Choose between Kling 4.0 modes by deciding what your video must preserve: an idea alone → text-to-video; a specific look → image-to-video; assets you need to lock → Omni Reference; exact timeline moments → Keyframes. Omni and Keyframes are separate entries and cannot be combined in one generation task. Pick the entry that gives you control over the part of the result that matters most.
Last verified: 2026-10-23
This guide helps you choose a starting point, prepare your inputs, and move to a detailed walkthrough. For the broader workflow, start with how to use Kling AI. For documented capabilities beyond mode selection, see the Kling 4.0 features guide.
How do you choose between Kling 4.0 modes?
Ask one question before writing your prompt: What would make this video wrong even if the motion looked good?
If the answer is “the scene does not communicate my idea,” begin with text. If it is “the subject looks different,” begin with an image. If several supplied assets must guide the result, choose Omni Reference. If a particular visual moment must occur at a planned point, choose Keyframes.
| Mode | Best when | What you provide | Watch out for |
|---|---|---|---|
| Text-to-video | You have an idea but no starting image | A written description of the subject, action, scene, and camera movement | Appearance depends on your description; keep the first concept focused |
| Image-to-video | A subject’s existing look matters | A photo or illustration, plus motion instructions; first/last frame options where appropriate | A reference does not guarantee identical faces throughout |
| Omni Reference | Several assets should guide appearance, movement, or sound | Combined references with clear roles in the prompt | Total and per-type caps apply; cannot share a task with Keyframes |
| Keyframes | Specific visual moments should structure the sequence | Up to 10 keyframe images defining important moments | Inspect the motion between images, especially turns, hands, and scale changes |
Choose based on the input you already have and the control you need. Adding more references is useful only when each one has a clear purpose.
When should you use Kling text to video instead of image to video?
Use Kling text to video when the concept comes before the visual asset. You might want a quiet bookstore at sunrise, a cyclist crossing a bridge, or a close-up of rain hitting a window. A written prompt lets you describe the scene without first preparing a photograph or illustration.
Start with one subject, one action, and one camera instruction:
A cyclist rides slowly across a stone bridge at sunrise. Soft morning light, calm atmosphere, camera tracking alongside the cyclist.
This gives the first attempt a clear purpose. Review whether the action and framing communicate your idea before adding more scene changes.
Use Kling image to video when you already have a visual you want the shot to follow. A character portrait, product photograph, or illustration supplies an appearance reference that text alone would need to describe.
With an image, focus the prompt on what should happen:
The subject turns slightly toward the camera. Keep the movement gentle, with a slow camera push-in.
For recurring characters, reuse the same sharp reference photo and consistent character and lighting descriptions. Keep motion small when facial consistency is the priority. First-and-last-frame options can help with shots whose endpoints need to match your plan.
No setting guarantees an identical face. Review the clip and regenerate weak results using the image-to-video consistent faces guide.
What is Omni Reference for?
Omni Reference is the entry for directing a video with multiple assets. Think “use this asset as guidance”: one image may establish appearance, a video may guide movement, and a voice profile may supply a sound reference.
Its value comes from assigning clear jobs to your inputs. Assets are automatically numbered with labels such as @Image1 and @Video1. State each asset’s role in the prompt rather than leaving the relationship ambiguous.
For example:
Use @Image1 as the subject’s appearance reference. Use @Video1 to guide the movement. Keep the scene focused on that subject.
Treat asset locking as the goal of your instructions, then inspect whether the result follows them.
The verified reference limits are:
| Reference type | Cap |
|---|---|
| All references combined | ≤15 total across images, videos, voice profiles, and elements |
| Images | ≤10 |
| Videos | ≤5; each 3–30 seconds; combined video duration ≤30 seconds |
| Voice profiles | ≤3 |
| Elements | ≤7, including ≤3 video elements |
The combined ceiling still applies when individual categories have unused capacity. Five videos must also fit within the combined video-duration limit.
Use the smallest set that explains your intended result. For asset preparation and prompt structure, continue to the Kling 4.0 Omni Reference guide.
What are Keyframes for?
Keyframes are for planning specific visual moments across a timeline. Instead of asking only for a certain appearance or movement style, you define what the sequence should look like at important points.
Kling 4.0 supports up to 10 keyframe images to guide story progression. These can establish a first frame, a last frame, or a sequence of important moments.
For example, a short scene might need three visual stages:
- A person stands beside a closed door.
- The person turns toward the opening.
- The person steps into the room.
Use Keyframes when those stages are essential to the story. Prepare images that communicate the progression clearly, with particular attention to subject position, framing, and scale.
Then review the transitions. Matching the planned images is only part of the job: hands, turns, and sudden scale changes can make the movement between moments feel inconsistent.
See the Kling 4 Keyframes guide for the detailed workflow.
How is Omni Reference different from Keyframes?
The distinction is asset guidance versus planned visual moments.
Omni Reference answers: “Which supplied assets should guide how this video looks, moves, or sounds?”
Keyframes answer: “What should be visible at these particular moments?”
Choose Omni when the main requirement is carrying asset characteristics into the result. Choose Keyframes when the main requirement is the sequence of visual states.
Omni and Keyframes cannot be combined in one generation task. If your concept needs both kinds of control, first decide which requirement defines the current clip. A simpler clip with one clear priority is easier to evaluate than a prompt with competing requirements.
What is the easiest beginner path?
Begin with one small goal: one subject performing one action in a short scene.
If you have no image, write a text-to-video prompt. If you already have the right subject image, use image-to-video and describe restrained movement. Move to Omni when multiple assets need distinct roles, or to Keyframes when the sequence requires specific visual moments.
Kling 4.0 documents generation durations of 3–30 seconds and resolution options of 720p and 1080p. Pick a duration that fits the action, then review appearance, motion, and framing before expanding the concept.
Follow the beginner Kling AI walkthrough for the overall process. For lip-sync and audio details, see the Kling 4.0 lip-sync guide.
FAQ
Which Kling 4.0 mode should I choose without an image?
Choose text-to-video. Describe the subject, action, setting, and camera movement. Keep the first attempt focused on one scene.
Is image-to-video better for keeping a character consistent?
It provides a concrete appearance reference. Reuse a sharp image and consistent descriptions, keep movement modest, and review the result; identical faces are not guaranteed.
How many references can Omni Reference use?
Omni supports up to 15 combined references. Separate caps apply to images, videos, voice profiles, and elements, including limits on video duration.
Can I combine Omni Reference and Keyframes?
No. They are separate entries and cannot be combined in one generation task. Choose asset guidance or specific timeline moments for that task.
How many keyframe images does Kling 4.0 support?
Kling 4.0 supports up to 10 keyframe images. Use them for important visual moments and inspect the motion connecting those moments.
Sources
- Official user guide: https://kling.ai/quickstart/kling-4-model-user-guide
- Official feature page: https://kling.ai/feature/kling-4-ai-video-generation
- Official model release page: https://kling.ai/dev/model-release/kling-4
- Kuaishou IR launch announcement: https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-officially-launches-next-generation-kling-40-raising/
Start with free credits on sign-up, enough for one video. Then Try Kling AI Today