Kling AI Image to Video: Start Frames, End Frames and Keyframes

Kling AI turns one still photo into a video clip: you upload the image in Image to Video, describe the motion, and generate. In November 2026 the newest model, Kling 4.0 (released to all users on October 15, 2026), runs image-to-video clips of 3 to 30 seconds and takes a start and end frame or up to 10 keyframe images in one generation ^[1]. Kling 3.0 stops at 15 seconds, but unlike Kling 4.0 it has a published credit rate: 12 credits a second at 1080p with native audio. One image is enough. Free accounts can try it at 720p with a watermark. Client or ad work needs a paid plan.

Choose the input mode before you write a word

Kling accepts images for video in four ways. A single image becomes the first frame, and the model invents what happens next. Start and end frames pin both ends of the clip. Keyframes, on Kling 4.0 only, fix pictures at chosen moments along a timeline of up to 30 seconds. Elements are reference images of a person or product that keep a subject consistent while a frame image sets the scene. The model you pick decides which of these you get.

The Flash row is the trap. It is the faster, cheaper variant of Kling 4.0, yet it does not support start and last frames or multiple keyframes, and Kling recommends the full Kling 4.0 for longer clips, higher resolution or precise frame control ^[2]. Flash suits quick drafts from one photo. Kling 3.0 is the practical middle ground: it handles start and end frames, binds extra elements, and has a public per-second cost.

Get the photo right before it costs a credit

Source images must be JPG, JPEG or PNG, up to 10 MB, and the Omni and O1 guides set a minimum of 300 pixels. That minimum is not a target. Kling's quality guide recommends at least 1024 by 1024 pixels, preferably 2K or higher, with the subject clear and well lit. Every flaw the model has to guess at shows up in motion. Kling also notes that pictures which already imply movement, such as a figure mid-pose or hair near wind, water or moving light, animate more convincingly than flat arrangements.

Frame the photo for the format you need. In Kling 4.0 Keyframes mode the aspect ratio follows the first keyframe once images are uploaded, so a landscape product shot yields a landscape clip. Crop first. If you lack a usable photo, Kling's Image 3.0 Omni produces 2K and 4K stills natively, and a generated image can go straight to Image to Video through the Generate Video action.

For products, start from a real photo: Kling's own e-commerce pages say text to video produces a generic item. Kling says Kling 3.0 keeps text from uploaded images, such as signs, captions and logos, consistent across the clip. Its fashion workflow adds a fair warning, though. Reference images guide the result but do not guarantee that every seam, print or logo survives intact, so check those details frame by frame before anything ships.

Start and end frames: transitions, loops and before-and-after

Start and End Frames takes two images as the first and last frame and generates the transition between them; you turn it on with Add End Frame inside Image to Video. Kling suggests it for transformations, craftsmanship and before-and-after comparisons; raw dough and the finished loaf as the two frames would be one example of ours. Its older guide adds the rule that matters most. The two frames should be similar, because large differences may trigger a shot switch instead of a smooth change.

The same tool builds clean loops. Upload the identical image as both start and end frame, and the clip begins and ends on the same picture, which suits a looping music visual or a website hero. Kling warns that motion in between can be uneven; continuous movement such as drifting smoke or a slow rotation hides that best. Since January 2026 you can also pull a still from an existing video asset and use it as a start or end frame, which helps when one shot must continue another.

Elements add consistency on top of frames. In Kling 3.0, after you upload a start frame or start and end frames, you can bind up to three additional elements, and each must appear in the reference frames. A multi-image element holds two to four pictures. VIDEO O1 has one quirk. It cannot generate from an end frame alone, and you choose any length from 3 to 10 seconds.

Keyframes in Kling 4.0: a storyboard in one generation

Kling 4.0 splits creation into two entry points, Omni Reference and Keyframes, and they cannot be combined in one task. Kling's advice is to choose Keyframes when visuals must be locked at specific moments, and Omni Reference to mix several references or explore ideas. Keyframes mode accepts first, last, start-and-last and middle frames anywhere on the timeline, up to 10 images. Without keyframes, the model decides order and timing itself. With them, you hold the beats.

The prompt changes shape in this mode. Kling's guide maps a short description of each keyframe to a time range on the timeline, a format built for projects that already have a storyboard; its sample runs 15 seconds. Omni Reference works differently: images are auto-numbered in upload order as @Image1, @Image2 and so on, and the prompt should state what each one is for, with up to 10 images among 15 assets in total.

Thirty seconds is the ceiling for a single Kling 4.0 generation. Video extension for Kling 4.0, forward or backward up to 2 minutes in total, is listed as coming soon in November 2026, as is 10-bit HDR output. Older models already extend clips by 4 to 5 seconds per pass, up to about 3 minutes, so a longer piece is either built there or assembled in an editor.

The motion prompt: one movement, explicit camera

In image to video the picture already supplies the scene, so the prompt describes only what changes. Kling's formula is Subject plus Movement, then Background plus Movement if the setting should move too. Its sharper rule is to name one exact physical motion and state the camera behavior outright, adding a camera move only after a static-camera version works ^[3]. Keep one main camera move per shot.

Two failure modes recur in Kling's guides. Descriptions that stray far from the image may cause a sudden camera cut, and complex physics, such as a bouncing ball, remain hard to render. The built-in DeepSeek-R1 prompt generator has a gap here: for image to video its suggestions describe the scene rather than the motion, so you add the motion line by hand. Kling states that a specific prompt improves the odds without guaranteeing a result. Our Kling AI prompt guide covers the full formulas.

What an image-to-video clip costs in credits

Kling publishes per-second rates for Kling 3.0 and 3.0 Omni without a video input: 6 credits at 720p without native audio, 9 with it; 8 at 1080p without audio, 12 with it; and 30 for native 4K. A 5-second 1080p clip with audio costs 60 credits, a 10-second one 120, and 10 seconds of native 4K costs 300. Standard is $8.80 a month for 660 credits. By our arithmetic, 660 credits cover 82 seconds of Kling 3.0 at 1080p without audio, or 55 seconds with native audio.

Kling 4.0 is different. Kling does not publish a per-second credit rate for Kling 4.0 on its plans or guides; the cost appears on the Generate button before you confirm. Read it there each time, because settings change it. Budget for retries, too. In Kling's own image-to-video test each attempt took 5 to 10 minutes, and its guides advise several attempts plus a check of hands, text and product details.

Paid plans return up to four videos per generation, against one on the free Basic tier, which helps when you want options from a single photo. Credits spent on failed generations come back as credits, but a clip that renders and simply disappoints is paid for. Monthly plan credits expire after 31 days without rollover; our Kling AI pricing breakdown covers every plan. Free use stays watermarked and closed to commercial use, as our guide to whether Kling AI is free explains.

The bottom line

Kling's image to video rewards preparation more than prompting. Pick the mode first: one photo for a quick move, start and end frames for a transformation or loop, Kling 4.0 keyframes when the beats of a 30-second take must land on time. Feed it a sharp, well-lit image cropped to the target ratio, ask for one motion with a stated camera, and read the credit cost on the Generate button before each run. Use Kling 3.0 when a predictable per-second rate matters more than length.