Each model receives the same input, identical prompt, 5-second target duration, aspect ratio, and one generation attempt per scenario. We publish latency, output, motion, temporal coherence, prompt adherence, failures, and cost per usable result only after both outputs are captured.
This page currently separates catalogue facts from a ready-to-run controlled protocol. The paired benchmark has not been executed, so blank result cells are intentional and no speed or quality winner is claimed.
Protocol published 2026-08-01; controlled outputs pending
Identical inputThe same consented waist-up portrait with face and both hands visible
Identical settings9:16, 5s
Exact prompt sent to both modelsThe subject takes one calm breath, blinks once, then turns their eyes slightly toward camera. Hair moves gently in a light breeze. Keep the same face, age, hands, clothing, background, lighting, and framing. Static camera, natural motion, no new objects.
Result evidence
Kling 3.0
No controlled result published yet
- Catalogue price per generation
- 95 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Wan 3.0 Text to Video
No controlled result published yet
- Catalogue price per generation
- 80 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Identical inputThe same centred product packshot with readable label and clean background
Identical settings1:1, 5s
Exact prompt sent to both modelsA slow 20-degree camera orbit around the product while a narrow highlight travels across the packaging. Preserve the exact product shape, logo, label text, cap, colours, shadows, and background. No morphing, no extra objects, premium studio motion, continuous geometry.
Result evidence
Kling 3.0
No controlled result published yet
- Catalogue price per generation
- 95 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Wan 3.0 Text to Video
No controlled result published yet
- Catalogue price per generation
- 80 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Identical inputThe same full-body character frame in a simple street environment
Identical settings16:9, 5s
Exact prompt sent to both modelsThe character takes two measured steps forward as the coat hem reacts to movement. The camera tracks sideways at walking speed. Preserve identity, body proportions, clothing details, street layout, and light direction. Continuous foot contact, no cuts, no duplicated limbs, no scene change.
Result evidence
Kling 3.0
No controlled result published yet
- Catalogue price per generation
- 95 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Wan 3.0 Text to Video
No controlled result published yet
- Catalogue price per generation
- 80 coins with default settings
- Cost per usable result
- No controlled result published yet
- Observed speed
- No controlled result published yet
- Motion
- Not measured in a controlled side-by-side run yet.
- Temporal coherence
- Not measured in a controlled side-by-side run yet.
- Prompt adherence
- Not measured in a controlled side-by-side run yet.
Failure cases- Not assessed until the controlled output is generated and reviewed.
Recommendation by production task
Test bothPortrait identity
Run the same short portrait prompt in both models and choose by identity drift at the middle and final frames; catalogue specifications cannot answer that question.
Test bothProduct geometry and readable packaging
Use the controlled product scenario and count a result as usable only when shape, logo and label remain intact; the cheaper generation is not necessarily the cheaper usable result.
Test bothFull-body character motion
Compare foot contact, limb continuity and background stability under the identical tracking prompt before committing to a longer or more expensive scene.
Test bothFast first draft under a fixed budget
Start with the lower current catalogue price, but record retries and divide total coins by usable outputs; that is the cost metric that matters for production.