Prompt Like a Pro: How to Create Better AI Avatars, Voices & Motion in HeyGen
# Prompt
# Video Avatars
# Voices
# Motion Avatars
# Adding Motion
Create 100% AI avatars. Prompt their look, motion, voice, and even add products.
Create better AI avatars, voices, motion, and more with clear and simple prompts
Prompting is how you tell HeyGen what to do. With the right words, you can control how your avatar looks, speaks, and moves- no complex tools required.
Visit this guide for step-by-step instructions on how to generate a custom avatar. Then, review the steps below at the prompting stage.
Start with a prompt that covers pose, clothing, and setting.
✅ Good example:
Avatar holding a clipboard, wearing a white lab coat, standing in a hospital hallway with soft lighting
📌 You can be more detailed if needed — especially for style-driven characters:
✅ Descriptive example:
A joyful female art teacher in her late 30s, with curly auburn hair in a
messy bun, wearing paint-splattered overalls and large round earrings.
She’s holding a palette and brush, standing in front of a colorful easel
withsunlight streaming through classroom windows.
Tip: Choose “Realistic” style for your the initial avatar you create, as this gives you more flexibility when generating animated or stylized looks later.
Once your avatar is trained, you can generate up to 300 additional Looks which can be deleted and replaced at any time. Avatar Looks are additional outfits, background and poses you can create to ensure your spokesperson looks the part no matter what. Refer to our Generate Looks guide for step-by-step instructions.
✅ Format that works well:
Avatar [pose or action], [outfit], [setting or background]
✅ Examples:
Avatar sitting at a wooden table, wearing a leather jacket, in a cozy café with warm lighting
Avatar standing in Times Square, wearing a navy suit, holding a briefcase
Having trouble with achieving consistency of your avatar's appearance across the Looks you've generated? Use this retraining method:
Generate 50 looks
Pick your most consistent 10
Retrain the avatar on those
Repeat if needed
Generating new Looks with a product
Use HeyGen's Product Placement feature to upload photos of your product. Give it a clear name (e.g. “green mug”), and then use that name in your prompt:
Avatar holding green mug with both hands, smiling slightly
Young male, British accent, relaxed tone, mid-pitch, friendly delivery
Try generating multiple voices from the same prompt, as they can vary a lot. You can also adjust the stability and style exaggeration settings in Studio for more control.
Generative motion models work best when you describe physical actions clearly and directly. Keep it simple and focus on what the subject should do, and skip abstract or overly detailed visuals.
❌ the subject embodies the essence of joyful greeting
✅ the avatar smiles and waves
❌ a young woman with curly brown hair wearing a green sweater and jeans reaches out her hand for a handshake
✅ the woman extends her arm to shake hands and nods politely
❌ the camera doesn’t move
✅ the camera remains still
Generating motion will create a maximum of 10 seconds of movement.
HeyGen offers two types of motion prompting: consistent and expressive.
Consistent motion is ideal if:
Visual fidelity is crucial: It ensures that characters and scenes retain their original look without distortions.
Cinematic aesthetics are a priority: Suitable for dramatic, high-quality sequences with smooth transitions.
Minimal animation complexity is needed: Works best for polished, professional clips where movement is secondary to composition.
Expressive motion is preferable if:
Realism is the top priority: Ideal for lifelike human movement and organic animations.
Creative control is essential: It adheres closely to your artistic vision and prevents unintended artifacts.
Dynamic movement is key: Best for storytelling, character-driven animations, and energetic sequences.
For the best results, consider limiting custom motion you prompt to short clips under 10s. Use default motion (i.e., no text prompt) for longer scenes, and intercut with custom gestures for more natural pacing.
Below are suggestions for resolving the most frequent questions users have when generating avatars, voices, and motion.
Avatar doesn’t show full body
Use the “Full Body” pose option in the style panel. Mentioning clothing items like “white sneakers” can also encourage full-body framing.
Glasses or key features are missing
Retrain your avatar using only images that include the desired feature (e.g., glasses, hairstyle). Consistency in training images improves output reliability.
Voice sounds flat or unnatural
Adjust ElevenLabs settings such as "style exaggeration" and "stability" in Studio. These sliders affect energy, clarity, and delivery.
Avatar looks different in new Looks
Use the retraining method: generate 50+ Looks, select the most consistent 10–20, and retrain your avatar. Repeat to improve identity retention.
Product doesn't appear correctly in image
Upload the product to HeyGen using our Product Placement feature, give it a clear name (e.g., “green mug”), and reference that name in your prompt. Keep instructions simple and direct.
Motion loops awkwardly in long videos
For the best results, consider limitingLimit custom motion you prompt to short clips (under 10s). Use default motion (i.e., no text prompt) for longer scenes, and intercut with custom gestures for more natural pacing.
Style looks too cartoonish
Start by choosing realistic style when creating your original avatar. If you want only slight animation, avoid selecting “animated” style. Instead, prompt for “subtle animation” or “stylized realism.”
Final reminder
Prompting takes practice. Don’t worry if it takes a few tries — that’s part of the process. Keep your prompts simple, try different variations, and trust your eye.