How to Import Custom Audio and Optimize Portrait Framing in Hedra for Social Media Avatars

Creating talking avatar videos for TikTok, Instagram Reels, and YouTube Shorts requires more than generating a generic talking head. Viewers scroll past stiff, synthetic-looking avatars within two seconds. To produce high-converting organic clips and user-generated content (UGC) ads, creators and marketers need full control over two critical elements: custom audio that delivers an authentic voice, and precise portrait framing that holds viewer attention without colliding with platform interface overlays.

Hedra provides dedicated pipelines—both through an autonomous conversational agent in Spaces and through direct manual controls—to animate still portraits using your own imported voice tracks. Here is how to import custom audio, frame portraits for vertical mobile feeds, and refine finished social video assets.

1. Preparing and Importing Custom Audio Tracks

In Hedra, your audio track acts as the choreographic engine: the model analyzes phonemes, vocal energy, and pauses to generate mouth shapes, eye blinks, and natural head movement. Flawed audio leads directly to unnatural lip synchronization.

Audio Preparation Standards

• Isolate vocal tracks: Ensure voice recordings are clean and dry. Background music, ambient coffee shop noise, or heavy room reverb confuse the lip-sync engine, resulting in erratic mouth fluttering. If you have a mixed recording, strip background audio before importing.
• Keep pauses intentional: Natural speech contains micro-pauses between sentences. The avatar model uses these silent intervals to deliver subtle breathing cycles, eyebrow resets, and natural head nods. Avoid over-compressing pauses unless crafting rapid-fire UGC hooks.
• Pacing and pronunciation: For synthesized voiceovers (such as voice tracks exported from external tools or Hedra's built-in Speech tool), spell out acronyms phonetically and use line breaks to establish clear conversational rhythm.

Two Ways to Import Custom Audio

Hedra accommodates custom audio through two core workflows:

Workflow A: The Agent in Spaces (Prompt-Assisted)

1. Open a Space and initiate a project using either the presenter or ugc Skill.
2. Select Add reference asset and upload your external audio file (such as MP3, WAV, or M4A).
3. Attach your character portrait alongside the audio.
4. Prompt the agent with your delivery requirements (e.g., "Animate this character speaking with an energetic, friendly creator tone matching the uploaded audio. Maintain steady eye contact.").
5. The agent generates candidate takes within the canvas, allowing you to request targeted revisions while retaining approved assets.

Workflow B: Manual Video Tool (Direct Model Control)

1. Navigate to Manual tools → Video and switch the mode to Avatar.
2. Open the live model selector to choose an avatar model (such as Hedra Avatar or specialized presenter models) based on duration limits and resolution capabilities.
3. In the reference panel above the prompt, upload your custom audio track into the audio reference field and your portrait image into the start frame/image reference field.
4. Input concise performance guidance in the prompt box, define your output aspect ratio and resolution, and click Generate.

Duration Tip: While newer avatar models on Hedra accommodate long-form audio up to several minutes, social media creators achieve better creative control by generating clips in 15- to 30-second segments. Shorter generations conserve credits, allow faster iterations on opening hooks, and simplify clip swaps in post-production.

2. Optimizing Portrait Framing for Vertical Social Feeds (9:16)

Framing dictates whether an avatar looks like an authentic mobile creator or an awkward digital cutout. Social media platforms place strict demands on composition due to vertical aspect ratios and persistent on-screen UI clutter.

The Ideal Shot Size: Chest-Up Medium Close-Up

The most reliable framing for talking avatars is the bust shot (chest-up medium close-up):

• Why not extreme close-ups? Tight crops focused strictly on the face leave no margin for natural jaw movement or chin drop during emphasized speech, often producing distortion around the lower edge of the frame.
• Why not full-body or wide shots? In wide shots, facial features occupy too few pixels. Subtle lip movements, teeth definition, and micro-expressions become blurry and lose lip-sync precision on mobile screens.
• The chest-up sweet spot: Including the upper shoulders and collarbone provides the AI model with anatomical anchors. This allows the avatar to tilt, shift weight, and express chest rise during breathing without distorting the surrounding background.

Designing for Mobile UI Safe Zones

Vertical short-form video (9:16) requires strict adherence to screen safe zones:

+-----------------------------------+
|     Top 10-15%: Headroom          |  <- Keep clear of status bars & search
|                                   |
|   +---------------------------+   |
|   |                           |   |
|   |     Eyes & Face Zone      |   |  <- Upper-middle third (focal center)
|   |                           |   |
|   +---------------------------+   |
|                                   |
|       Chest & Shoulders           |  <- Right side: Likes/Share buttons
|                                   |
|   Bottom 20-25%: Safe Zone        |  <- Keep clear of captions & handle
+-----------------------------------+

• Upper-Middle Third Focus: Place the character's eye line roughly one-third down from the top edge. Viewers instinctively lock onto eyes in the upper half of mobile screens.
• Headroom: Leave 10% to 15% vertical headroom above the hair. If the top of the head touches the frame border, upward head tilts during speech will clip through the top margin.
• Bottom Safe Zone (20–25%): Avoid placing the mouth or chin near the lower third. Platform overlays—including captions, creator usernames, audio tags, and comment input fields—will obscure the speaker's lip synchronization and distract the audience.
• Lateral Safe Margins: Keep the character centered horizontally. The right margin of TikTok, Reels, and Shorts houses engagement icons (like, comment, share, bookmark), which can block the avatar's face if placed off-center.

Portrait Source Image Checklist

Before generating video, inspect your source portrait against these parameters:

• Lighting: Balanced, soft front lighting. Strong shadows across the mouth or one half of the face degrade lip tracking.
• Facial Occlusions: No hands touching the chin, no microphones blocking the lips, and no tinted sunglasses obscuring eye gaze.
• Camera Angle: Straight-on, eye-level orientation. Severe low angles or high angles cause perspective anomalies when the model animates head turns.
• Image Clarity: Crisp, high-resolution source image (PNG or JPG). Blurry inputs compound into generative visual artifacts.

3. Maintaining Consistent Personas with Hedra Elements

For marketing campaigns, brand channels, and ongoing episodic social content, maintaining character consistency across dozens of videos is essential. Regenerating a new face for every prompt destroys brand recognition.

Hedra solves this with Elements:

1. Go to Library → Elements and click Create element.
2. Upload up to 10 clear reference images of your chosen creator or avatar across consistent lighting and angles.
3. Label the element as a Character and assign a unique tag, such as @BrandSpokesperson.
4. In future Space briefs or manual video prompts, simply reference @BrandSpokesperson alongside your newly imported audio track.

Using Elements ensures consistent facial structure, hair color, skin tone, and styling across multi-part social series, UGC ad variants, and localization campaigns.

4. Assembling and Polishing in Composer

In professional social media marketing, the generated avatar clip is rarely the finished asset. To maximize retention, raw talking-head takes should be edited into a multi-layered composition.

Hedra’s built-in timeline editor, Composer (Manual tools → Edit Video), lets you finalize videos without switching software:

1. Import Generated Clips: Bring your generated 9:16 avatar segments into the timeline.
2. Pacing and Trim: Snip any excess frames at the beginning or end of speech to ensure instant hook delivery within the first 1.5 seconds.
3. Layer Background Audio: Add low-volume background music (-18 dB to -24 dB relative to vocal dialogue) on a separate audio track to create energy while preserving vocal clarity.
4. Intercut B-Roll and Product Shots: Alternate between the avatar explaining a pain point and visual demonstrations of the product or software interface. Cutting away from the talking head every 3 to 5 seconds dramatically increases average watch time on short-form platforms.
5. Export: Render the final composition at native resolution (1080p vertical) ready for direct publishing.

Production Summary & Quick Checklist

By pairing clean, dedicated voice tracks with disciplined vertical portrait framing, creators and marketers can turn Hedra from a basic animation novelty into a repeatable, high-performing social video engine.

References

• Create an avatar or presenter video - Hedra
• Create a UGC ad - Hedra
• Give the agent visual context - Hedra
• Speech - Hedra
• Edit Video - Hedra
• Video - Hedra