Kling 4.0: thirty seconds in one go, $4.71 a ticket at 720p

In short

• On fal, Kling 4.0 costs $0.157 per second at 720p, $0.235 at 1080p. A 30 s clip: $4.71 or $7.05; if you make 10, $47.10 or $70.50 (our math).
• Stereo sound is announced, but the price list has only one price and the API no "audio" setting.
• Seedance 2.5, which also does 30 s in one go on fal: about $13.87 per clip at 720p according to fal's example, versus $4.71; if you make 10, $138.70 versus $47.10.

What came out

Kling 4.0 is Kling AI's video model. Its website presents it as "All-New Kling 4.0", with no price or release notes we could read: everything below comes from fal, which offers it.

Four modes: text, image (first frame and, if you want, last frame), references, editing a video. From 3 to 30 s, 5 by default; 720p or 1080p, 1080p by default; it can be split into shots, each with its own prompt. According to fal, Kling 4.0 generates dialogue, singing, sound effects and ambience in stereo, synced with the picture; still according to fal, Kling has announced, for later, a way to extend a video up to 2 minutes, plus 4K and HDR.

When the sources contradict each other

• The API. fal's presentation page, updated on October 21, says "API coming soon". The technical pages already give the API address and code. We didn't run anything.
• Seedance 2.5 at 1080p. fal writes "up to 720p", then gives a 1080p price of roughly $1.164 per second, as we had noted in our FLUX 3 Max versus Seedance bill (in French).
• Seedance 2.5 at 720p. Roughly $0.473 per second, i.e. $14.19 for 30 s; but fal's example says $13.87. I'm keeping the example.

What actually changes: the bill

The 30 s clip. 30 × 0.157 = $4.71 at 720p; 30 × 0.235 = $7.05 at 1080p. 1080p costs half as much again: $2.34 more per clip, $23.40 if you make 10. And it's the default setting.

With or without sound? One price list, no "audio" line. The API schemas offer prompt, images, shots, duration, resolution, format; nothing for sound. For Seedance 2.5, fal says it plainly: same price with or without. For Kling 4.0, I only have one price, sound included if it's there.

The discount trap. With reference videos, or to edit a video, the per-second price drops to $0.094 (720p) and $0.141 (1080p)… but fal bills input plus output, at least twice the output. 30 s of result: 60 s billed, $5.64 and $8.46, more than the normal price ($4.71 and $7.05). 10 s of result at 720p with 30 s of reference: $3.76 versus $1.57. A per-second discount is not a discount on the bill.

Against the others, on fal. Seedance 2.5 also does 30 s in one go, for about $13.87 at 720p, according to fal's example. The others I looked at stop earlier: two videos and a cut. FLUX 3 Max (20 s at most): $1.28 for 30 s at 720p during its promo, until November 7, then $5.10. MiniMax H3 (15 s at most): $1.80 at 768p; H3 Max, $3.12 at 1080p until October 31, then $4.80. Once the promos end, Kling 4.0 at 720p catches up with FLUX 3 Max: $4.71 versus $5.10; if you make 10, $47.10 versus $51. And in a single take.

And on a rented GPU? Kling 4.0 doesn't run there: these are other models. On 10/26, on a GPU at $2.49/h, we measured a 5 s vertical shot (704 × 1280 to 768 × 1344): $0.019 with LTX 2.5, $0.059 with Wan 2.2 (silent), $0.135 with MiniMax H3. Times six for 30 s: about $0.12, $0.36 and $0.81; according to our clip kit of 10/24, LTX 2.5 at 1088 × 1920 would cost $0.24 (our math). If you make 10 clips: $1.15 with LTX 2.5 versus $47.10 with Kling 4.0 at 720p. Not the same game: GPU time only, model already loaded, no setup or failed takes, shots put end to end, different resolution and quality. Our 3 min 06 clip at 1080p (in French), all in: about $3.40.

Our own tests (Kling 4.0 on Higgsfield)

This article first said Kling 4.0 had not been tested: that was wrong. We ran it on Higgsfield, not on fal, on three films. Higgsfield bills it 1 credit per second, at 1080p, and we pay $0.05 per credit: $0.05 per second for us, versus $0.235 on fal at 1080p.

The house verdict:

• The chase: "catastrophic", in Kling 4.0 as in Kling 3.0 pro. Too many actions per shot, and one take in two can add a cut or invented text.
• The heist: not convincing. A "very, very saturated" image, "insane" contrast, "ultra soft" pacing. Two moments saved: the display case shattering, and the close-up on the robber's face with a dolly-out.
• The same heist with Seedance 2.5: "much better", not perfect. It cost 250 credits, i.e. $12.50, for 2 blocks of 15 s, draft then 1080p.

Other findings from our shoots:

• Sound: it's really there, native (rain, thunder, engines), even though no "audio" setting shows up.
• Extras: ask for three robbers, it adds four or five, unless you write "exactly three".
• Reference face: it pasted the reference portrait as a shot inside the video.

The videos are coming out on our YouTube channel: Kling 4.0 on the chase on 11/05, "Neon Rain" on 11/03, the heist on 11/07, then the comparisons.

What we don't know yet

• The output on fal: not tested there. Our tests went through Higgsfield, in shots of 3 to 15 s, never a single 30 s take.
• The API and the sound: open or not? Sound always there, billed separately one day?
• Failed takes: billed? Not stated. The price at Kling: pages unreadable without a browser. Kling 4.0 Flash (720p, 20 s, early access): no price.

Key terms

• Per-second billing: length of the video produced × the price for its resolution.
• Multi-shot: one call, several shots, each with its own prompt and duration.
• Reference video: a video that guides the acting, the camera, the pacing; on fal, its length is paid for too.

About the source

• Who: Kling AI makes the model; fal offers it and publishes the prices.
• Date: fal page updated on 10/21/2026; fal post on X on 10/23/2026, spotted but not used as proof.
• Links: Kling 4.0 on fal · text to video · image to video · reference to video · Kling AI

Source passages

Pages read on 10/26/2026 between 9:57 and 10:00 UTC.

Show the 15 sources

Ref.	Quote (original language)	URL
S1↑1↑2↑3↑4↑5↑6+8	"Kling 4.0 generates videos from text prompts at 720p or 1080p, with durations from 3 to 30 seconds and support for per-shot prompts."; "Video costs $0.157 per second at 720p and $0.235 per second at 1080p."; "Per-shot prompts for multi-shot generation. The video duration is the sum of the shot durations."; "Video duration in seconds (3-30). Default value: 5"; "Output resolution. Default value: "1080p""; "Endpoint: https://fal.run/kling/v4/text-to-video"	https://fal.ai/models/kling/v4/text-to-video
S2↑1↑2↑3↑4↑5	"Kling 4.0 animates a starting image into a 3 to 30-second video at 720p or 1080p, with optional ending frames and per-shot prompts."; "Video costs $0.157 per second at 720p and $0.235 per second at 1080p."; "Endpoint: https://fal.run/kling/v4/image-to-video"	https://fal.ai/models/kling/v4/image-to-video
S3↑1↑2↑3↑4↑5↑6+4	"Kling 4.0 is an AI video generation model from Kling AI, now on fal in the playground and fal Agent, with API access coming soon."; "Kling 4.0 generates dialogue, singing, sound effects and ambience as two-channel stereo, in sync with the picture."; "Kling has announced video extension up to 2 minutes as coming soon."; "Kling has announced 4K and 10-bit HDR output as coming soon."; "Four Kling 4.0 modes"; "Image to video pins the first frame, or both the first and the last."; "A reference video guides the performance, action, camera movement and pacing of a new AI video."; "All videos on this page were generated with Kling 4.0 and published by Kling AI. Not generated via fal API."; "Kling 4.0 Flash is the fast member of the lineup, built for frequent creation at 720p and up to 20 seconds. It is in early access on fal."; "Last updated: October 21, 2026."	https://fal.ai/kling-4
S4↑1↑2↑3	"With video references or video elements, it costs $0.094 per second at 720p and $0.141 per second at 1080p, billed on the input plus output length (minimum 2× the output)."; "Reference videos (3-30s each, 30s combined with element videos)"; (edit) "Video costs $0.094 per second at 720p and $0.141 per second at 1080p, billed on the input plus output length (minimum 2× the output)."	https://fal.ai/models/kling/v4/reference-to-video
S5↑1↑2↑3	(API schemas) "x-fal-order-properties": ["prompt", "multi_prompt", "duration", "resolution", "aspect_ratio"]; ["prompt", "start_image_url", "end_image_url", "multi_prompt", "duration", "resolution"]	https://fal.ai/api/openapi/queue/openapi.json?endpoint_id=kling/v4/text-to-video
S6↑	"All-New Kling 4.0"; "Audio-Visual Performance for Realistic Results"	https://klingai.com
S7↑1↑2↑3↑4↑5↑6+1	"Dreamina Seedance 2.5 animates a single still into a native 30-second clip at up to 720p"; "for 1080p, roughly $1.164 per second of generated video"; "30s at 720p 16:9648,000~$13.87"; "for 720p, you will be charged roughly $0.4730 per second of generated video"; "The cost of video generation is the same regardless of whether audio is generated or not."; "Supports 4 to 30 seconds"	https://fal.ai/models/bytedance/seedance-2.5/image-to-video
S8↑1↑2	"duration 5-20 seconds (default 5)"; "generate_audio (default true)"; "Video costs $0.0425 per second at 720p and $0.0725 per second at 1080p."; "The discount runs from October 15 to November 7, after which 720p is $0.17/second and 1080p is $0.29/second."	https://fal.ai/models/blackforestlabs/flux-3-max/image-to-video
S9↑1↑2	"Video costs $0.05 per second at 480p, $0.06 per second at 768p, $0.13 per second at 2K and $0.16 per second at 4K."; "Range: 5 to 15"	https://fal.ai/models/minimax/h3/image-to-video
S10↑1↑2	"Video costs $0.0325 per second at 480p, $0.052 per second at 768p, and $0.104 per second at 1080p."; "The discount ends October 31, after which 480p is $0.05/second, 768p is $0.08/second, and 1080p is $0.16/second."; "Range: 0.92 to 15"	https://fal.ai/models/minimax/h3-max/image-to-video
R1↑	Lead, not proof: fal post of 10/23/2026, "Kling 4.0 is now publicly available on fal, with API access open to all customers."	https://x.com/fal/status/2113646807997620396
R2↑1↑2↑3↑4↑5↑6+2	Our Kling 4.0 math: 30 × 0.157 = $4.71; 30 × 0.235 = $7.05; ×10 = $47.10 and $70.50; gap $2.34 ($23.40 for 10); 5 × 0.235 = $1.175; video reference 60 × 0.094 = $5.64, 60 × 0.141 = $8.46; 40 × 0.094 = $3.76 versus 10 × 0.157 = $1.57	— (our math)
R3↑1↑2↑3↑4↑5↑6+1	Our math for other models: Seedance 2.5 ×10 = $138.70; 30 × 0.473 = $14.19 ($0.32 gap with the example); FLUX 3 Max 30 × 0.0425 = $1.275, 30 × 0.17 = $5.10 (×10 = $51), 30 × 0.0725 = $2.175, 30 × 0.29 = $8.70; MiniMax H3 30 × 0.06 = $1.80; H3 Max 30 × 0.104 = $3.12, 30 × 0.16 = $4.80	— (our math)
R4↑1↑2↑3↑4↑5↑6	Video bench of 10/26/2026, RTX PRO 6000 at $2.49/h, medians for a 5 s shot: LTX 2.5 "$0.0192", Wan 2.2 fast "$0.0594" (silent), MiniMax H3 "$0.1349"; studio kit "1088x1920: about 3 min of compute for 15 s (≈ $0.12)"; clip "about $3.40 ($2.97 compute, about $0.33 setup, $0.10 images)"; totals for 30 s and 10 clips: our math (our French records, translated)	https://siliciumrose.fr/articles/clip-3-minutes-3-dollars-40.html
R5	Illustration and charts: Python/PIL script, no generative AI, no reused image	—

Correction of 10/28/2026 at 19:50 UTC: the first version said Kling 4.0 had not been tested by us. That was wrong: we had used it on three films on 10/26 and 10/27. Added the "Our own tests" box, cover redone with a real frame from our heist.