langchain-tanvo

LangChain tools for Tanvo, an AI image, video and music studio. Give an agent the ability to make images (Nano Banana, Seedream, GPT Image, Qwen, Grok Imagine), short videos with sound (Kling, Veo, Seedance, Wan, MiniMax) and songs (Suno V6), or to run one of 80 ready-made photo apps (pet portraits, figurines, old photo restoration, trend videos) on the user's photos.

• Works with LangChain 1.x agents (create_agent) and LangGraph.
• No API key needed to try it: the house image engine runs on Tanvo's free tier.
• Each tool returns a short text for the model (status, cost, file URLs) and the full run as an artifact for your code.
• When a setting is rejected, the error lists what the model does accept, so the agent can correct itself.

pip install langchain-tanvo

Tools

TanvoToolkit().get_tools() returns all six, sharing one client.

Setup

export TANVO_API_KEY="sk_..."   # optional; create one at https://tanvo.ai/settings/apikeys

Without a key the tools use Tanvo's anonymous free tier, which runs the house image engine (studio-image-v1) only. A key runs on your account: every model, the same prices as the website, and every run in your history. Failed runs are refunded automatically.

Use with an agent

from langchain.agents import create_agent
from langchain_tanvo import TanvoToolkit

agent = create_agent("anthropic:claude-sonnet-5-5", tools=TanvoToolkit().get_tools())
result = agent.invoke({"messages": [{"role": "user", "content": "Make my dog a royal portrait: https://example.com/rex.jpg"}]})
print(result["messages"][-1].content)

The agent finds the pet portrait app with tanvo_find_apps, runs its royal look with tanvo_run_app and answers with the image URL.

Use a tool directly

from langchain_tanvo import TanvoImageGeneration

tool = TanvoImageGeneration()
print(tool.invoke({"prompt": "a tiny lighthouse on a rocky island at dusk, film photo", "aspect_ratio": "4:3"}))

# As a tool call, to also get the artifact (the full run)
msg = tool.invoke({"type": "tool_call", "name": tool.name, "id": "1", "args": {"prompt": "…", "model": "nano-banana-2", "resolution": "2K"}})
print(msg.content)            # "Done (image, nano-banana-2, 25 credits). Files: - https://media.tanvo.ai/…"
print(msg.artifact["outputs"])

Video waits for the clip by default (usually 1–8 minutes). For long-running agents, return at once and check later:

from langchain_tanvo import TanvoGetGeneration, TanvoVideoGeneration

tools = [TanvoVideoGeneration(wait=False), TanvoGetGeneration()]   # or TanvoToolkit(video_wait=False)

Configuration

Models, their options and prices: tanvo.ai/developers/api.

Also from Tanvo

• tanvo-python: the Python client these tools are built on
• tanvo-mcp: the same apps and models in Claude, Cursor and other MCP clients
• tanvo-ai-provider: a provider for the Vercel AI SDK

License

MIT