How to Use Kling 4.0 for Long Cinematic AI Video

Kling 4.0 stretches the Kling line to 30-second clips and adds a proper editing route. On ManifoldGen it runs as four routes behind one API key: text-to-video, image-to-video, reference-to-video, and video edit. All four render at 720p or 1080p, accept 3 to 30 seconds, and support per-shot prompting. This guide covers which route to pick, how to write shot-by-shot prompts, and how pricing works.

Pick the route by input

• Text is the fastest way to explore. Describe subject, action, camera, and light, then iterate on the winner.
• Image to Video locks the first frame. Generate the still with any image model, then animate it when a character or product must look exactly right. An optional end image pins the last frame.
• Reference to Video takes reference images and videos so characters, objects, and motion carry into a new shot. Call them @Image1, @Video1 and so on in the prompt.
• Video Edit rewrites an existing 3 to 30 second clip from a text instruction: change the background, swap wardrobe, shift the weather. Refer to the source as @Video1.

Prompt shot by shot

Longer clips need structure. Write one line per shot with its own camera distance and action, and keep the total within the clip length:

Shot 1 (0-5s): Wide - a detective in a black trench coat crosses a rain-soaked rooftop toward a telescope.
Shot 2 (5-10s): Close-up - her eye at the eyepiece, neon reflecting in the lens.
Shot 3 (10-15s): Medium - she lowers the telescope and turns toward the stairwell door.

One subject, one action, and one camera per shot keeps boundaries clean. For clips past 15 seconds, give each beat a distinct visual change so the model has something to move toward instead of repeating itself.

Resolution, aspect ratio, and cost

• 720p is the drafting resolution. 1080p costs about 50% more per second; spend it on the shots that survive the edit.
• Aspect ratios are 16:9, 9:16, 1:1, and 21:9. Reference to Video also accepts auto, which follows the references. Image to Video and Video Edit follow their source.
• Pricing is per second of output. Reference runs that include a reference video, and every Video Edit run, bill on input plus output length, so they cost about twice the plain per-second rate.
• Kling 4.0 does not generate a separate audio track. Add sound in the Studio timeline.

Troubleshooting

• Shots melt together - shorten each shot description to one action.
• Character drifts - move to Image to Video or Reference to Video with a clean reference image.
• Edit changes too much - name what must stay: "keep the subject and camera move, only change the background".
• Long clip feels slow - split into two 10 to 15 second generations and cut them together.

Open the workflow at /tools/kling-4, /tools/kling-4-image, /tools/kling-4-reference, or /tools/kling-4-edit. The API mirror of each lives at /api/video-generators/kling-4 and friends.