How to Generate AI Videos with Your Agent (OpenClaw + Veevid)

You can generate AI videos by talking to your agent. No API calls, no dashboards, no manual polling - just say what you want and get a video back.

The Veevid skill connects your OpenClaw agent to Veevid's API, giving it access to 10 video generation models. Your agent handles model selection, cost estimation, generation, and delivery.

What You Need

• An OpenClaw agent
• A Veevid account with credits (sign up here)
• An API key from veevid.ai/settings/api-keys

Step 1: Install the Skill

One command:

npx clawhub@latest install veevid

Then save your API key:

mkdir -p ~/.config/veevid
echo "vv_sk_your_key_here" > ~/.config/veevid/api_key

Your agent now knows how to generate videos.

Step 2: Talk to Your Agent

Text to Video

Describe what you want:

You: "Generate a cinematic video of a golden retriever running through sunflowers"

Agent: "I'll use Veo 3.1 standard mode. Cost: 30 credits. You have 451 credits. Proceed?"

You: "Yes"

Agent: "Done! Here's your 8-second video: [video URL]"

What the agent does:

1. Picks the best model for your request
2. Calls /api/quote to get the exact cost
3. Asks you to confirm before spending credits
4. Generates the video and polls until ready
5. Returns the video URL

Image to Video

Send an image and ask your agent to animate it:

You: "Turn this product photo into a video" (attach image)

Agent: finds your image in chat history "Is this the right image?" quotes the message

You: "Yes, use Kling 3.0, 10 seconds"

Agent: "Kling 3.0 Standard, 10 seconds with audio. Cost: 250 credits. You have 400 credits. Proceed?"

You: "Go ahead"

Agent: "Video ready! [video URL]"

The agent handles image upload, model configuration, and polling - you just talk.

On a Budget

You: "Make a quick test video, keep it cheap"

Agent: "One of the cheapest options is Grok Imagine at 480p for 6 seconds - 15 credits. Want me to use that?"

The 10 Models

Your agent can access all of these:

Model	Best For	Credits
Veo 3.1	Budget-friendly, native audio	15-285 credits
Grok Imagine	Fastest generation, native audio	15-135 credits
Seedance 2.0	Multimodal references, native audio, up to 4K	30-2800 credits
Seedance 2.5	Longest single takes (30s), native audio	112-5700 credits
Kling 3.0	Multi-shot, audio, 3-15s	48-495 credits
LTX 2.3	Up to 4K (2160p), audio	72-1280 credits
Kling 2.6	Cinematic, multi-shot, audio	70-280 credits
Gemini Omni	Native audio, up to 4K, up to 7 input images	45-180 credits
Wan 2.7	Start/end frames, image and video references	40-900 credits
Wan 2.6	Video-to-video, multi-shot, audio	100-450 credits

You can let the agent choose, or specify:

"Use Kling 3.0 for this one"

"Give me the cheapest option"

"I need a 20-second video, what can do that?"

Advanced: Multi-Shot and References

Kling 3.0 Multi-Shot

Kling 3.0 supports multi-shot sequences - multiple prompts with different durations in one video:

You: "Make a multi-shot product ad: first 3 seconds close-up of the phone, then 5 seconds of someone using it, then 2 seconds logo reveal"

The agent constructs the multi_prompt array and handles it for you.

Seedance 2.0 Multimodal References

Seedance 2.0 takes up to 9 images, 3 video clips and 3 audio clips as references in one generation:

You: "Make a 10-second video of this character dancing to this song, copying the moves from this clip" (attach a photo, a dance clip and an audio file)

Start + End Frames (Kling 3.0)

Send two images for precise control:

You: "Use these two images as start and end frames for a 10-second transition" (attach two images)

How It Works

The Veevid skill teaches your agent this workflow:

1. Parse user intent - pick model + parameters
2. POST /api/quote - get exact credit cost
3. Confirm with user - "250 credits, proceed?"
4. POST /api/generate-video - start generation
5. Poll /api/video-generation/{id}/status - wait
6. Return video URL to user

Your agent reads the skill's SKILL.md and follows these steps. No coding needed.

Use Veevid MCP (Claude Code, Cursor, Claude Desktop)

Not using OpenClaw? Veevid is also a remote MCP server at https://veevid.ai/api/mcp, so other MCP clients can use the same models with the same API key. In Claude Code it's one command:

claude mcp add --transport http veevid https://veevid.ai/api/mcp \
  --header "Authorization: Bearer vv_sk_your_key_here"

Your assistant gets eight tools: list_models, get_credits, get_quote, generate, get_generation, upload_file, create_upload_url and list_creations. The flow is the same as with the skill: quote first, confirm the cost with you, generate, then wait for the result. Generations cost the same credits as on the website (see pricing).

Setup for Cursor and Claude Desktop, what each tool does, and a full quote-to-result example are in the MCP section of the developer docs.

Tips

• Costs are confirmed first - The agent quotes before generating, so no surprises
• Be specific - "10-second vertical product video with Kling 3.0" works better than "make a video"
• Pick the right model - Veo 3.1 for quick and cheap, Kling 3.0 for quality, Seedance 2.5 for long single takes
• Check your balance - Ask "How many credits do I have?"

Get Started

1. Install: npx clawhub@latest install veevid
2. Get your API key: veevid.ai/settings/api-keys
3. Save it: echo "vv_sk_xxx" > ~/.config/veevid/api_key
4. Tell your agent: "Generate a video of a cat playing piano"

One skill, 10 models, no friction.

---

The Veevid skill is on ClawHub. For API docs, see veevid.ai/developers.