How to Use the Kling 4.0 Reference to Video API: reference to video integration guide (2026)
Learn how to call the Kling 4.0 Reference to Video API on NamiFusion: parameters, pricing (from ≈$0.785) and runnable copy-paste code — try it in the playground first.
Kling 4.0 Reference to Video is a reference to video model by Kling: Generate 3–30 second videos using image, video and audio references, with native prompt references and 720p or 1080p output. On NamiFusion you can try it in the playground (exact quote shown before you submit), then call it through one unified REST API — about $0.785 per typical run.
TL;DR
· Task type: reference to video, provided by Kling
· Pricing: about $0.785 per typical run ($1 = 100 credits), pay-as-you-go
· Getting started: playground (exact quote before submit) → create an API key → one POST request
· Tunable parameters: 7 (see the table below)
· No platform watermark; commercial use allowed (per the Terms of Service)
What can you build with Kling 4.0 Reference to Video?
Short-form video and ad clips: go from a prompt or a single image to publishable motion content.
Product demos and concept previews: validate storyboards and visual direction cheaply before any shoot.
A social content factory: batch vertical clips and ride trending topics fast.
Remixing existing assets: image-to-video brings static work to life and extends its lifespan.
How do you call this API?
- Sign up on NamiFusion and create a key on the API Keys page (signup bonus credits can only be used on selected models).
- Submit a task with the cURL example below, passing your key as a Bearer token.
- Poll the task status with the returned task_uuid and read the output URLs when completed.
# 1) Submit — returns { "task_uuid": "..." }
curl -X POST "https://www.namifusion.com/api/v1/marketplace/run/kwaivgi/kling-v4.0/reference-to-video" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"input": {
"prompt": "A cinematic portrait, dramatic rim light, shallow depth of field",
"keep_original_sound": false,
"resolution": "720p",
"duration": 5,
"aspect_ratio": "auto"
}
}'
# 2) Poll until status is "completed", then read the output URLs
curl "https://www.namifusion.com/api/v1/marketplace/run/tasks/TASK_UUID" \
-H "Authorization: Bearer YOUR_API_KEY"What parameters does it take?
prompt (Prompt): required
images (Reference images): optional
audios (Reference audio): optional
keep_original_sound (Keep original sound): optional, default false
resolution (Resolution): optional, default "720p"
duration (Duration): optional, default 5
aspect_ratio (Aspect ratio): optional, default "auto"
How do you write prompts for Kling 4.0 Reference to Video?
- Subject first: lead with the actor and action, then add environment, lighting and style ("an astronaut sprinting through a neon street in the rain, cinematic, shallow depth of field").
- Concrete nouns beat stacked adjectives: "85mm portrait lens, golden-hour backlight" works far better than "very beautiful, ultra high quality".
- Change one variable per iteration: pin the seed to reproduce a result, tweak a single phrase at a time so you know what actually moved the output.
- State negatives positively: instead of "no blur", ask for "sharp focus, crisp details".
How much does it cost?
Pay-per-use in credits. About $0.785 per typical run ($1 = 100 credits); the exact price varies with parameters like resolution or duration, and the playground shows the exact quote before you submit.
Kling 4.0 Reference to Video
Create a new 3–30 second video from a prompt with optional image, video and audio references. Output duration is chosen independently of reference-video duration.
Inputs
- prompt: required string.
- images: optional image-URL array, at most 10.
- videos: optional video-URL array, at most 5. Each source must be at least 3 seconds; at most 30 seconds of video is used in total.
- audios: optional audio-URL array, at most 3; each clip must be 5–30 seconds.
- keep_original_sound: boolean, default false. Set true to preserve reference-video audio; requires videos.
- resolution: "720p" (default) or "1080p".
- duration: integer from 3 through 30, default 5.
- aspect_ratio: "auto" (default), "16:9", "9:16", "1:1" or "21:9".
Limits and references
Use at most 15 reference files across images, videos and audios. Longer source footage is trimmed to the 30-second total budget. In prompt, @Image1, @Video1 and @Audio1 refer to the first item in their respective arrays; numbering starts at 1 separately for each media type. Explain the role of each reference.
List price and initial credit charge
| Input video | 720p | 1080p | Billing basis |
|---|---|---|---|
| None | $0.157 | $0.235 | Per output second |
| Present | $0.094 | $0.141 | Per counted second |
Without reference video: Cost = output seconds × resolution rate. A 5-second clip costs $0.785 at 720p or $1.175 at 1080p.
Let D be output seconds and V = ceil(min(total input-video seconds, 30)). With video input, cost = rate × (D + max(D, V)). At 720p, D=5 and an 8.2-second input give V=9 and a cost of $1.316; the initial charge is 131.60 credits.
1 USD = 100 Nami credits. Round the total charge upward to 0.01 credit: credits = ceil(cost_usd × 10000) / 100. These are list prices; no promotional discount is applied.
Images, reference audio, aspect ratio and audio-preservation settings add no separate charge under this formula.
FAQ
Can I try it for free first?+
Signup bonus credits can only be used on selected models, and eligibility may change. The playground shows the exact quote before submission.
Can I use the output commercially? Is there a watermark?+
Yes, outputs are yours to use commercially, and NamiFusion adds no platform watermark — subject to the Terms of Service and applicable law.
What is the content policy?+
Pornographic content, sexualized content involving a minor, unauthorized use of real people, deceptive impersonation, fraud and false endorsements are prohibited. Lawful, consensual, non-explicit adult themes may be permitted; see the Terms of Service.
How long does a Kling 4.0 Reference to Video task take?+
It depends on parameters — images usually take seconds, videos tens of seconds to minutes. The API returns a task_uuid for polling.
What are the resolution and duration limits?+
See the parameter table above — each model lists its available resolution and duration options there and in the playground.