One prompt · three models
Three results
9:16 · 8 sec. · 720pThree AI video generation models got the same prompt about a salmon wearing a GoPro and an underwater rave, and the results were very different.
The takeaway at a glance
Quick summary
Judging by six frames from each model, Gemini Omni Video delivers the clearest and most stylistically consistent salmon rave, while Seedance 2.0 Mini is let down by a completely black frame.
Same conditions
One prompt
A wildlife researcher gently secures a tiny waterproof action camera to a healthy salmon and releases it into a crystal-clear mountain river. Create a single continuous first-person GoPro POV shot from the fish. The salmon swims rapidly downstream through bubbles, rocks, reeds and shafts of natural sunlight, with believable underwater motion, water resistance and rushing river audio. It dives through a narrow rocky passage and unexpectedly enters a hidden underwater cavern. Inside, dozens of salmon wearing colorful 1990s sunglasses are dancing naturally at a secret underwater rave under neon pink, cyan and violet lights. Fish swim past the camera and react playfully. The reveal should feel surprising, funny and highly shareable while remaining photorealistic. Energetic 1990s dance music with believable underwater acoustics. Wildlife-documentary detail, seamless spatial continuity, no cuts, no visible camera rig, no captions, no logos, no watermark.Show translation
A wildlife researcher carefully attaches a tiny waterproof action camera to a healthy salmon and releases it into a crystal-clear mountain river. The result is a single continuous first-person shot, as if filmed by the fish's own GoPro. The salmon swims swiftly downstream through bubbles, rocks, reeds and shafts of natural sunlight, with believable underwater motion, water resistance and the sound of a rushing river. It dives into a narrow rocky passage and unexpectedly enters a hidden underwater cavern. Inside, dozens of salmon in colorful nineties-style sunglasses are dancing at a secret underwater rave under neon pink, cyan and violet light. Fish swim past the camera and react to it playfully. The reveal should be surprising, funny and highly memorable while staying photorealistic. Energetic nineties dance music with believable underwater acoustics. Wildlife-documentary detail, seamless spatial continuity, no cuts, no visible camera mount, no captions, no logos, no watermarks.
Score comparison
| Criterion | Kling 3.0 | Seedance 2.0 Mini | Gemini Omni Video |
|---|---|---|---|
| Prompt | 7 | 5 | 8 |
| Consistency | 6 | 4 | 7 |
| Movement | 6 | 5 | 7 |
| Physics | 6 | 5 | 7 |
| Camera | 6 | 4 | 7 |
| Artifacts | 5 | 3 | 6 |
| Cost/result | 6 | 4 | 8 |
| Overall result | 6 | 4.3 | 7.1 |
| Generation time | 176 sec. | 143 sec. | 48 sec. |
| Cost on AI OGO | 15,976 credits | 11,520 credits | 10,481 credits |
Model breakdown
Kling 3.0
An average result with noticeable distortion in the climactic scene.
The frames move from the camera being attached to the salmon, through a swim along a rocky passage, to a rave in a neon-lit cavern. The overall style comes close to photorealism, and the scene with the sunglasses and lighting reads clearly. In the final frames, however, the fish silhouettes look deformed, which undermines the scene's credibility. A set of still frames can't show how smoothly the video transitions between these moments.
Seedance 2.0 Mini
Poor prompt adherence due to a failed frame and cartoonish fish.
The wide shot of the river and the splash on entry are rendered with reasonable energy, but one of the six frames is completely black, which looks like a generation failure. The fish at the rave look cartoonish, with almost human facial features, rather than photorealistic as the prompt required. The first-person POV barely comes through: several frames look more like they were shot from the side than from a camera mounted on the fish.
Gemini Omni Video
The most consistent visual sequence of the three models.
The frames form a logical sequence, from the riverbank through a narrow rocky passage to a cavern where salmon in sunglasses dance at a neon rave. The photorealistic style holds up even in the surreal scene with dancing fish. Light shafts and water splashes look convincing. One of the cavern frames is dark, but the fish are still distinguishable in it. The actual motion dynamics can't be judged from individual stills.
How we compared
Test methodology
The evaluation is based on six still frames from each model, without access to the video files themselves. The frames capture key points in the story: attaching the camera to the salmon, the swim through the rapids and the climax at the cavern rave. This set lets you compare composition, style, lighting and story continuity across frames, but it does not prove smooth motion, fully accurate water physics or how the camera actually behaves between the captured frames.
- Test date
- September 28, 2026
- Format
- 9:16
- Duration
- 8 sec.
What to choose
Winners
Gemini Omni Video holds photorealism more consistently in the climactic scene, builds more logical transitions between frames and avoids obvious failures like the competitor's black frame.
