Kling 4.0 Guide: 30s Clips, Prompts, Price and Comparison
Kling 4.0 explained: 3-30 second clips, 720p/1080p, @image and keyframe references, copy-ready prompts, VdoBloom credit prices plus Veo 3.1 comparison.
Kling 4.0 is Kuaishou’s video model, officially launched on 15 October 2026. It makes clips of 3 to 30 seconds in a single pass, at 720p or 1080p through the API. It can also take up to 10 reference images, 5 reference videos and up to 10 keyframes. On VdoBloom you can use it today for text-to-video and image-to-video.
Updated October 2026.
Disclosure: VdoBloom is our platform.
What Kling 4.0 is
Kling 4.0 follows Kling 3.0. Its biggest change is length: one generation can run up to 30 seconds, so you don’t have to stitch two or three short clips together for a longer shot.
The model works in three main ways. You can describe a scene in text, start from an image, or steer the clip with tagged references and keyframes. You call those references in your prompt by name, for example @image_1 or @keyframe_2.
If you’re still deciding between models, our roundup of the best AI video models in 2026 gives the wider picture.
Kling 4.0 specs at a glance
| Spec | Kling 4.0 (model capability) |
|---|---|
| Launch | 15 October 2026 (Kuaishou press release) |
| Clip length | 3 to 30 seconds, single pass |
| Resolution via API | 720p, 1080p |
| Aspect ratios (text prompts) | 16:9, 9:16, 1:1, 21:9 |
| Prompt length | Up to 8,000 characters |
| Reference images | Up to 10 (@image_1, @image_2 …) |
| Reference videos | Up to 5 (@video_1 …), each 2 to 30 s, 30 s combined |
| Elements and voices | Up to 7 elements (subjects) and 3 voices |
| Total assets | Maximum 15 per request |
| Keyframes | @first_frame, up to 8 @keyframe_N, @last_frame |
| Native audio | Generated in reference mode whenever no reference video is supplied |
Kling’s own app is reported to offer 4K output. The API stops at 1080p, and that’s the top resolution on VdoBloom.
How the @ reference system works
In reference-to-video mode, every file you upload gets a tag. Images become @image_1 to @image_10, and videos become @video_1 to @video_5. You then write those tags into the prompt where you want each asset used.
One rule matters more than the others: every @mention must match an asset you actually supplied. If your prompt says @image_3 and you only uploaded two images, the request is rejected. Check your tags before you submit.
A reference video can be up to 30 seconds long, and all reference videos together can’t pass 30 seconds. Using a reference video to rebuild a scene with new characters, sometimes described as “video recreation”, is the same reference-to-video mode, not a separate feature.
Keyframes: the duration-to-frame cap
Keyframe-to-video lets you pin the start (@first_frame), the end (@last_frame) and up to 8 points in between (@keyframe_1 to @keyframe_8). The finished video takes the aspect ratio of the first frame.
Most guides skip the detail that the total number of frames you can supply depends on the clip length. Short clips accept fewer frames:
| Clip duration | Max frames (first + middle + last) | Middle keyframes possible |
|---|---|---|
| 3 s | 4 | 2 |
| 4 s | 5 | 3 |
| 5 s | 6 | 4 |
| 6 s | 7 | 5 |
| 7 s | 8 | 6 |
| 8 s | 9 | 7 |
| 9 s or longer | 10 | 8 |
The middle column assumes you use both a first and a last frame. In practice that works out to roughly one frame per second for short clips. If you need eight middle beats, plan for a clip of at least 9 seconds.
Copy-ready Kling 4.0 prompt templates
These are templates to adapt, not finished results. Replace the bracketed parts with your own details. Templates 1 to 4 work with plain text or image prompts. Templates 5 to 8 show the @ syntax as model-level examples.
1. Single 30-second continuous shot
“One continuous handheld shot, 30 seconds. A [subject] walks through [location] at [time of day]. The camera follows from behind, then circles to the front as they stop at [landmark]. Ambient sound of [sounds]. Natural colour, no cuts.”
2. Product reveal
“Slow dolly-in on a [product] resting on [surface]. Soft side light from the left. The camera rises to a top-down view over 8 seconds. Quiet room tone, a soft click as the lid opens.”
3. Vertical social clip
“9:16 vertical. A [person] looks into the camera and smiles, then turns to point at [object] beside them. Bright daylight, shallow depth of field, upbeat street sound.”
4. Cinematic landscape (21:9)
“21:9 wide. Aerial drift over [landscape] at dawn, mist in the valleys. The camera glides forward for 20 seconds and slowly tilts up to reveal [feature] on the horizon. Wind and distant birds.”
5. Character from reference images
“The woman from @image_1 wears the jacket from @image_2. She walks through the market in @image_3, stops at a fruit stall and laughs. Medium tracking shot.”
6. Motion from a reference video
“Follow the camera movement and pacing of @video_1. Replace the dancer with the character in @image_1. Keep the setting of @video_1 but change the lighting to warm sunset.”
7. First and last frame
“Start on @first_frame: a closed flower bud on a windowsill. Time-lapse as it opens, ending exactly on @last_frame. Static camera.”
8. Multi-beat keyframe story
“10 seconds. @first_frame: a runner at the start line. @keyframe_1: mid-stride on the track. @keyframe_2: crossing the finish tape. @last_frame: arms raised in the crowd. Smooth transitions, one camera move.”
For more phrasing ideas, the Kling 3.0 prompts guide covers camera and lighting language that carries over well.
Settings cheat-sheet
| Goal | Mode | Duration | Resolution | Aspect |
|---|---|---|---|---|
| Quick idea test | Text-to-video | 3 or 5 s | 720p | Any |
| Reels, Shorts, TikTok | Text or image-to-video | 8 to 15 s | 1080p | 9:16 |
| Product clip from a photo | Image-to-video | 5 to 10 s | 1080p | Follows your photo |
| Long single take | Text-to-video | 20 to 30 s | 720p first, then 1080p | 16:9 or 21:9 |
| Square feed post | Text-to-video | 8 to 12 s | 1080p | 1:1 |
A sensible habit is to draft a long prompt at 720p and a short duration, then rerun the version you like at full length and 1080p.
Using Kling 4.0 on VdoBloom
VdoBloom currently offers two Kling 4.0 modes:
- Text-to-video: write a prompt, choose 16:9, 9:16, 1:1 or 21:9, and get native audio with the clip.
- Image-to-video: your photo becomes the first frame, and the output follows the photo’s shape.
Both modes offer durations of 3, 5, 8, 10, 12, 15, 20, 25 and 30 seconds, at 720p or 1080p. The multi-reference (@image, @video, elements, voices) and middle or last keyframe features described above are model capabilities. They aren’t part of VdoBloom’s Kling 4.0 tools at the moment.
Kling 4.0 price on VdoBloom
Prices are the same for text-to-video and image-to-video. One credit is $0.05 at the Lite rate ($15 for 300 credits).
| Duration | 720p (11.4 credits/s) | 1080p (17 credits/s) |
|---|---|---|
| 3 s | 35 | 51 |
| 5 s | 57 | 85 |
| 8 s | 92 | 136 |
| 10 s | 114 | 170 |
| 12 s | 137 | 204 |
| 15 s | 171 | 255 |
| 20 s | 228 | 340 |
| 25 s | 285 | 425 |
| 30 s | 342 | 510 |
See the pricing page for credit packs.
Kling 4.0 vs Kling 3.0
The clearest difference is length. Kling 3.0 Omni on VdoBloom runs 3 to 15 seconds, while Kling 4.0 doubles that to 30 seconds in one pass. Kling 3.0 Omni on VdoBloom also offers a 4K option, which Kling 4.0 does not have here (it tops out at 1080p).
Choose Kling 4.0 when you need a longer continuous shot. Choose Kling 3.0 when a short clip at 4K matters more than length.
Kling 4.0 vs Veo 3.1 and Seedance 2.5
Veo 3.1 is built around short clips with sound. If your shot fits in a few seconds and you want Google’s model, it remains a strong pick. Kling 4.0 is the better fit when the shot itself needs to last 20 or 30 seconds. Our Kling 3.0 vs Veo 3.1 comparison covers how the two families handle motion and prompts.
Seedance 2.5 also reaches 30 seconds, so length alone won’t decide between them. Try the same prompt on both and keep the result that suits your scene. The best AI video generators page lists the other options.
When not to use Kling 4.0
- You need 4K. The API, and VdoBloom, stop at 1080p.
- You only need a 2 to 3 second loop. A long-clip model adds nothing, and a short clip from a lighter model is enough.
- You need a talking presenter reading a script. Dedicated tools such as the AI talking avatar or AI presenter video are designed for that job.
- You need many references in one clip on VdoBloom. VdoBloom’s Kling 4.0 tools currently take a text prompt or one starting image.
FAQ
When did Kling 4.0 come out?
Kuaishou officially launched Kling 4.0 on 15 October 2026, announced in a press release. It followed Kling 3.0 and arrived with longer single-pass clips, a tagged reference system and keyframe control. API access through providers came around the same time, which is how platforms such as VdoBloom offer it.
How long can Kling 4.0 videos be?
A single generation can run from 3 to 30 seconds without stitching. On VdoBloom you can choose 3, 5, 8, 10, 12, 15, 20, 25 or 30 seconds for both text-to-video and image-to-video, at 720p or 1080p. Longer clips cost more credits, since pricing is per second.
Does Kling 4.0 support 4K?
Kling’s own app is reported to offer 4K, but that isn’t confirmed in the API documentation. Through the API the model outputs 720p or 1080p, so 1080p is the highest setting on VdoBloom. If 4K is essential, Kling 3.0 Omni on VdoBloom offers a 4K option for clips up to 15 seconds.
How many references can Kling 4.0 use?
In reference-to-video mode the model accepts up to 10 images, 5 videos, 7 elements and 3 voices, with a total of no more than 15 assets in one request. Reference videos can be 2 to 30 seconds each but 30 seconds combined. These are model capabilities, not current VdoBloom features.
How do @image_1 and @video_1 work?
Each uploaded file gets a numbered tag in upload order: images become @image_1, @image_2 and so on, videos become @video_1 onwards. You place the tags in your prompt to tell the model which asset to use where. Any tag without a matching asset causes the request to be rejected.
Does Kling 4.0 make audio?
Yes. In reference-to-video mode, native audio is always generated when you don’t supply a reference video. On VdoBloom, Kling 4.0 text-to-video includes native audio. Audio in keyframe mode isn’t clearly documented, so don’t count on it there. Describe the sounds you want in the prompt for best results.
How much does Kling 4.0 cost on VdoBloom?
It costs 11.4 credits per second at 720p and 17 credits per second at 1080p, the same for text and image modes. A 5-second 720p clip is 57 credits, and a 30-second 1080p clip is 510 credits. One credit is $0.05 at the Lite rate ($15 for 300 credits).
Kling 4.0 vs Veo 3.1: which should I use?
Use Veo 3.1 for short clips with sound where a few seconds is enough. Use Kling 4.0 when you need one continuous shot of up to 30 seconds, or when you plan to use reference images, videos or keyframes through the API. Running the same prompt on both is the simplest way to decide.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →