 50 words - which, combined with the 3 sentence limit - will keep the Interactive Avatar’s replies to a generally acceptable length.

LLMs have a tendency to introduce themselves too often. You can include in your instructions a directive to “As a general rule, don’t introduce yourself outside of the beginning of the conversation. Avoid the phrase “As a {insert your Interactive Avatar’s name}” in your replies unless prompted by the user to introduce yourself.”
    

You can also include tonality in these instructions. For example, maybe you’d prefer your Interactive Avatar to appear more or less formal, depending on the Persona it is adopting.

Do you want your Interactive Avatar to decline to respond to questions that don’t fall within the purview of the Knowledge you have provided it? If so, consider adding that to your Instructions: “Politely decline to answer questions about anything other than the history of the Batman character and the Knowledge provided above.

What would you like the user to get out of the interaction with your Intreactive Avatar? There is no hard and fast rule, but if the Interactive Avatar is only meant to answer questions, you can include an Instruction to “Offer to answer questions about the history of the Batman character.”
    

If the ideal outcome of your user’s conversation with a Interactive Avatar is something more concrete, consider adding a more specific guide for the conversation to your Interactive Avatar. For example, you could write, “To begin the conversation with the user, ask which products of our company they are interested in. Once you have a specific product or products that the user is interested in, tell them the detail about those products that have been provided to you in the Knowledge section. If the user asks any questions to which there are no direct answers provided in the Knowledge section, direct them to contact our sales team.

Both Anthropic and OpenAI have detailed suggestions and instructions for crafting prompts. Feel free to review them and, if you observe any undesirable behavior based on our suggestions, please let us know!

This article helps understand how to best prompt your Interactive Avatar

Prompting your Interactive Avatar

Get ready to chat! We have recently launched HeyGen Interactive Avatars; these Avatars are optimized for real-time interaction, and can speak in real-time, making them the perfect virtual support agent, salesperson, translator, expert, or any number of other roles for your business.

Think of HeyGen Interactive Avatars as a digital worker, with a realistic and engaging human presence, that you can put to any task you can dream up. Intrigued? Let's dive in!

HeyGen’s Interactive Avatar API is a technology that allows you to initiate ‘streaming’ video sessions in which a HeyGen Avatar interacts and can say text on command in real-time, with low latency.

It’s like a FaceTime call with an Avatar!

The most common use case with HeyGen’s Interactive Avatar s to connect it, via code, to a Large Language Model like ChatGPT. For example, you could initiate an Interactive Avatar video session, and then send text it text from ChatGPT for the Avatar to say. In this way, the Interactive Avatar acts as the ‘body’ of the Large Language Model, while the ‘brain’ and source of the interaction remains ChatGPT.

A developer can recreate our interactive demo experience with relatively little effort, adding custom branding and interaction patterns of their own based on your specific business needs

Interactive Avatar in Action: Real-Life Examples

You can check out our demo to see what an Interactive Avatar interaction is like: 

 added an Interactive Avatar to their Chatbot solution by visiting their 

 and clicking the ‘Chat’ widget in the bottom right hand corner:

To get started, please find the Labs tab on the bottom left side of the screen, and choose 'Interactive Avatar'.

There, your next step would be to click on 

and follow the instructions. If you'd like to use one of our Public Avatars instead, we have different ones available to you free of charge!

Please note that the output of your Interactive Avatar will not be immediate. Below are 

Engage with facial expressions (e.g., raise eyebrows, smile) to demonstrate you are attentively listening.

Speak clearly and confidently, ensuring your message is concise and to the point.

Maintain a neutral expression and simply nod occasionally, without additional facial expressions.

 Show active engagement through facial expressions, as if you are listening attentively.

 Keep a neutral demeanor and only nod, demonstrating attentive presence without extra expressions.

Try to keep the same body-position for the whole 120 seconds of the input video!

How much does it cost to make an Interactive Avatar? 

Each Interactive Avatar costs $49 per month.

How much does it cost to use my custom Interactive Avatar? 

You can use your Interactive Avatar in multiple ways. We include a low-code 'Embed' snippet that you can use to add a chat experience with your Interactive Avatar to your webpage.

Alternatively, for technical users, you can deploy your Interactive Avatar in different mediums using HeyGen's 

You can test the Streaming API for free by using your 

 to create sessions. Sessions are unique instances of the Interactive Avatar being displayed in a window on a website or app. There is no cost to creating sessions or using the Streaming API when the sessions are created with your Trial Token. 

If you are ready to move beyond testing, and plan to integrate the Interactive Avatar on your website or in your product, you will need to purchase Streaming Credits. Streaming Credits cost $0.10 for every minute of the Interactive Avatar speaking, and when you complete your purchase of Streaming Credits, you receive access to an Enterprise API Token in your Space Settings. When you use the Enterprise API Token in concert with the Streaming API, you can create up to 100 concurrent sessions, with unlimited monthly usage and session length. If you anticipate needing more than that, please get in touch with our Sales team.

Technical integration of an Interactive Avatar

We have a starter project on GitHub that developers can use to get started with Interactive Avatars from scratch, as well as an NPM Package that can easily add Interactive Avatar functionality to existing web apps:

Interactive Avatar Starter Project on GitHub

I want to add this to my site, but I can’t code. What do I do?

 that will enable most web developers to add the Interactive Avatar functionality to your website or product. Currently, adding Interactive Avatar functionality to your site requires some programming, as there is no easy “copy-pasting” our Interactive Demo to your site with any customization.

I want my users to be able to speak with the Interactive Avatar out loud. How do I do that?

Currently HeyGen does not provide this feature; you can use a library such as OpenAI’s Whisper to convert an End User’s audio to text and send that text to their LLM.

We have added Whisper ASR implementation to our 

, but an OpenAI API Key is required to enable it. Please see the 

 for instructions on where to put your OpenAI API Key to enable both Whisper functionality as well as Chat Completions API functionality.

Can the Interactive Avatar speak any language?

Firstly, remember that your user isn’t speaking with our Interactive Avatar - the use is speaking with ChatGPT or whichever LLM (Large Language Model) you have connected to our Interactive Avatar. So if that LLM can converse in multiple languages, the Interactive Avatar will speak those languages too.

The second answer is that the accent of the Interactive Avatar is actually controlled by the Voice that you assign it. The different Voice IDs on HeyGen come from different Text-to-speech providers, including ElevenLabs, OpenAI, Azure, and Google, and the accent and sound of the voices will differ between those providers.

In keeping with the aforementioned analogy, the Interactive Avatar is only a ‘body’, and you use the Streaming API to instruct the Avatar what to say. Because this text can come from anywhere, the developer needs to write code to send the text from your GPT or LLM to the HeyGen Streaming API.

Can I use a Photo Avatar as an Interactive Avatar?


Only Instant Avatars and Studio Avatars are supported. Photo Avatars are not supported.

Which voice can I use with the Streaming API?

You can use different *Voice IDs with the Interactive Avatars. To browse through the available Voice IDs, you can visit the AI Voices on the HeyGen Homepage.

The developer can use the voices provided by HeyGen Public Voices, Private Voices that are created when they create avatars or Voices brought from 3rd party sites such as ElevenLabs.

My Interactive Avatar makes weird motions. How do I improve it?

When filming footage to turn into an Interactive Avatar, specifically, it is recommended to:


Restrict movement of hands, arms, and head to ‘micro movements’
    

For additional Avatar creation tips please visit our 

We currently offer one remake of Interactive Avatars if you are not satisfied with the end result. This is the only remake that we offer free of charge. Please contact 

HeyGen Avatars that respond in real-time!

Interactive Avatar 101: Your Ultimate Guide

When creating your Interactive Avatar, we offer two options to record your footage:

Upload a Google Drive or local video file

Please keep in mind that uploading your footage from a higher quality camera is the better option to create a higher quality avatar.

 While the back-facing camera of a modern smartphones is adequate to record footage, use a professional-grade camera for best results.

 Choose a quiet space with good lighting and a simple background without too much detail.

 For an Interactive Avatar, we require you to submit 2 minutes of footage depicting the subject in three different ‘modes’: listening mode, idle mode, and speaking mode.

 We recommend filming in 1080p for higher avatar quality. 4K at 60FPS is even better.

 Do not edit or cut the footage in the middle. It should be a continuous shot.

 Stay still and avoid sudden head motions or unnatural body movements. Be natural and calm. Your avatar will perform exactly as you do.

 Maintain direct eye contact with the camera at all times. Avoid looking around or up and down during the recording.

 We recommend keeping your hands mostly out of frame while filming a Interactive Avatar. If you choose to include your hands in the footage, keep them below chest level, and try to keep them mostly in a clasped or still position. Do not make exaggerated gestures while filming.

 Take pauses with lips closed between sentences and look at the camera.


Check out our Interactive Avatar 101 Guide

Here you can find some advanced and detailed instructions for creating an optimal Interactive Avatar.

Interactive Avatar Advanced Instructions

Create personalized videos and send customized videos to your contacts

HeyGen Labs Features