Hailuo Olympic diving cats: origin, prompt and current models

Runbo Li
Runbo Li
·
· 4 min read
ca

Quick answer

Hailuo Olympic diving cats are AI-generated clips of cats performing stylized platform dives; they are not recordings of animals doing the stunt. The format became associated with MiniMax's Hailuo 02 physics demonstrations in 2025. To recreate it now, use a current MiniMax video model such as Hailuo 2.3 or MiniMax H3 where available, describe one dive and one camera setup, and label a realistic result clearly.

The original article treated virality, hashtag views, generation speed, and visual realism as measured facts without preserving sources or analytics. Those claims have been removed. This guide separates the observable format from hypotheses about why people share it.

Recreate the idea with a current video model

Use the prompt as a starting point, generate one short clip, and inspect the complete result before iterating.

Try Text to Video

What the trend looks like

cat

A typical clip shows one animal approaching or leaving a diving board, a single leap or rotation, a pool landing, and a sports-broadcast or slow-motion visual treatment. Variations swap the animal, venue, camera position, costume, or commentator.

Which Hailuo model should you use now?

MiniMax launched Hailuo 02 in June 2025 with text-to-video and later image-to-video support. Its current API catalog still lists Hailuo 02, but it also lists Hailuo 2.3 and Hailuo 2.3 Fast. Magic Hour currently offers MiniMax H3, a newer reference-driven MiniMax video model with native audio.

  • MiniMax H3: use when current Magic Hour access, reference-led control, longer supported durations, or native audio fits the intended clip.
  • Hailuo 2.3: use when working through MiniMax's current direct video API and you want text-to-video or image-to-video.
  • Hailuo 2.3 Fast: use for image-to-video when the faster direct-API variant fits the job.
  • Hailuo 02: keep for reproducing the original 2025 workflow or when a provider specifically exposes it.

Provider names and model controls differ. Confirm the exact model, input mode, aspect ratio, duration, resolution, audio behavior, price, and terms in the interface you use.

A practical diving-cat prompt

A tuxedo cat on a regulation springboard above an indoor competition pool. Side broadcast camera, locked framing. The cat takes two careful steps, makes one clean forward dive with a single rotation, enters the water paws-first, and creates a small realistic splash. Bright arena lighting, spectators softly out of focus, no text, no logo.

Keep the action simple. A prompt that asks for an approach, several flips, a camera orbit, a reaction shot, a scoreboard, and a perfect landing in one short generation creates more opportunities for motion or anatomy errors.

Image-to-video gives you more visual control

Create or supply a rights-cleared starting image with the cat, board, pool, camera angle, and framing already established. In image-to-video, describe only the motion you want. This can reduce scene drift, but it does not guarantee correct anatomy or physics.

MiniMax's current image-to-video API supports Hailuo 2.3, Hailuo 2.3 Fast, and Hailuo 02, with documented duration and resolution combinations. It also documents camera commands such as static, push, pull, pan, tilt, tracking, and zoom. Use no more camera movement than the shot needs.

Prompt variations

  • "A tabby cat performs one backward dive from a low platform, fixed wide camera, arena pool, gentle splash, no text."
  • "A calico cat in a tiny swim cap pauses at the edge, then completes one pencil dive, slow-motion side view, realistic water, no logo."
  • "A fluffy white cat performs a deliberately clumsy belly flop from a low board, wide comedic broadcast shot, safe fictional scene."
  • "A black cat completes one clean dive at an outdoor pool at sunset, static camera, subtle crowd reaction, cinematic but physically plausible motion."

How to review the result

  • Inspect the cat's legs, tail, eyes, body shape, and contact with the board frame by frame.
  • Check whether the takeoff, rotation, entry, splash, and water surface connect coherently.
  • Reject clips that imply a real animal was put at risk or that resemble a real event in a misleading way.
  • Check all text, logos, flags, uniforms, venue marks, and audio for errors or unauthorized use.
  • Add a clear synthetic-media label or context when viewers could reasonably mistake the scene for real footage.
  • Use only music, voices, images, and references you have permission to use.

Why the format can be shareable

The concept combines a familiar sports broadcast with an impossible animal performance, and it communicates the joke quickly without much setup. Loopability, surprise, recognizable framing, and easy variation are plausible reasons creators reuse the format. They are hypotheses, not proof that a clip will earn reach.

Measure each post through qualified views, completion, rewatches, shares, profile or site visits, and the action that matters to the account. Compare it with similar posts. A hashtag total or one viral example does not predict the next generation's performance.

Frequently asked questions

No. They are synthetic video scenes. Do not present them as footage of real animals performing dives.

No evidence supports that broad claim. The format became associated with Hailuo examples, but similar clips can be made with multiple video models. Identify the actual model only when the creator or generation record establishes it.

No. MiniMax's current API catalog lists Hailuo 2.3 and Hailuo 2.3 Fast as well as Hailuo 02, and Magic Hour currently offers MiniMax H3.

Use the shortest duration that contains one complete action. Direct MiniMax duration and resolution combinations vary by model; Magic Hour's MiniMax H3 has its own current platform limits. Verify the selected route rather than copying an old 5-to-10-second rule.

Official sources checked

Historical Hailuo 02 facts come from MiniMax's launch announcement. Current direct models and controls come from the MiniMax model catalog and image-to-video API documentation. Current Magic Hour access comes from the MiniMax H3 model page. Sources were checked September 13, 2026.

Runbo Li
Runbo Li
CEO of Magic Hour
Runbo Li is the Co-founder and CEO of Magic Hour, where he builds AI video and image tools for content creation. He is a Y Combinator W24 founder and former Data Scientist at Meta, where he worked on 0-1 consumer social products in New Product Experimentation. He writes about AI video generation, AI image creation, creative workflows, and creator tools.
View author →

Continue Reading

ghibli trend
Studio Ghibli AI trend: what happened and safer prompts (2026)
30+ AI Art Statistics: Key Insights and Trends To Watch
30 AI art statistics on artists and collectors
AI Video Editing Trends and Tools
AI video editing trends in 2026: six workflow shifts
tHE GEN AI CREATIVE ECONOMY
Generative AI & creator economy: 28 verified stats (2026)
HAILUO 02 SIMPLE GUIDE
Hailuo 02 guide: current status vs MiniMax H3
Editorial comparison of MiniMax H3 modular workflows and Kling 3.0 multi-shot video production
MiniMax H3 vs Kling 3.0: open weights or managed control?