Avatars
April 8, 2026 ¡ Last updated on July 9, 2026
How to get the best results with Avatar V and Avatar IV in HeyGen
# Editing
# Avatar
# Avatar V
# Avatar IV
Avatar V and Avatar IV best practices
Want to create a professional, authentic-looking avatar? This guide covers everything you need to record high-quality footage for your Digital Twin so your avatar captures your real expressions, gestures, and delivery style.
Once your avatar is ready, pair it with a voice that sounds just like you. See best practices for cloning your voice to complete the experience.
ďťż
What's covered in this guide
- How the Digital Twin works
- Camera setup
- Audio
- Lighting and environment
- What to wear
- How to perform on camera
- Uploading your footage
- After recording: looks and personal model
- Common mistakes to avoid
ďťż
How the Digital Twin works
A Digital Twin is your personal AI avatar, built from a short 15-second video of you. HeyGen uses that clip to learn your specific gestures, expressions, cadence, and mannerisms, then lets you animate any photo of yourself with that motion.
This means you can change your outfit, your setting, or your angle as many times as you want. The appearance is flexible. The motion stays unmistakably yours.
Once created, your Digital Twin is powered by a motion engine that controls how it moves and performs. Avatar V is the recommended engine for real human Digital Twins and is selected automatically when you have a video-based look. Avatar IV is available for virtual or non-human avatars and scenarios where custom prompting is needed.
For most users creating a personal Digital Twin, Avatar V is what you'll be working with.
Getting the most out of your Digital Twin comes down to three inputs you control:
Input | What it affects |
|---|---|
Motion recording | How your avatar moves, gestures, and expresses itself |
Base look photo | The identity reference for every outfit and scene you generate |
Voice clone | How natural and authentic your avatar sounds |
Optimize all three and the results are noticeably stronger than getting just one right.
ďťż
Camera setup
You don't need a professional studio. A modern smartphone, webcam, or camera all work. Avatar V is primarily extracting your motion and expressions, not your environment, so your setup matters much less than your face being clearly visible and well lit.
Minimum requirements:
- Resolution: 1080p at 30fps or higher (4K at 60fps gives better results)
- Orientation: landscape (16:9) or portrait (9:16) both work
- Camera at eye level, not angled up or down
- Distance: about 2 to 3 feet from the lens to avoid distortion
If using a smartphone:
- Use the best camera it has, cinematic mode at 4K gives the best results
- Lock focus by tapping and holding your face until a lock icon appears
- Lock exposure the same way to prevent brightness shifts during recording
- Do a short test recording with some movement first to confirm focus and exposure stay consistent
ďťż
Audio
- Record in a quiet room with no background voices, music, or echo
- A good laptop mic or smartphone mic works, an external microphone improves quality but is not required
- Your recording must include your voice speaking, even if you plan to use a different voice later. The model needs your audio for accurate lip sync.
ďťż
Lighting and environment
Lighting
Good lighting is the most important environmental factor. The model needs to see your face clearly to capture your features accurately. Poor lighting makes it harder to maintain your likeness across generated looks.
- Natural light from a window positioned directly in front of you is ideal
- Use soft, indirect light, avoid direct sunlight, which causes harsh shadows and overexposure
- Avoid backlighting, which silhouettes your face
- Avoid lighting that shifts or changes during recording
ďťż
Background
Your background matters much less than it did with older avatar models. A plain wall with natural light is perfectly fine. That said, a clean, simple background makes it easier to remove or replace later if needed.
ďťż
What to wear
- Solid colors work best, avoid stripes, busy prints, or heavy textures
- Avoid large logos or text, which can cause artifacts or distortions in generated looks
- A good rule of thumb: dress the way you would for a video call where you want to look put together
- Glasses, jewelry, and accessories are fine but may appear slightly less precise
- Facial hair and makeup are both fine, record exactly as you are
ďťż
How to perform on camera
Energy and expression: the most important tip
This is the single biggest factor in avatar quality, and the most common mistake. The energy you put in is exactly the energy you get out. A flat, stiff recording produces a robotic avatar. Genuine expression and energy produces an avatar that feels natural and believable.
Be more expressive than feels natural, not over the top, just a noticeable step up from how you'd normally sit in front of a camera.
Avatar V is audio-driven. Your vocal delivery directly controls how expressive your avatar looks. An energetic, expressive delivery produces an expressive avatar. A monotone delivery produces a flat one.
ďťż
Eye contact
Look directly into the camera lens throughout the recording. Avoid looking at the screen or scanning around. If you're using a script, try to spend the first 15 seconds looking straight at the lens before referring to it, this ensures your opening looks natural and direct.
ďťż
Facial expressions
Use visible expressions, especially during pauses. Staying expressive during quiet moments helps the model create natural idling states between sentences.
ďťż
Hand gestures
Gestures are encouraged and make your avatar more natural. Keep them within the frame, avoid covering your face, and keep movements subtle and generic rather than big and fast. The gestures you use in your recording are what your avatar will replicate.
ďťż
Body movement
You can move naturally, but controlled, centered movement produces the most consistent results.
ďťż
First 15 seconds
Minimize large movements in the first 15 seconds, you can move naturally, but avoid sudden wide gestures or position changes in this window. After that, feel free to incorporate natural movement.
ďťż
Do a second take if needed
Review your footage before submitting. If the energy was flat or the delivery felt stiff, re-record. It takes 15 seconds and it's worth it.
Keep in mind: Your avatar reflects exactly what you record. Coughs, unusual movements, robotic delivery, or echoey audio will carry over. Clean, natural, expressive footage produces the best results.
ďťż
s.
ďťż
Uploading your footage
- The recording must be one continuous clip. Do not edit or splice clips together.
- You can adjust brightness, contrast, or saturation before uploading if needed.
- HDR footage is fine if that's your camera's default setting.
Consent: The consent step is baked into the recording itself. At the end of the clip, you'll read a short script that includes your unique code. This confirms you are aware of and agree to the avatar creation, in line with HeyGen's AI safety protocols. Make sure to read the code, this part is required for the process to complete.
ďťż
After recording: looks and personal model
Choosing your base look
Once your avatar is processed, you'll choose a base look. This is the identity photo Avatar V references every time it generates a new look for you, it's the foundation everything else is built from.
- A half-body or close-up shot works best
- Your face should be clearly visible
- Pick a photo where you like how you look
- Subtle expression, minimal accessories
If the auto-generated base look isn't quite right, you can upload your own photo instead. Experimenting with a few different base looks is worth it, a strong base consistently produces better outputs.
ďťż
Generating new looks
You have two options:
Scene library: Pre-built looks with different outfits, environments, and settings. One click puts your face into the scene.
Custom prompt: Describe exactly what you want, outfit, location, lighting, style. You can prompt a side-angle look, a specific setting, a particular outfit, and the model generates it while maintaining your likeness.
If a generated look is close but not perfect, use the edit option to fine-tune the outfit, background, or lighting.
Tip: Always use the same base look when generating new ones. Switching base looks can cause slight inconsistency across your look library.
ďťż
Training a personal model (for best results)
For the highest consistency, train a personal model on your uploaded photos:
- Minimum 10 photos, 30 or more recommended
- Use uploaded real photos as the foundation â you can include a few generated looks you like, but base it on real photos
- Include variety: different angles (front, side, 45 degrees), different lighting, different expressions
- Include any angle or lighting you plan to generate looks in â the model can only work with what it has
- Training takes around 15 to 20 minutes
The instant pipeline is fast and often good enough. The personal model produces more consistently accurate results, especially for side angles and less common lighting conditions.
ďťż
Motion reference in AI Studio
When creating a video in AI Studio, open the advanced settings next to the motion engine selector. You'll find a motion type option that lets you control how expressive your avatar moves. You can also assign different motion reference clips to different scenes.
Match your motion reference to the tone and angle of what you're generating:
- Use a high-energy recording as your reference for dynamic, expressive scenes
- Use a calm recording for natural, subtle output
- Use a side-angle recording when generating from a side-angle photo â mismatched angles produce noticeably less consistent results
ďťż
Common mistakes and how to fix them
Problem | Likely cause | Fix |
|---|---|---|
Avatar looks stiff or robotic | Flat, low-energy motion recording | Re-record with more expression, gestures, and vocal energy |
Avatar looks nothing like me | Poor lighting or weak base look photo | Improve lighting so your face is clearly visible; try a different close-up base look photo |
Generated looks are inconsistent | Base look photo is blurry, distant, or has heavy accessories | Use a clear close-up with a subtle expression and minimal accessories |
Voice doesn't sound like me | Using audio from the motion clip instead of a dedicated voice clone | Record a standalone voice clone and iterate until it sounds right |
Output doesn't match the photo angle | Motion reference and photo angle are mismatched | Use a motion reference recorded at the same angle as your photo |
Avatar gestures randomly or unnaturally | Motion recording had erratic or overly complex movements | Re-record with subtle, generic gestures |
Background noise in the output | Noisy recording environment | Re-record in a quiet room |
Looks drift or change across videos | Switching base looks between generations | Stick to the same base look for consistency across your library |
ďťż
Like
Comment (1)
Popular
ďťż
Dive in
Related
Guide
Seedance Avatar Shots prompting guide: get the best results in HeyGen
Apr 8th, 2026 ⢠Views 2.1K
1:30
Video
HeyGen latest UPDATE with Avatar IV and AI Studio is AMAZING!
By Cifer Fusion â˘Â Jun 5th, 2025 ⢠Views 1.3K
58:04
Video
HeyGen Product Updates: Avatar IV, HeyGen's Most Lifelike AI Avatar
By Nik Nolte â˘Â May 22nd, 2025 ⢠Views 1.7K
Guide
Seedance Avatar Shots prompting guide: get the best results in HeyGen
Apr 8th, 2026 ⢠Views 2.1K
1:30
Video
HeyGen latest UPDATE with Avatar IV and AI Studio is AMAZING!
By Cifer Fusion â˘Â Jun 5th, 2025 ⢠Views 1.3K
58:04
Video
HeyGen Product Updates: Avatar IV, HeyGen's Most Lifelike AI Avatar
By Nik Nolte â˘Â May 22nd, 2025 ⢠Views 1.7K