Portrait + Script → Talking Avatar

A portrait photo + a script → a lip-synced talking-head clip.

Category: motion · Difficulty: Intermediate · ~Time: 12 min

What you'll make

A talking-head video where a still portrait speaks your script with synced lips and natural head motion. Use it for a spokesperson clip, a course intro, a product explainer, an avatar host, or a faceless-channel presenter. Input is one portrait plus the words you want said.

Models used

🎬 Video (avatar) — Kling avatar (resolve live; e.g. kwaivgi/kling-v2.6-pro/avatar)
🎬 Video (alt) — Hailuo 02 image-to-video (resolve live; e.g. minimax/hailuo-02/i2v-pro)
🎨 Image — Nano Banana Pro (resolve live), only if you need to generate the portrait first
💬 LLM — Claude, to tighten the script to a target duration

Steps

Get a clean portrait — front-facing, well-lit, mouth visible. Need one? Generate it with Nano Banana Pro via atlas_generate_image, or use the Character Reference Sheet recipe.
Upload the portrait — ask the agent to upload it so it becomes a URL → maps to atlas_upload_media.
Tighten the script — "Trim this to ~20 seconds of natural spoken English, conversational tone." → maps to atlas_chat with Claude.
Generate the talking clip — provide the portrait URL + the script (and a voice/audio if the schema asks for one) → maps to atlas_generate_video with Kling avatar.

Tip: ask the agent to atlas_search_docs for the avatar model first — it has specific fields for the audio/voice and the driving image; these vary by model and change over time.
Frame & caption — ask the agent for an ffmpeg crop to 9:16 and burned-in subtitles for silent autoplay.

Prompt starters

Driving portrait: [uploaded URL]. Have the person say, warmly and clearly:
"Hey — welcome back. Today I'll show you the three settings that make the
biggest difference. Let's jump in." Natural head movement and blinks, eyes to
camera, neutral studio background.

Spokesperson read, confident and upbeat: "Introducing the new Atlas — faster,
lighter, and built for teams. Try it free today." Subtle nods on emphasis,
friendly expression.

Tips & variations

For longer monologues or a wider range of head motion, try Hailuo 02 as the alt path.
Keep the source portrait neutral-mouthed and forward-facing — extreme angles or open mouths hurt sync.
Bring your own voice: record/clone the audio elsewhere and pass it in if the model accepts an audio input, rather than relying on text-to-speech.