Video
Kling O1
Video
Available on Ofrai — chat, images, and video in one place with shared credits.
Example prompts
Choose an example to load its prompt and recommended settings into the Kling O1 creator above.
How it works
Create with Kling O1 in three steps
- 01
Choose your inputs
Keep Kling O1 selected, then add a prompt and any supported reference assets.
- 02
Set the output
Choose the aspect ratio, resolution, and available Video options before generating.
- 03
Generate and refine
Review the result, adjust the prompt or settings, and generate another version from the same place.
Kling Video O1 is documented as a reasoning-enhanced model that produces the highest quality output in its line, and that single sentence explains most of its behaviour. It does not invent a shot and then look for material to match it; it resolves the material you give it and renders accordingly, which makes the input lists the real workload of every job.
It uses the official Omni reference structure, so images, one reference video, and elements are supplied as ordered lists and addressed from inside the prompt with numbered placeholders. Quality still comes from the familiar standard and professional modes at 720p and 1080p, and duration runs from three to ten seconds. The costs of this approach are structural rather than cosmetic: nothing is ever filled in for you, and a reference video that misses one of its format limits is the most common reason a job fails.
Reasoning over references is the whole point
Read the interface notes again and the emphasis lands on three phrases: reasoning-enhanced, highest quality, and support for reference video editing and feature reference. That combination describes a model whose differentiator is what it does with supplied material rather than how fast it produces an unconditioned clip. Feature reference and reference video editing both ask the same underlying question — can the model carry traits or motion from an input into a new shot — and the reasoning pass is what makes that transfer deliberate instead of incidental.
In practice this changes how you plan a job: you are not writing a prompt and hoping, you are assembling evidence and then instructing. The flip side is that the quality ceiling is not reachable from text alone on this entry, which is why the same family also ships simpler text-to-video routes for the shots that do not need a subject held in place.
Where it visibly fails, and why
Four failure modes account for nearly all of it, and none of them is resolution. First, nothing is inferred: the system will not insert a start frame and will not add placeholders you did not write, so a prompt that references an index you never supplied fails or misbehaves rather than being repaired.
Second, ordering is your responsibility — the order of each input list must match the order in which the prompt cites its placeholders, and a citation to the third item while that item sits second is a silent correctness bug. Third, the reference video slot accepts one video and enforces hard media limits: a publicly accessible URL, MP4 or MOV only, at least three seconds long, between 720 and 2160 pixels, 24 to 60 frames per second, and no more than 200MB.
Phone footage routinely breaks the resolution floor, the frame rate ceiling or the size cap, and the fix is a transcode, which has to happen before submission. Fourth, duration stops at ten seconds, so a longer scene must be split.
Prompt structure this model responds to
The prompt is required and uses the official Omni placeholder syntax, with numbering that starts at 1. Each image slot, the video slot and each element slot are cited by index, and the documentation's own example is the shape to imitate: ask the person in the cited image to wave toward the camera. That example is instructive precisely because it is almost entirely about action.
The model already has the face, the wardrobe, the lighting and the framing — from the reference — so the prose is only worth spending on what the model cannot already see. Describe the change, not the subject. Keep camera language in its own sentence so that when a render fails you can tell whether the reference or the camera move caused it, and keep index numbers contiguous from 1 so that a later edit that inserts a reference does not silently repoint every citation after it.
Parameters, preparation cost, and billing
The parameter set is small: mode at standard or professional, defaulting to standard, decides 720p or 1080p; duration accepts three through ten seconds and defaults to five; aspect ratio takes the common 16:9, 1:1 and 9:16 with 16:9 as default; and the three reference lists are optional in the sense that they may be omitted, but a job with none of them is just a text-to-video job on a more expensive model. The cost that does not appear in the parameter list is preparation.
Getting a video into the accepted shape — public URL, container, duration, frame size, frame rate, file size — is real work, and it is work you pay for again every time the reference changes. Audio is not among the documented fields for this entry, so treat sound as something this endpoint does not deliver and plan it in another pass rather than expecting it to appear alongside the video.
When to pick a sibling instead
Two situations in particular argue against this model. If the job needs a shot longer than ten seconds, the Turbo entry in the same line accepts every second up to fifteen and needs no reference lists, so it is the simpler instrument. If the deliverable must arrive with sound, the v2.6 entry documents automatic audio generation in its professional mode, while this endpoint's documented fields do not include it.
Between the reference-driven entries, choose O1 when the reference material includes motion — a reference video to edit, or features to carry across shots — because that is what the reasoning pass and the video list exist for; when the requirement is many still references for multi-angle consistency instead, the Vidu Q3 base model is documented as the strongest option for that and accepts up to seven images. The cost difference between these choices is smaller than the rework cost of the wrong one.
Moving an existing reference job onto O1
Coming from any text-driven workflow, the migration is mostly additive: you now supply references instead of describing everything, and the quality ceiling rises, but you inherit three obligations. First, audit every existing prompt for unindexed descriptions — anything that describes an asset you are also uploading is a second, competing description and should be deleted. Second, verify each media file against the reference video limits before submitting, because a rejection at that layer costs a round trip and the fix is not obvious from the error.
Third, re-check your duration assumptions against the ten-second ceiling, and cut scenes that assumed longer takes. One habit from this model is worth keeping permanently: store the reference lists and the prompt together in version control, since the correctness of this endpoint depends on a correspondence between two things that live in different fields.
FAQ
- How many reference videos can I send, and what format do they need?
- One video maximum. It must be a publicly accessible URL in MP4 or MOV, at least three seconds long, between 720 and 2160 pixels, 24 to 60 frames per second, and no larger than 200MB.
- Does the model insert a start frame or placeholders automatically?
- No. The system will not supply a first frame and will not add placeholders you did not write. Every index your prompt cites must exist in the corresponding list, and each list's order must match the order of citation.
- What durations does Video O1 support?
- Three through ten seconds, defaulting to five. If a scene needs a longer continuous take, it has to be split, or rendered on an entry that accepts durations beyond ten seconds.
- What distinguishes this entry from other reference-driven models?
- O1 is documented as reasoning-enhanced for the highest quality output in its line, and it carries reference video editing and feature reference. Those are the reasons to pick it when the reference material includes motion rather than stills alone.
FAQ
- Can I use Kling O1 directly on this page?
- Yes. The creator above is the working Ofrai studio with Kling O1 selected by default.
- Where are my finished results saved?
- Signed-in generations are kept in your Ofrai library so you can download or revisit them.
- Which prompts work best with Kling O1?
- Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
- Can I edit the example prompts before generating?
- Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.
Related models
Compare other Video models that can handle similar creative work to Kling O1.
Related guides
Tools
Commercial use depends on the selected model and model terms.





