p-video-avatar

Prerequisites

Install and load these skills before generating (skip if already in context via @pruna):

Or install the full suite once: npx skills add PrunaAI/pruna-skills@pruna -y

Follow each skill's Before generating / craft sections — do not restate guide content here.

Agent habit

In the first reply, name `p-video-avatar` in backticks, confirm PRUNA_API_KEY (or stop with signup links from pruna-api), then ask for required inputs. Open intake → generation-diversity clarification intake before the first POST. Multiple talking-head scenes with the same person → redirect to avatar-multi-scene (this skill is one clip only). Draft host motion with Prompt craft (dynamic + faithful) — do not paste skill examples.

Skill boundary

This skill = one p-video-avatar prediction per invocation.

Out of scope (stop and redirect):

• Several host segments with continuity → avatar-multi-scene
• Multi-scene assembly, concat, or parallel scene batches → workflow skills (avatar-multi-scene, narrated-multi-scene, …)
• Silent B-roll / no talking head → p-video
• Motion transfer from a template video → p-video-animate

Prompt craft (dynamic + faithful)

video_prompt (and optional voice_prompt) must be fresh per clip and faithful to the user's host beat. Diversity applies to camera nuance and delivery wording — not to changing who speaks or what they say.

Fidelity check (before pay): the clip must still be the user's speaker, script/audio, and approved host beat. If mouth visibility or pacing drifts from the brief, rewrite.

When NOT to use

Use a different skill instead:

HTTP (curl)

Upload portrait

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: ${PRUNA_API_KEY}" \
  -F "content=@/path/to/portrait.png"

Use urls.get as input.image.

Create (async — recommended)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-video-avatar' \
  -d '{
    "input": {
      "image": "https://api.pruna.ai/v1/files/FILE_ID",
      "voice_script": "Hey — so we shipped something I've wanted for a while.",
      "voice": "Puck (Male)",
      "voice_language": "English (US)",
      "voice_prompt": "Natural conversational tone — relaxed pacing, real pauses.",
      "resolution": "720p",
      "video_prompt": "Medium close-up speaking directly to lens, subtle push-in",
      "negative_prompt": "subtitles, captions, on-screen text, watermark, logo, typography, letters, words",
      "negative_prompt_strength": 0.35
    }
  }'

Poll and download: follow pruna-api.

Complete the random seed ritual from generation-diversity before writing prompts — omit seed unless the user supplied api_seed. Confirm voice_language with the user.

For multiple clips: create all jobs in parallel (async, no Try-Sync), then batch-poll.

Create (sync — quick test only)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-video-avatar' \
  -H 'Try-Sync: true' \
  -d '{
    "input": {
      "image": "https://api.pruna.ai/v1/files/FILE_ID",
      "voice_script": "Hey — so we shipped something I've wanted for a while.",
      "voice": "Puck (Male)",
      "voice_language": "English (US)",
      "voice_prompt": "Natural conversational tone — relaxed pacing, real pauses.",
      "resolution": "720p",
      "video_prompt": "Medium close-up speaking directly to lens, subtle push-in"
    }
  }'

Uploaded narration (audio wins over voice_script)

Generate gemini-3.1-flash-tts → upload to /v1/files. Pass as input.audio with portrait image. If both audio and voice_script are set, audio wins.

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-video-avatar' \
  -d '{
    "input": {
      "image": "https://api.pruna.ai/v1/files/PORTRAIT_ID",
      "audio": "https://api.pruna.ai/v1/files/NARRATION_ID",
      "resolution": "720p",
      "video_prompt": "Medium close-up, natural head motion matching narration"
    }
  }'

Before generating

1. Complete Prerequisites guide reading order (generation-diversity → video-prompting).
2. Ritual seed → draft a dynamic + faithful video_prompt (section above) → confirm image URL, voice_script (or audio), voice / voice_language, voice_prompt, video_prompt, and resolution. Explicit user confirmation before any paid call.
3. Pruna notes: P-API uses snake_case (voice_script, video_prompt, …). Mouth must be visible on the plate. Unique video_prompt per clip — do not reuse one string across a multi-scene reel. Default The person is talking. is quick-test only.

Negative prompt (experimental — suppress on-screen text)

Starter: subtitles, captions, on-screen text, burned-in text, watermark, logo, typography, letters, words. Start strength around 0.3–0.4. See avatar-single-scene for gated host workflows.

Typical next steps

Common follow-ons after this skill: