> ## Documentation Index > Fetch the complete documentation index at: https://docs.bey.dev/llms.txt > Use this file to discover all available pages before exploring further. # Custom Avatars > Create an avatar exclusive to your account from a training video or a single image. (New) You can now create custom avatars yourself in the [Studio's Avatars section](https://bey.studio/avatars) 🎉 A custom avatar belongs to your account and can be used for brand representation, product walkthroughs, or a persistent digital persona. There are two ways to create one: * [From a training video](#avatars-from-videos): A few minutes of recorded footage of one person. This produces the highest quality result and is what most of this page covers. * [From a single image](#avatars-from-single-images): A single photograph, with no recording session required. ## Avatars from Videos The quality of a video-trained avatar depends almost entirely on the training video. The sections below cover how to record it, what to send, and what happens next. ### How It Works Follow the setup and recording instructions below to record a good avatar training video. Most people get a good take within two or three attempts. Go to the [Studio's Avatars section](https://bey.studio/avatars), click *"Create Custom Avatar"*, and upload the video. Follow the Studio's prompts to record a consent verification video. Submit your video and wait until the avatar has been created. ### 1. Setup #### Camera and Framing * A good phone camera or laptop webcam works well. Photo Booth or QuickTime Player on a newer Mac are solid options. Even better is a DSLR or mirrorless camera on a tripod, which will give you the sharpest result. * Position the camera at **eye level**. Shooting from below or above distorts the face and degrades the result. * Record at **1080p**. * Frame from mid-chest up, with a small amount of headroom above the head. * Lock focus and exposure if your camera allows it. Auto exposure shifting mid-take causes problems. * Don't zoom, pan, or reframe at any point, and don't move closer to or further from the camera during the take. Record locally to a file. Avoid streaming software or any tool that sends video over an internet connection before saving. The compression artifacts aren't recoverable. #### Lighting * Use a well-lit room with the light coming **from the front**, so your face is evenly lit. * Avoid strong shadows across the face. * Watch what's behind you. A very bright window or lamp behind your head will pull exposure down and darken your face. If you can't move, add light in front of you to compensate. * Keep the lighting consistent for the whole recording. Don't let it change intensity, direction, or colour partway through. #### Background * The background should be **static**. Nothing moving, no one walking behind you. * **Only one person in frame.** No other faces anywhere in the video, including in the background, on a screen, or in a poster or photo on the wall. * Choose a background that contrasts with your hair and clothing. Dark hair against a dark wall is a common failure. * It doesn't have to be a blank wall. A normal room works, as long as it's still and doesn't compete with your face. * Green screen is optional and isn't required for a good result. #### Wardrobe and Hair Anything that moves independently of your body creates artifacts. * Pin down anything loose: open collars, scarves, lanyards, drawstrings, jacket flaps. * No dangling earrings or swinging necklaces. * No hats with brims that shadow the eyes, and no sunglasses. * Keep hair off the forehead and away from the face. Tie long hair back so nothing falls across your face mid-take. * Avoid clothing that matches the background colour. * Wear something that contrasts clearly with your skin tone. Skin-coloured clothing may blur the boundary between neck and clothing. **Long beard?** Wearing a turtleneck or an undershirt in a similar colour to the beard helps the model separate it cleanly from the neck. Long loose hair and full beards are still harder for the model than short hair and clean-shaven faces. The tips above reduce the risk. We are actively working on improving quality across all hair types. #### Glasses Glasses generally work. To reduce risk: * Eliminate reflections of lights or windows in the lenses. Reposition the light rather than tilting your head. * Avoid transparent or very thin wire frames. * Avoid frames that sit close to the eyelid or cut across the eye. If you're unsure, record one take with glasses and one without, and try both. #### Setup Checklist **Room** * [ ] Quiet: HVAC, fans, and notifications off * [ ] No one else in frame or earshot * [ ] Background static, no other faces visible **Camera** * [ ] Eye level, chest-up framing * [ ] 1080p, recording locally to file * [ ] Focus and exposure locked * [ ] Storage and battery for a 5-minute take **Light** * [ ] Lighting from the front, face evenly lit * [ ] No shadow across the face, nothing bright behind your head * [ ] No reflections in glasses **You** * [ ] Clothing contrasts with skin tone and background * [ ] Hair back and off the face * [ ] Nothing loose or dangling * [ ] Hands resting below the neck ### 2. Recording the Video #### Video Specifications | | | | -------------- | ------------------------------------------------------ | | **Length** | 2–3 minutes | | **Resolution** | 1080p recommended | | **Format** | MP4 or MOV | | **Structure** | 20s silent opening, 2-3 mins speaking, 5s still ending | | **Camera** | Eye level. Phone, webcam, or DSLR on a tripod | | **Framing** | Chest-up, one person only, hands below the neck | | **Background** | Static, contrasting with hair and clothing | | **Audio** | Clean and quiet, no other voices | #### Recording Structure We recommend structuring the video in three phases: Record in one continuous take. Cuts, trims, and splices anywhere in the video cause problems. Start the recording and stay **silent and still for about 20 seconds** before you speak. * Keep your mouth closed the entire time. Don't open it to speak, yawn, or reset. * Imagine someone is talking to you and you're listening attentively. That's the expression we want. * Gentle, natural movement is welcome: a small smile, slight head nods, subtle shifts of expression. * Avoid large or exaggerated movement. This opening gives us the material we need for a natural-looking avatar when it isn't speaking, so give it the full 20 seconds. **Summary:** silent, mouth closed, listening expression, gentle nods After the 20 seconds of silence, ease into speaking. * Start relaxed. Something like "Hey, my name is ..." works well. * Try not to jump in abruptly with high energy, like "Hey! What's up everyone?!" * Aim for **2–3 minutes** of speech. Shorter recordings give the model less to work with. * Hold steady eye contact with the camera throughout. * Keep your head and body stable. Try not to tilt your head or shift your posture. * Keep your shoulders fixed. Avoid shrugging or lifting them. * Natural micro-expressions are good. Gentle nods and subtle smiles are fine. Large gestures are not. * **Close your mouth completely after each sentence**, with a short natural pause of about half a second. This produces noticeably smoother speech animation. * Aim to be positive and smiling from the inside, while keeping your outward expression neutral, calm, and friendly. * Keep your hands below your neck so they stay out of frame. You don't need a script. Talk about your work, your product, your day, anything you can speak about comfortably for a few minutes without stumbling. Natural speech works better than a formal read. **Summary:** Speaking naturally for 2-3 minutes, eye contact, stable posture, mouth closed between sentences * Wrap up simply. A short sign-off is enough. Don't trail off mid-thought or end on a rushed sentence. * After you finish speaking, stay still and silent for a few seconds in the same position. * Keep your shoulders and head steady through the whole pause. #### Recording Checklist * [ ] 20 seconds silent and still at the start, mouth closed * [ ] Easing in around 0:21 with a relaxed opener * [ ] 2–3 minutes speaking, eye contact, shoulders steady * [ ] There are pauses between sentences * [ ] Simple close, then 3–5 seconds still * [ ] One continuous take, no cuts * [ ] Face is visible at all times and looking directly at the camera * [ ] There were no other faces or voices in the background during the recording ### Best Practices #### Record Clean Audio To obtain an avatar with good lip-synchronization, the audio needs to be audible and clean. * Record in a **quiet space with no background noise**: no air conditioning, traffic, keyboards, fans, or notifications. * Avoid echo. A soft-furnished room is better than an empty hard-walled one. * Speak clearly at a moderate pace. Natural pauses are good. Avoid rushing and filler words. * A lavalier or USB microphone gives the best result, but built-in camera or laptop audio is fine in a quiet room. **No other voices at any point.** If someone else speaks during the recording, your voice won't be captured cleanly and the voice clone will be inaccurate. Don't call "action" or "cut", and don't have anyone feeding you lines. #### Test Setup Before Full Recording Record 30 seconds, stop, and watch it back at full size before doing the real take. Check: * Is the face sharp when you pause on a frame? * Is there a shadow across either side of the face? * Does your hair separate clearly from the background? * Is the audio clean, with no hum, echo, or room noise? * Are your lenses free of reflections? * Are your hands out of frame? Two minutes here saves a full reshoot later. #### Avoid Common Mistakes In rough order of frequency: * Missing or shortened silent opening * Very short recordings * Cuts or trims anywhere in the take * Another voice audible in the recording * Camera above or below eye level * Shadow across one side of the face, or a bright light behind the head * Hair or clothing too close in tone to the background * Another face visible in the background * Hands entering frame * Shoulders shrugging or posture shifting during the read * Loose hair falling across the face partway through * Auto exposure or white balance shifting mid-take ## Avatars from Single Images Custom avatar creation from single images is only available on demand for customers on paid plans. Contact [support@beyondpresence.ai](mailto:support@beyondpresence.ai) with your image. You can also create an avatar from a single photograph, with no recording session required. Quality won't match a video-trained avatar, but it's fast and needs nothing more than a good headshot. The photo does more work here than a single frame does in a video, since everything the avatar produces is derived from it. A neutral, evenly lit, front-facing image gives the model the most room to animate. | Requirement | Details | | ------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Aspect ratio** | 16:9 landscape works well. Square (1:1) or portrait (9:16) is also fine. | | **Head & body position** | Straight and upright, facing the camera directly. Minimal head tilt or rotation, shoulders square to the frame. | | **Framing** | Head-and-shoulders, face centered, with a little headroom. Nothing cropped (full head visible), face not too small and not an extreme close-up. | | **Eyes** | Both eyes fully open and clearly visible, looking toward the camera. Not squinting, mid-blink, or covered by hair. Images with closed or partially closed eyes won't work. | | **Lips** | Closed, or a gentle closed-mouth smile. Avoid open mouths and wide teeth-showing grins. | | **Expression** | Neutral or very lightly smiling, with a relaxed, natural face. Avoid strong or exaggerated expressions — a broad smile, laughter, or raised eyebrows will carry into every frame the avatar produces. | | **Lighting** | Soft and even, lighting the face from the front. Avoid harsh shadows, strong backlight, and heavily coloured or rim lighting. | | **Subject** | One person only. Nothing covering the face or mouth. Glasses are usually fine. | | **Resolution** | Sharp, high-resolution image with the face taking up a good portion of the frame. 1080p or higher preferred. | | **Format** | JPG or PNG. | Avoid heavily filtered or retouched photos, and images that have been upscaled from a small original. Both tend to produce soft or unstable results. ## Consent and Ethics Every avatar requires the explicit, informed consent of the person being replicated, whether it's created from video or from a single image. We verify this during review. If the subject is a paid performer or an employee, make sure your agreement with them covers the creation and ongoing use of an AI likeness, the scope and duration of that use, and what happens if they leave. We don't create avatars of public figures, or of anyone who hasn't personally consented. ## Related topics - [Beyond Presence Avatar Models](/concepts/avatars/index.md) - [Glossary](/learn/glossary.md) - [Concurrency & Quotas](/production/concurrency.md) - [Troubleshooting](/production/troubleshooting.md) - [Studio](/concepts/studio.md)