Overview
Kling V3 is the recommended default video model. Made by Kuaishou, it offers the best balance of quality, flexibility, and cost. It supports continuous duration from 3-15 seconds (integer only) and both text-to-video and image-to-video generation.
Both variants support first-frame and last-frame keyframe control via
end_image_url.
Quick start
Capabilities
Text-to-video
Image-to-video
Animate a still image with a motion prompt:Keyframe control (first + last frame)
Provide two images to control the start and end frames of the video. The model interpolates the motion between them.cURL
image_url (start frame) and the second becomes end_image_url (end frame).
Parameters
string
required
Text description of the video to generate. For image-to-video, describe the desired motion.
number
default:"5"
Video duration in seconds. Integer values from 3 to 15. No decimals —
5.5 will fail.string
default:"16:9"
Output aspect ratio. Common values:
16:9, 9:16, 1:1.array
Array of
{ url: string } objects. One image for image-to-video. Two images for first+last frame keyframe control.Pricing
Standard tier is 33% cheaper with slightly lower quality. Good for drafts and iteration.
Tips
- Image-to-video is significantly better than text-to-video. Always generate a reference image first when possible.
- Keyframe control is powerful for transitions — provide a start and end frame, and Kling interpolates the motion between them.
- Keep prompts concise and motion-focused. Describe what moves and how, not static scene details (those belong in the reference image).
- Standard tier (
kling-v3-standard) is recommended for iteration and testing. Switch to pro for final renders. - Flexible duration is Kling V3’s advantage over other models that only support 5 or 10 seconds.
Related models
Seedance 2
ByteDance’s premium model. Excellent quality, supports video editing.
Sora 2
OpenAI’s video model with remix capability.
Wan 2.5
Budget alternative at 80 credits. Good motion quality.
Kling Legacy
Previous Kling versions (V2.6, V2.5, V2.1, V2).