Kling 4.0 vs Higgsfield Genjutsu: does the swap respect height?

Results

Price on KeelRouter

per second. Prices follow our catalog and update when providers change theirs.

Can I Ride dance: a 70 cm pug and a 50 cm kitten replace the two dancers

Replace both people completely, for the entire clip, with exactly two real pets. The pug from <<<image_1>>> replaces the LEFT dancer wearing a beige beanie. Viki the kitten from <<<image_2>>> replaces the RIGHT dancer in the black cap who starts nearest the camera. Both pets STAND UPRIGHT ON THEIR HIND LEGS AND DANCE on their hind paws for the whole clip, never on all fours. The pug's total standing height, from the ground to the top of the head, is exactly 70 cm. Viki's total standing height is exactly 50 cm. They are small animals at ground level in the same parking spaces as the source dancers, not human-sized animals. Viki is visibly shorter than the pug, about five sevenths of its height. Keep the camera fixed in its original position; keep the original houses, cars and pavement at their original scale, so the pets occupy much less of the frame than the humans did. Do not enlarge the pets to fill the human silhouettes and do not zoom in. Each pet keeps its own face, fur colour, markings, ears, body proportions, exact clothes and accessories from its own reference photo. The pug keeps its punk outfit and Viki keeps her cape and clasp. Their forelegs and real furry paws copy the arms and hand gestures; no human hands, human skin or human torsos. No beige beanie, black cap, logo shirt or cross-pattern trousers from the original people. No extra people or animals; no human limbs left behind. Copy the source choreography and timing exactly: the right dancer steps back beside the left, then both bounce on bent knees in sync and swing their bent forelegs and paws back and forth near waist height, lifting their paws as in the original toward the end of these five seconds. Faces and mouths follow the original expressions, without inventing speech. Keep the dusk lighting, pale blue sky, grey houses, parked cars, pavement and camera. Photorealistic phone footage. No added text. The source video is <<<video_1>>>. Edit that video; preserve its exact choreography, timing, setting and original camera perspective.

The same clip with two people, Genjutsu only

To see why the pets came out wrong, we also ran Genjutsu with two people of ordinary height. Their heights look right. Our reading is that Genjutsu fills the outline of each source dancer with the new subject, so whatever goes in comes out about the dancer’s size. It even kept the dancers’ beanie and cap, although the prompt said not to.

Replace the left dancer in the beige beanie with the subject in reference image 1, and the right dancer in the black cap with the subject in reference image 2. Replace both dancers completely for the whole video. The source gives only the choreography, positions, timing and camera. Each subject keeps its own face, fur or hair, markings, age, build, height and proportions, and wears the exact clothes and accessories in its own photo. Never copy the beige beanie, black cap, black logo shirt or cross-pattern trousers from the source unless that subject's own photo shows them. Never exchange identities or outfits between subjects. Pets dance upright on their hind legs with their own furry paws and animal proportions, never human hands or a human body. Copy the source performance beat by beat, including its different expressions and mouth shapes: - 0–1.5 s: the dancer on the right steps back from the camera into line beside the left dancer. - 1.5–4.5 s: both bounce with bent knees, swinging their bent arms and fists near waist level in sync. - 4.5–7 s: both lift their hands, clap and wave with open palms beside their heads, following the exact hand shapes and timing in the source. - 7–9 s: hands return toward the waist and sweep across the body as both keep bouncing. - 9 s to the end: the left dancer throws both hands overhead; the right dancer finishes their own source moves. Keep the dusk parking lot, parked cars, grey houses, pale blue sky and fixed phone camera. Exactly two subjects, no extra dancers. Photorealistic. No text.

When to use which

FAQ

Does Genjutsu follow heights written in the prompt?

Not in this test. Genjutsu Motion Transfer sized each pet to the dancer it replaced, so a 50 cm kitten came out as tall as a 70 cm pug. With two people it looked right. It seems to fill the outline of each source dancer, so to use it for subjects of very different sizes you would need a separate source clip for each size, such as adults, children, large dogs and small pets. Kling 4.0 Video Edit kept the kitten shorter, as asked, from one clip.

How are the two billed?

Kling 4.0 Video Edit counts the source seconds plus the output seconds, at $0.12 per second at 720p on KeelRouter. Genjutsu Motion Transfer counts output seconds only: $0.20 per second at 480p, $0.44 at 720p. Our test clip was 5.07 seconds, which Genjutsu billed as 6, so the run came to $1.20 for Kling at 720p and $1.20 for Genjutsu at 480p; Genjutsu at 720p would have been $2.64.

Why are the two clips at different resolutions?

We ran each the way our app would. Our Genjutsu dance templates run at 480p. Kling 4.0 Video Edit only has 720p and needs a source at least 720 pixels on both sides, so we gave it the 720 × 1280 original.

How do the prompts point at the photos?

Kling 4.0 Video Edit names inputs in the prompt: <<<video_1>>> for the source clip, <<<image_1>>> and <<<image_2>>> for the photos. For Genjutsu we wrote "reference image 1" and "reference image 2", in the order the photos are sent. The rest of the prompt was the same.

Is one test enough to choose a model?

No. This is one clip with one pet run per model, plus one run with people on Genjutsu. If your swap changes body size, as pets in a human dance do, try your own clip on both.

More comparisons

• Wan 3.0 vs Seedance 2.5: one photo, one prompt, side by side
• Kling 4.0 vs Seedance 2.0 Mini: a pug takes the clapping cat’s place
• Claude Fable 5.2 vs Opus 5.5: price per token vs cost per task
• Gemini 4 Carbon vs Claude Opus 5.5: benchmarks, cost per task and routing
• Seedream 5.5 vs Seedream 5.0 Flash: editing curved, multi-line and RTL text in images
• Seedream 5.5 Layerize vs Seedream 5.0 Pro Layerize: occlusion order and 20+ Live2D parts