How to Use Kling 4 in Claude
Claude can generate Kling 4 videos through Creative Claw. Add the plugin, connect your account, and ask for Kling 4.0 Omni. Claude handles the model call and brings the finished clip back into the conversation. No separate Kling or Pika API key is needed.
The model is useful for a scene with directed camera movement, recurring characters, product references or dialogue. Its 3 to 30 second range gives a short narrative room for an establishing shot, action and a clear ending. See the Kling 4 model page for the official Kling demo and all current integration limits.
Add Kling 4 to Claude
- Open Creative Claw in Claude’s plugin directory and choose Add.
- Open the plugin’s Connectors tab, connect Creative Claw, and complete sign-in. The setup guide also covers the custom-connector option and client-specific steps.
- Ask Claude to use Kling 4.0 Omni. Include the resolution, length and aspect ratio, then provide a creative brief and any references.
- Request the credit estimate and approve generation. Kling 4 requires a first Creative Claw credit purchase. Your Claude subscription does not include Creative Claw generation credits.
Claude Code
Install the marketplace plugin with:
/plugin marketplace add CreativeClawCo/creative-claw-marketplace
/plugin install creative-claw@creative-claw-marketplace
Complete the MCP sign-in when prompted. If using the hosted MCP connection directly, the endpoint is https://app.creativeclaw.co/mcp. The same Creative Claw account and purchased credits work in Claude, Claude Code and ChatGPT.
Let Claude plan a scene before rendering
For a story, ask Claude to separate planning from generation:
Plan a 15-second Kling 4.0 Omni scene at 720p, 16:9.
Two old friends recognize each other at a small train station at dusk.
Use my attached character photos only as identity references.
Give me three chronological beats with one consistent lighting style.
Keep dialogue to one short English line per person.
Show the prompt and cost estimate. Wait for my approval to generate.
This makes the sequence reviewable before spending credits. A 15-second text/image-based take at 720p is 471 credits. Ask for a shorter five-second take if you want to test one performance or camera idea first.
A prompt with actions, camera and sound
After reviewing the plan, the generation prompt could be:
Use Kling 4.0 Omni. 15 seconds, 720p, 16:9.
Keep the two friends' identities and clothing from the reference photos.
0 to 4s: wide shot of a quiet station platform at dusk. One friend steps
out of a train and looks up. The other waits near a bench.
4 to 10s: a slow handheld approach as they recognize each other,
smile and walk toward one another. Keep the platform layout consistent.
10 to 15s: medium two-shot as they share a brief hug.
The arriving friend says in English, softly: "You waited."
The other replies in English, warmly: "Of course."
Warm practical lights, cool evening sky, natural movement.
Sound: footsteps, distant train rumble and quiet dialogue, no music.
Name the speaker for each line. Specify tone and language, and leave enough time for both speech and movement. Dialogue is generated with the picture, so exact timing and pronunciation still need review.
Kling’s official guide describes improvements to movement, reference consistency and longer storytelling. Those are capabilities to explore, rather than a guarantee that every take will preserve every face, logo or action perfectly.
Give each reference one job
An identity photo should guide the person’s appearance. A literal first frame should fix the opening composition. A driving reference clip should guide the movement or camera. Tell Claude these roles in plain language.
For tool-level control, image_urls become ordered @Image1, @Image2 references and video_urls become @Video1, @Video2. image_url is the literal opening frame, and last_frame_url is the ending frame. Advanced subject elements use their own handles. Ask Claude to inspect the model parameters and retain those handles exactly rather than copying the older Kling @Element1 contract.
A reference request might say: “Keep the actor from @Image1. Borrow only the handheld camera from @Video1. Keep the location and clothing from my written brief.” Video references increase the billing reserve, so estimate with the actual footage attached.
For source edits, name the change and what should stay: “Change only the coat to dark green. Retain identity, trousers, platform, lighting and camera movement.” Editing can change timing or sound; it is a generation rather than a pixel-preserving edit.
Kling 4 pricing in Claude
For text or image inputs, the rate is 31.4 credits per second at 720p or 47 at 1080p, rounded up once per job. Native audio is included.
| Clip length | 720p | 1080p |
|---|---|---|
| 3 seconds | 95 credits | 141 credits |
| 5 seconds | 157 credits | 235 credits |
| 10 seconds | 314 credits | 470 credits |
| 30 seconds | 942 credits | 1410 credits |
Creative Claw prices checked October 23, 2026. Text and image inputs, with native audio included. Credit purchase prices vary by plan, purchase and discounts. Video inputs and edits use different billing; ask for an estimate with your actual references. Kling 4 requires a first credit purchase.
When you supply video, the reserve uses output duration plus the greater of output duration and measured combined input length. Each billed second is 18.8 credits at 720p or 28.2 at 1080p. For example, a five-second output using a twelve-second video input reserves 320 credits at 720p. Video-based subject elements count too. Final credits follow settled usage within the approved reserve.
At ten seconds without video inputs, Kling 4 costs 314 credits at 720p or 470 at 1080p. Seedance 2.5 costs 534 or 1,314; Wan 3.0 costs 130 or 260; H3 Max costs 160 at 768p or 320 at 1080p. Compare the required inputs and operation as well as the price. All model pricing.
Limits that affect your shot plan
- 3 to 30 whole output seconds, at 720p or 1080p. Aspect ratios: 21:9, 16:9, 9:16, 1:1 or adaptive. This route does not expose 4K or HDR.
- 15 combined assets, with overlapping caps of ten image materials, five video inputs and seven subject elements. No more than three subjects can be video-based.
- 30 seconds of combined input video. Specify the useful source interval for a longer clip. Creative Claw can trim before submission. Video-based subjects have a stricter 3 to 8 second limit and require 1080p MP4/MOV footage of a realistic person, up to 200 MB.
- Native audio is always on. No separate audio-file references, seed or negative-prompt field. For exact supplied speech, use a separate speech and lip-sync workflow.
- One prompt for the sequence. Write chronological beats. Do not send the old
multi_promptarray or numeric keyframe timestamps. The route accepts up to 26,000 prompt characters, while the model also has an 8,000-token limit.
The official capability guide lists Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French and Hindi. Hebrew is unverified. For an unverified language, ask Claude to prepare speech separately and choose a compatible lip-sync model.
When to choose another workflow
Use Kling 4 for new general Kling scenes. Keep Kling 3.0 Motion Control Pro as a separate option for transferring a recorded performance from a driving clip. For a film longer than 30 seconds, ask Claude to plan separate shots, check character continuity and assemble the results. For a simple trim or reframe, use a media-editing tool rather than regenerating the footage.
Frequently asked questions
How do I use Kling 4 in Claude?
Add Creative Claw from Claude’s plugin directory, connect it and sign in. Ask Claude to use Kling 4.0 Omni, specify the shot and approve the estimate. Claude submits the job and returns the finished video.
Is there a Kling 4 MCP server?
Creative Claw is a hosted MCP server with Kling 4.0 Omni. Its endpoint is https://app.creativeclaw.co/mcp and it uses OAuth sign-in. You do not need a provider API key.
How much does a 30-second Kling 4 video cost?
With text or image inputs in Creative Claw, 30 seconds costs 942 credits at 720p or 1,410 credits at 1080p, including native audio. Video inputs and edits use different billing.
Does Kling 4 work in Claude Code?
Yes. Install the Creative Claw marketplace plugin in Claude Code and authenticate its MCP connection. Ask for Kling 4.0 Omni by name. The same Creative Claw account and credits work across clients.
Sources checked October 23, 2026: Kling’s official model guide, Pika’s current endpoint contract, and Creative Claw pricing. The demo on the model page is attributed to Kling AI and is not a Creative Claw generation.
Swap a person in an existing video with Kling 4
Ask for “Swap the person in this video using Kling 4.0”, attach your source clip and reference photo or saved Character, and name who should change and who should stay. Creative Claw uses the Swap workflow with three free scene previews before rendering. In Studio, choose Kling 4.0 under Swap version.
Kling Swap supports 4 to 30 whole seconds, 720p (default) or 1080p, and up to seven combined photo or Character references. Longer clips are cut from the start; fractional lengths are shortened without changing speed. It has no 480p draft or later native draft completion. The original soundtrack is restored by default, and silent sources stay silent.
For matching input and output lengths, a 10-second swap is estimated at 376 credits at 720p or 564 at 1080p. A 30-second swap is 1,128 or 1,692 credits, respectively. Ask for the estimate using the actual clip and resolution. Review the previews, approve the price in Review mode, and inspect the final motion, identity and camera before requesting another take.
Ready to try it?
Connect CreativeClaw to Claude and create your first result.
Connect CreativeClawContinue the workflow