Two Strangers Made an AI Film. Won $150K?

Results are in for the biggest AI film competition ever. 8,752 entries poured in from 139 countries. The #1 winner, 'GRANDMA vs WASP,' was made by two people living in Detroit and Germany.

Carat EditorMar 20, 20264 min read

What's up — Carat crew here. 🤗 Today we've got 8,752 AI short films from 139 countries, Microsoft pulling out its own image AI, and more. Let's get into all 5 stories.

🔥 Two Strangers Made an AI Film. Won $150K?

Results are in for the biggest AI film competition ever. 8,752 entries poured in from 139 countries.

The #1 winner, 'GRANDMA vs WASP,' was made by two people living in Detroit and Germany. They've never met in person. They worked async — one person creating while the other slept. The prize? $150,000 (roughly $200M KRW).

The kicker? Third place went to 'SCRATCH' — two brothers who made it in just 5 days. Viewers DM'd them asking 'where's the full movie link?' They thought it was an actual theater trailer.

The second-place winner plans to use the $100K prize to make a superhero film. A project that would normally cost millions — attempted solo.

139 countries. 8,752 films. That number tells you one thing: AI filmmaking isn't an experiment anymore — it's real production. On Carat, you can start making your own short films with 11 video models including Sora 2, Kling, and MiniMax.

📌 3 Stories You Should Know

1️⃣ MiniMax Built an AI That Evolves Itself?

Chinese AI company MiniMax just dropped M2.7. This time it's not video — it's an LLM. The standout feature: 'self-evolution.' The AI analyzes its own failures, fixes its code, runs evaluations, then decides what to keep and what to toss.

After 100+ iterations, performance jumped 30%. On coding benchmarks, it scored on par with GPT-5.3 Codex. No human fine-tuning needed — the AI just gets better on its own.

They also added 'Agent Teams' — multiple AI agents splitting roles and collaborating, catching errors by challenging each other's work.

2️⃣ Microsoft Drops Its Own Image AI — Debuts at #3

Microsoft just released MAI-Image-2, its own image generation model. Until now, Microsoft relied on OpenAI's DALL-E. Not anymore.

It debuted at #3 on the AI image generation leaderboard (Arena). Right behind Google Gemini (#1) and OpenAI GPT Image (#2). Their first-gen model cracked the top 10 last October — 5 months later, they're already at #3.

Strong points: photorealism, text rendering inside images, and cinematic composition. Rolling out to Copilot and Bing Image Creator soon.

Every big tech company is now building its own image model. Google, OpenAI, and now Microsoft. The AI image space is heating up fast. You can try all these latest AI image models on Carat.

3️⃣ 10 Images Is All AI Needs to Clone Your Art Style

A new trend is picking up in AI image generation. Upload 10–30 of your own works, and the AI learns your entire style — line weight, color palette, lighting, character traits. All of it.

For designers and illustrators, this opens the door to mass-producing work while keeping their signature style intact.

Big tech platforms are rolling this out one by one. Image AI is evolving from 'a tool that generates random images' to 'a partner that learns your brand.'

This matters most for marketers and designers who need brand consistency. Once AI learns your style, cranking out drafts gets way faster. On Carat, you can use the 'Consistency' feature to generate images that stay true to your look.

🧪 Prompt Tip of the Day

One Prompt to Make Game Boss Fight Animations

More and more creators on X are using Seedance to make game-style animations. @NACHOS2D_ finished a Brawl Stars-style animation solo in 10 days, and the Tournament of Power clip hit 29K views and 159 bookmarks.

The trick is writing prompts like a cinematic director: scene setup → character action → camera movement → mood. That's how you get trailer-quality results. Try the prompt below on Carat with Kling O3.

A surreal battlefield in the sky. Floating rock islands drift through a thunderstorm, clouds swirling below like an ocean.

A masked swordsman sprints across the drifting rocks. Behind him, a colossal winged beast closes in. A vortex of storm clouds and lightning spins inside its chest.

The camera hurtles from island to island, chasing the swordsman. Rocks tilt, spin, and crumble beneath his feet. Every wingbeat sends shockwaves through the air, blasting debris and rain straight at the screen.

Rapid handheld cuts capture the swordsman leaping impossible gaps. His blade carves arcs of light through the darkness.

Final shot: the swordsman launches off the last crumbling rock, rides a bolt of lightning straight into the beast's chest vortex. One last slash detonates the storm from within. A blinding flash clears the sky.

Here's a video we made on Carat using Kling O3 with the prompt above ↓

Generated with Kling O3 on Carat

The key is being specific about 'camera movement' and 'physical reactions.' Directives like 'handheld cut' or 'shockwave shakes the frame' help Kling O3 nail dynamic shots. Dramatic action descriptions like 'rides the lightning and charges' also boost output quality big time.

You can try this with Kling O3 on Carat right now. Swap the character to a ninja, pirate, or wizard — or change the setting to lava, outer space, or cyberpunk. Totally different results every time.

Try today's prompt tip — just swap the character and hit generate. You might be surprised by what comes out ☺️