How to Use Kling AI: Models, Prompts, Lip Sync and Credits

Using Kling AI comes down to five moves: open a free account on the web, phone or desktop app, pick a model, give it a prompt or images, check the credit cost shown on the Generate button, then download. In November 2026 Kling's newest video model is Kling 4.0, released to all users on October 15, 2026, and it makes 3 to 30 seconds per generation. For talking characters, either let the model generate speech natively or run the separate Lip Sync tool on a video you already have. The free account is fine for learning. Real work needs a paid plan.

Open an account and pick where you work

Kling runs in the browser at kling.ai, in iPhone (iOS 15 and up) and Android apps, and in a Windows or Mac desktop app that syncs with the web platform. The web app is the most complete place to start: its header lists the tools as skills, including Motion Control, Avatar, Lip Sync, Video to Audio and Restyle. Two jobs are tied to one surface. Recording a character video to create an element works in the app only, while the Team plan and invoices live on the web only.

A free Basic account claims monthly free credits at login, valid 31 days; Kling does not publish the amount. Basic generates at 720p, one video per generation, keeps the Kling watermark, and its outputs may not be used commercially. Every paid personal plan, from Standard at $8.80 a month, unlocks 1080p and 4K generation, watermark removal, commercial use, Professional mode, video extension and up to four videos per generation ^[1]. Learn the interface on Basic. Upgrade before any clip is used commercially, for a client or your own business. The full free-tier rules sit in our guide to Kling's free plan.

Pick the model before you write the prompt

The model sets clip length, resolution and audio, so choose it first. Kling 4.0 is the default for most new work, and Kling 4.0 Flash trades resolution for speed and lower cost. Kling 3.0 and 3.0 Omni stay useful because they are the newest models with published consumer credit rates. VIDEO O1 and VIDEO 2.6 are older; Kling upgraded them to 3.0 Omni and 3.0.

Kling itself recommends 4.0 over Flash for longer clips, higher resolution or precise frame control, so keep Flash for drafts and volume where 720p is acceptable. One caution on 4.0. Kling lists 4K output for it in the user guide, but 10-bit HDR is still listed as coming soon, and so is extension. If you need to lengthen a clip today, generate it with an older model that supports extension.

Turn text or images into a clip

Kling 4.0 has two entry points, and they cannot be combined in one task. Omni Reference is the default and also covers plain text to video: you type a prompt and can attach up to 15 assets, including up to 10 images, up to 5 videos totaling no more than 30 seconds, up to 7 elements and voice profiles ^[2]. Keyframes is the image-only flow, with first, last, start-and-last and middle frames, up to 10 images placed on the timeline. Kling's rule is simple. Use Keyframes when a visual must land at a set moment, Omni Reference when mixing references or exploring ideas.

Uploaded assets are numbered in upload order, as @Image1 or @Video1, and your prompt should say what each one is for. Otherwise the model guesses. Kling recommends opening with one sentence covering subject, scene, action and emotion, then adding only details that matter: camera work, timing, audio and what must stay unchanged. Most scene prompts work best at 50 to 100 words. Swap vague phrases like a premium feel for concrete lighting, materials and composition.

The remaining settings sit under the prompt box: length, aspect ratio (21:9, 16:9, 1:1, 9:16 or Auto on Kling 4.0), resolution and mode. Standard mode generates faster at lower cost; Professional mode gives richer detail at higher cost and needs a paid plan. In Keyframes mode the aspect ratio follows the first keyframe. On Kling 4.0, multi-shot, audio-visual sync and Smart Storyboard are on by default, so the model splits a long clip into shots unless your prompt asks for one continuous take.

For image to video on older models, Add End Frame turns two images into a transition. Keep the two frames similar, because a big difference can trigger a shot switch instead of a smooth move. The image already supplies the scene, so the prompt should describe motion and camera direction. Our Kling prompt guide collects Kling's own formulas.

Keep characters consistent with elements

Elements are reusable subjects, such as a person, product, animal or scene, that you define once and call with the @ symbol. Build one from up to four images taken from different angles, or from a 3 to 8 second video of a single character. A character element can also carry a voice, bound from a 5 to 30 second clip of one person speaking, so the same voice returns in later scenes. Kling 4.0 accepts up to 7 elements per task, no more than 3 of them video-based.

Each plan caps element creation: 30 on Basic, 50 on Standard, 150 on Pro and Premier, 500 on Ultra+, 300 per seat on Team. Downgrade above the cap and existing elements stay usable, but you cannot edit them or add new ones. Build the library around the characters and products you reuse most. One-off subjects do not need one.

Make characters talk: native audio and Lip Sync

Speech has two routes. The first is native audio inside the generation. Kling 4.0 produces two-channel stereo with speech in nine languages: Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French and Hindi. Write the dialogue in the target language inside quotation marks in your prompt. Kling 4.0 accepts only voice profiles as audio input; music and sound effects cannot be uploaded and are generated from the visuals. Kling 3.0 speaks five languages and can give several characters in one scene their own lines.

The second route is the dedicated Lip Sync tool, for a character video you already have. You upload a voiceover or singing track, or type a script for an AI voice whose speed runs from 0.8x to 2x and whose emotion you can adjust ^[3]. It works on realistic, 3D and 2D human characters with the full face clearly visible. Animals are not supported. Audio longer than the video must be trimmed first. Lip Sync is a paid feature billed by video length, and the cost appears before you generate.

Choose by starting point. If the scene does not exist yet, write the dialogue into a Kling 4.0 or 3.0 prompt, ideally with a voice-bound element, since a standalone voice reference does not attach itself to a character. If the clip is finished and only the mouth must match new audio, use Lip Sync. A still portrait that should speak belongs in Avatar instead. Motion Control transfers movement and expressions, not speech. A silent clip can go through Video to Audio for matched sound effects.

Check the cost, extend and download

Credits are deducted when a generation starts, and the exact cost shows on the Generate button before you confirm. Check it every time. Kling does not publish a per-second credit rate for Kling 4.0 on its plans or guides, so the button is the only figure for that model. Kling 3.0 charges 8 credits a second at 1080p without native audio and 12 with it. By our arithmetic, a 10-second clip at those 1080p audio settings costs 120 credits, and Standard's 660 monthly credits cover 55 seconds of it, or 110 seconds of 720p without audio. Monthly credits expire after 31 days and fees are non-refundable once activated; see our pricing breakdown. Failed generations return their credits.

Extension is a paid feature on older models. Auto-Extend continues the clip without a prompt; Customized Extend follows one you write. Each pass adds 4 to 5 seconds with the same model and mode as the source, up to about 3 minutes in total ^[4]. Keep the extension prompt about subjects already on screen, because unrelated text can cause a camera cut. Expect several attempts. For Kling 4.0, extension up to 2 minutes is listed as coming soon in November 2026.

Downloads can be exported as lossless .mov files, and you can pull a frame from an existing video to start the next clip. Before publishing, review text, logos, hands and product details, which Kling itself flags. Kling's terms ask you to disclose Kling-made output when you share it. Free outputs keep the watermark.

The bottom line

Learning Kling AI takes an afternoon on the free account. The skill that pays off is choosing the entry point: Kling 4.0 Omni Reference for mixed references and dialogue, Keyframes for timed visuals, Flash for cheap 720p drafts, and Lip Sync only when a finished human-faced clip needs new audio. Elements keep faces and voices steady across scenes. Move to a paid plan once clips leave your desk, because 1080p, watermark removal and commercial use start there.