Kling Video 4.0 Reference-to-Video

Kling Video 4.0 Reference-to-Video Overview

Kling Video 4.0 Reference-to-Video generates video using a text prompt and optional image, video, Element and voice references. This Kling model belongs to the kling-v4 family and is available through each::labs. It exposes reference roles and reusable identifiers so a request can distinguish visual assets instead of treating every upload as a starting frame. Output framing can be adaptive or selected explicitly.

Start with a required prompt and add the references needed for the scene. Choose duration, resolution and aspect_ratio. Reference arrays are optional, but their combined limits still apply when you use them. The model supports output durations from 3 to 30 seconds at 720p or 1080p. Use it for short creative clips, concept development and application workflows where the requested motion can be described clearly. Review each result for the details that matter to your project before treating it as a final asset.

Capabilities

• Guide video generation with a required prompt plus optional structured references.
• Provide up to ten images with first, middle, last-frame or reference-image roles.
• Provide up to five videos, using feature_video or at most one base_video.
• Reference up to seven existing Elements, including at most three Video Character Elements.
• Supply up to three existing custom or system voice references.
• Choose a 3–30 second output at 720p or 1080p.
• Select adaptive framing or an explicit 21:9, 16:9, 9:16 or 1:1 aspect ratio.

Use Cases for Kling Video 4.0 Reference-to-Video

Creators can build scenes around a known visual subject. For example, "@Character crosses the courtyard while the camera follows at walking speed" states how a referenced character should participate rather than leaving the reference’s purpose implicit.

Marketers can combine a product reference and a setting reference for concept videos. "Place @Product on the table in @Setting, warm window light, slow camera push" separates the two roles. Check that appearance and product details remain suitable for the campaign.

Designers can use image keyframe roles to describe a visual progression while retaining other references for context. Keep the images mutually compatible and explain the intended change between the opening and closing compositions.

Developers can build reusable reference-driven video flows with existing Element IDs. Kling Video 4.0 Reference-to-Video is useful when the application needs more structured guidance than a text prompt alone.

Tips and Tricks

Give each referenced asset a short, unique id and describe its job clearly. Keep the subject, setting and motion instructions separate so the request does not ask two references to define incompatible details.

With an image id of Product, try: "Show @Product on a sunlit desk. Keep its overall appearance consistent while the camera slowly moves closer." With an image id of Character, try: "@Character walks through a quiet courtyard at sunrise, gentle forward tracking shot, one continuous action."

Add only references that contribute useful information. A compact reference set is easier to inspect when a result differs from your intention. Multiple middle_frame or refer_image entries require ids. Keep all supplied reference ids unique across input types and choose output parameters explicitly when you need a repeatable request.

Technical Specifications

• Output: generated video; the integration maps successful results from video_url.
• Duration: integer seconds from 3 through 30; default 10 seconds.
• Resolution: 720p or 1080p; default 1080p.
• Optional inputs: images, videos, elements and voices; prompt is required.
• Images: up to ten JPG/JPEG/PNG files, 30 MB each, at least 300 × 300 pixels; aspect ratio 2:5–5:2.
• Videos: MP4/MOV, 300 MB each, 3–30 seconds, 24–60 fps, dimensions 700–4553 pixels, at most 8,294,400 pixels.
• Framing: adaptive by default, or 21:9, 16:9, 9:16 and 1:1.
• Prompt limits: up to 26,000 English-only or 11,000 non-English-only characters; mixed-language prompts use a weighted limit enforced by Kling.
• Processing: asynchronous; no fixed completion time is published in the current catalog. No configurable codec or output frame rate is exposed.

Things to Be Aware Of

The individual array limits are not a license to fill every array to its maximum. All references together must not exceed 15. Videos plus Video Character Elements must not exceed five, and associated video duration constraints also apply. The provider checks media and Element-dependent rules at runtime. A valid JSON request can therefore still fail when a file is inaccessible or its actual properties exceed the supported limits.

Respect the provider’s language-weighted prompt limit and submit only fields exposed by the input schema. Check the final clip rather than relying on the prompt alone.

Key Considerations

Give each reference a clear role and a unique id. Keep combined reference counts within the limits. Framing defaults to adaptive, with four explicit aspect ratios also available.

Catalog estimates without input video are $0.157 per output second at 720p and $0.235 at 1080p. With input video, estimates are $0.094 per billable second at 720p and $0.141 at 1080p. Final charges use the provider-reported deduction or balance list price, rather than a flat per-request fee. Use only assets you are entitled to submit.

Limitations

References guide generation but do not guarantee exact identity, product geometry, speech delivery or pixel-level reproduction. The model accepts existing Element and voice IDs; this request does not create those assets. At most one base_video is supported. Video media and associated character clips must satisfy the combined 3–30 second duration limit. Output is limited to 720p or 1080p and durations of 3–30 seconds. Check important details before publication.