Build lip-synced talking-avatar videos from text or audio with the VisionStory REST API or Python SDK. Discover current models, avatars, and voices, submit an asynchronous job, and download the finished MP4—or automate the same workflow with the CLI, MCP server, or Agent Skill.
Help me set up VisionStory and generate a talking-avatar video.1. Install or update the CLI and Agent Skill:curl -fsSL https://developers.visionstory.ai/cli | bashnpx skills add visionstory-ai/skills --skill visionstory-api2. Run visionstory doctor. If I need a key, send me to https://developers.visionstory.ai/api-keys, wait while I create one, then ask me to run visionstory login in my terminal. Never ask me to paste the key into chat or expose it.3. Verify access with visionstory credits. Stop and help me if it fails.4. Discover current models, avatars, and voices, then use the CLI and visionstory-api skill to create and download the video. Never delete anything unless I explicitly request and confirm it.
New here?
Create an API key, then send this prompt to your coding agent. It will guide you through VisionStory login, verify credits, and create the video without exposing your key.
VisionStory developer tools
Choose the API integration for your stack.
REST, the Python SDK, CLI, MCP server, and Agent Skill all use the same authenticated VisionStory API and resource IDs. Start with one interface, then move the same avatar-video workflow into a backend, CI pipeline, or coding agent.
Create a server-side API key, verify your credit balance, and discover current model, avatar, and voice IDs. Submit one text_script or audio_script to POST /api/v1/video, poll about every five seconds until the job is created, then download the MP4. Completed video URLs are retained for seven days.
create_video.py
from pathlib import Pathfrom visionstory import VisionStoryClient, build_video_payloadclient = VisionStoryClient.from_env()payload = build_video_payload(avatar_id="YOUR_AVATAR_ID",text="Hello from my product.",voice_id="YOUR_VOICE_ID",)video = client.generate_video(payload)client.download(video["video_url"], Path("result.mp4"))
VisionStory model catalog
Choose the model for the output you need.
Start with the job you need to complete, then fetch the current model catalog at runtime. The API response is the source of truth for available IDs, accepted parameters, and output limits.
Use one API key to create talking presenters, synthesize natural speech, reuse custom avatars, voices, and media assets, and generate supporting video clips. Build localized explainers, personalized onboarding, education, sales outreach, or agent-created media without stitching together separate providers.
Direct answers about authentication, requests, asynchronous jobs, SDKs, CLI automation, MCP, and production use.
An AI avatar video API turns a digital presenter plus text or recorded audio into a lip-synced video that an application can generate programmatically. VisionStory accepts authenticated server-side requests, returns a video ID immediately, renders the job asynchronously, and provides an MP4 URL when generation is complete.
Create a VisionStory API key, verify your credits, fetch current model, avatar, and voice IDs, then submit either a text_script or an audio_script to POST /api/v1/video. The API returns a video ID immediately. Poll the job about every five seconds until its status is created, then download the MP4 from video_url.
Use REST when you need language-independent control, the Python SDK when you are building a Python service or notebook, and the CLI for terminal automation, CI, or batch generation. All three interfaces use the same VisionStory resources and asynchronous video workflow.
Yes. Install the VisionStory Agent Skill or connect the VisionStory MCP server. A supported coding agent can verify credits, list current models, avatars, and voices, prepare a valid request, start the video job, monitor it, and download the result while the API key remains in your environment or MCP secret storage.
Yes. Send an audio_script when you want the avatar to speak from recorded audio, or send a text_script with a voice ID when VisionStory should synthesize the speech. Use one script type per request rather than sending both.
Treat video creation as an asynchronous job. Store your client request ID and returned video ID, poll about every five seconds, and stop when the job reaches created or failed. Completed video URLs are retained for seven days, so copy finished media to your own permanent storage when required.
Keep VISIONSTORY_API_KEY in a server environment variable, CI secret, or MCP client secret store and send it through the X-API-Key request header. Never embed the key in browser code, a mobile application bundle, a public repository, screenshots, logs, or an AI chat message.
You can combine avatar video with text to speech, public or cloned voices, reusable image and audio assets, transcription, AI image generation, and frontier video models. These primitives support localized explainers, personalized outreach, automated learning content, and AI-agent media workflows.
Start with one authenticated request
Build your first AI avatar video integration.
Create an API key, verify your credits, and run the five-step talking-avatar quickstart. Start with REST or Python, then add CLI automation, MCP tools, or the Agent Skill as your workflow grows. Query live resource endpoints instead of hard-coding model, avatar, or voice IDs.
