Kling 4.0 and Seedance 2.5 split independent testers
SLOP TV ran no tests of its own: eight outside testers disagree on which model wins, and the release version each one tested decides most of the outcome.

Key takeaways
- The Paper's eight-test comparison found Seedance 2.5 strongest on complex instruction following and emotion but about three times the price of Kling 4.0: RMB 25 against RMB 7.5 for a 720p 15-second clip, published November 13, 2026.
- Curious Refuge's six side-by-side tests billed $24.00 for one desert sequence on standard Seedance 2.5 against $8.40 on Kling 4.0 at 1080p in Magnific.
- Kingy's 30-attempt ledger recorded Veo 3.1 Lite at 10 of 10 outputs usable for $1.20, against 3 of 10 for Seedance 2.0 Fast at $3.6666 and 4 of 10 for Kling 3.0 Standard at $4.20.
- LLM Review Pro scored four prompts blind on November 6, 2026, updated November 10, and split the win: Veo 3.1 on polish and audio, Kling 3.0 on human faces, Runway Gen-4.5 on control and editing.
Eight rounds of independent hands-on testing published in the past three weeks agree that the leading AI video models are close and disagree about which one is best, with the exact release each tester ran deciding most of the result.
SLOP TV ran none of these tests and owns none of the findings. Every verdict and number below belongs to the outlet that published it, and the version each one named stays with the finding.
What the testers ran
Shanghai's The Paper, publishing in Chinese on November 13, 2026, ran eight tests across Seedance 2.5 (ByteDance), Kling 4.0 (Kuaishou), Wan 3.0 (Alibaba) and Vidu Q4 Preview (Shengshu), scoring instruction following, material fidelity, speed and retries.
Curious Refuge ran six side-by-side tests with Kling 4.0 and Seedance 2.5, from a desert action sequence to a talking monkey driving a New York cab.
LLM Review Pro, in a review by Sarah Chen published November 6, 2026 and updated November 10, 2026, scored four prompts blind at 720p and eight seconds, text to video, three generations each, best of three, across Veo 3.1, Kling 3.0 and Runway Gen-4.5.
Kingy's AI Video Test Lab kept 30 charged attempts, 10 each for Veo 3.1 Lite, Seedance 2.0 Fast and Kling 3.0 Standard, on one controlled prompt at four seconds, 720p, with no audio, and left the failures, the cost and the latency in the ledger.
Four more tests fill gaps. Hermoso scored 45 clips from five models on November 1, 2026, CellCog ran ten reference-to-video tests read on October 29, 2026, Fuser ran four same-prompt tests on October 23, 2026, and DER added six prompts and their credit costs on November 4, 2026.
The version is load-bearing. Veo 3.1 is not Veo 3.1 Lite. Kling 3.0 is not Kling 4.0. Seedance 2.0 Fast is not Seedance 2.5. A test that names the release is testing a different product from one that names only the family.
Where the testers agree
Two findings repeat.
Seedance 2.5 leads on realism, texture and complex instruction following. The Paper found it strongest on long and complex prompts, on character realism and on emotion, and called it the best of the four overall. CellCog picked Seedance 2.5 in all ten of its tests, citing texture, physics and a less plastic look. Curious Refuge preferred Seedance across its most demanding action, camera and realism tests.
Kling 4.0 is the cheaper and steadier option, and every tester that priced both agreed on the size of the gap. The Paper billed the same 720p 15-second clip on Kling 4.0 at about a third of Seedance 2.5's price, CellCog read Kling 4.0 at $0.157 a second at 720p against about $0.473 for Seedance 2.5, and Fuser put it at about a third of Seedance 2.5 per second. The itemized figures are below.
CellCog's own verdict still went to Seedance 2.5, which it picked in all ten tests and scored 408 against 401 out of 700 on a stricter frame-by-frame review.
Where they split
The best-model claim depends on the version and on the platform.
The Paper's conclusion is Seedance 2.5 for quality, with Kling 4.0 usually in its top two, and Wan 3.0 the only model to reproduce a supplied parasol, a magnet and a digital scale one to one. Wan 3.0 was also the worst on physics there: its Newton's cradle came apart from the first second, and its slap became a punch. Vidu Q4 Preview ran away with the acting but needed the most rerolls, and a spin of pasta straightened itself out between shots.
LLM Review Pro split its four prompts three ways. Veo 3.1 won polish and audio, Kling 3.0 won realistic humans in the dialogue and crowd shots, and Runway Gen-4.5 won control and editing. All three produced garbled small text on the text-overlay prompt.
Hermoso, scoring 45 clips blind, put Kling 4.0 and Wan 3.0 level at the top at 3.82 and 3.85 out of 5, and Veo 3.1 last at 3.04 at the highest price of the five.
Kingy's ledger went to Veo 3.1 Lite, which returned 10 of 10 usable outputs and the lowest cost per usable clip, and Kingy declined to name an overall winner. It tested neither Runway nor Wan, which it lists as untested adapters rather than losers.
Fuser's four tests split by job. Kling 4.0 followed direction more closely and kept two fighters' bodies apart at speed, while Seedance 2.5 made the more atmospheric long take but lost the man poling the boat. Fuser also reported that Seedance 2.5 declined its two generated portrait references, saying they might show real people, even though both were invented.
What the tests cost
Rates below come from the testers' own platforms and pages, dated where they dated them.
- The Paper, on Chinese credit prices: RMB 25 for a 720p 15-second clip on Seedance 2.5, RMB 7.5 on Kling 4.0, RMB 7.76 on Vidu Q4 Preview (90 credits at 800 credits per RMB 69) and RMB 6.75 on Wan 3.0 (67.5 credits at 100 credits per RMB 10).
- Curious Refuge, in Magnific: $24.00 for Seedance 2.5 standard and $8.40 for Kling 4.0 at 1080p, and $6.00 for Seedance 2.5 Draft against $5.70 for Kling 4.0 at 720p.
- CellCog, on fal list price read October 29, 2026: Kling 4.0 at $0.157 a second at 720p and $0.235 at 1080p, Seedance 2.5 at about $0.473 and $1.164.
- Hermoso, at the makers' published prices: a six-second 720p clip at $0.60 on Wan 3.0, $0.94 on Kling 4.0, about $1.36 on Seedance 2.5 and $2.40 on Veo 3.1.
- Kingy, historical charges: $1.20 for ten Veo 3.1 Lite attempts, $3.6666 for ten Seedance 2.0 Fast attempts and $4.20 for ten Kling 3.0 Standard attempts, with $9.0666 in total provider charges.
On speed, The Paper timed a ten-second clip at about two minutes on Kling 4.0 and about ten on Seedance 2.5, excluding queue time, and Kingy logged mean latency of 55.3 seconds for Veo 3.1 Lite, 189.3 for Kling 3.0 Standard and 332.9 for Seedance 2.0 Fast.
| Tester | Tested | Verdict in brief | Cost they reported |
|---|---|---|---|
| The Paper | Seedance 2.5, Kling 4.0, Vidu Q4 Preview, Wan 3.0 | Seedance best quality, Kling best value, Wan best materials | RMB 25 against RMB 7.5 for a 720p 15s clip |
| Curious Refuge | Kling 4.0, Seedance 2.5 | Seedance on most tests, Kling on the anime style match | $24.00 against $8.40, 1080p desert sequence |
| LLM Review Pro | Veo 3.1, Kling 3.0, Runway Gen-4.5 | Veo polish and audio, Kling humans, Runway control | not reported per usable clip |
| Kingy | Veo 3.1 Lite, Seedance 2.0 Fast, Kling 3.0 Standard | Veo led this narrow screen | $0.12 against $1.05 and $1.2222 per usable clip |
| Hermoso | Kling 4.0, Wan 3.0, Seedance 2.5, MiniMax H3, Veo 3.1 | Kling and Wan level on top | $0.60 to $2.40 for a 6s clip |
| CellCog | Seedance 2.5, Kling 4.0 | Seedance on all ten picks | $0.473 against $0.157 per second |
Who each test suits, and who paid for it
A review is worth what its reviewer is. Hermoso sells access to all five models it tested and names Seedance 2.5 as its default, and its two judges were AI models reading frames and transcripts, not people. CellCog's video agent runs on Seedance 2.5, which it declares in the post. Fuser hosts both models as nodes and sells access to them. DER runs both models and bills in its own credits, which it declares up front. Curious Refuge closes by inviting readers into its filmmaking community. Kingy says its screen was human-confirmed but AI-assisted, not independent adjudication, and it ran the test through OpenRouter on existing account credit. The Paper tested these models in Chinese, which matters before comparing its realism findings with an English-language test.
What the roundup leaves open: no tester here ran every model, and the version each one named is the version its numbers hold for. A new release makes any line above a snapshot.
Every test in this roundup is linked above and free to read, and each names the model version it ran, so the fastest way to check a claim is to open the tester's own page and look at the number next to the version you would actually generate on.
Sources
- thepaper.cn - The Paper's own eight-test comparison of Seedance 2.5, Kling 4.0, Vidu Q4 Preview and Wan 3.0, November 13, 2026
- curiousrefuge.com - Curious Refuge's six side-by-side tests of Kling 4.0 and Seedance 2.5 and the Magnific cost figures
- llmreviewpro.com - LLM Review Pro's blind shootout of Veo 3.1, Kling 3.0 and Runway Gen-4.5, published November 6, 2026, updated November 10, 2026
- kingy.ai - Kingy's 30-attempt controlled ledger for Veo 3.1 Lite, Seedance 2.0 Fast and Kling 3.0 Standard, with cost, latency and usability
- hermoso.ai - Hermoso's 45-clip blind scoring of five models, November 1, 2026, and its published-price costs
- cellcog.ai - CellCog's ten reference-to-video tests of Seedance 2.5 against Kling 4.0 and its fal list-price read of October 29, 2026
- fuser.studio - Fuser's four same-prompt tests of Kling 4.0 and Seedance 2.5, October 23, 2026
- der.run - DER's six-prompt comparison and credit costs, tested November 4, 2026