Meta’s Secret Weapon, 9 Months in the Making

Nine months after creating a dedicated AI research lab, Meta just unveiled its first model: Muse Spark. It can read text, understand images, search the web, and split complex tasks across multiple workers automatically. Meta went from dead last on the AI leaderboard to 4th place overnight.

Carat EditorApr 9, 20263 min read

Hey — it’s the Carat team. 🤗
Today: Meta’s big comeback, 1.4x faster image generation, and a creator adding Hollywood-grade VFX to AI video.

🔥 Meta’s Secret Weapon, 9 Months in the Making

Nine months after creating a dedicated AI research lab, Meta just unveiled its first model: Muse Spark.

It can read text, understand images, search the web, and split complex tasks across multiple workers automatically. Think of it as an AI that can see, read, search, and get things done.

Meta went from dead last on the AI leaderboard to 4th place overnight. Some reviewers say it outperforms Claude Opus 4.6 and GPT-5.4 in certain areas.

You can try it now in the Meta AI app. The developer API starts as a partner-only preview, and Meta says they plan to open-source future versions.

Yesterday Anthropic said their model is "too powerful to release." Today Meta says "we’ll open-source ours." Two companies at a similar level, taking completely opposite approaches. The industry still hasn’t agreed on how open AI should be.

📌 Three Things Today

1️⃣ Image Generation Just Got 1.4x Faster

Black Forest Labs, the team behind FLUX.2, just dropped a speed upgrade. Same prompt, 1.4x faster output, less memory usage.

Quality stays nearly identical while speed goes up. Great news if you frequently generate high-resolution images.

FLUX.2 is available on Carat right now. Once this upgrade rolls out, your same prompts will return results even faster.

2️⃣ Grok Is Already Building Its Next Image Model

Elon Musk’s xAI is training 7 AI models simultaneously, including Imagine V2 — the next version of Grok’s image generator.

Grok Imagine’s quality mode just launched, and the next version is already in training. That’s how fast the image AI race is moving.

3️⃣ A Creator Is Adding VFX to AI Video

AI video creator @voxelplot added Hollywood-grade VFX to AI-generated video, pushing the quality to game trailer levels. Energy effects, lighting — the full package.

The workflow: generate images with AI, turn them into video, then color grade. Over 2,300 bookmarks. We’re entering an era where game-production-level post-processing is possible with AI.

🧪 Prompt Tip of the Day

One Prompt, One War Movie Scene

AI video creator @umesh_ai fed a single prompt into Seedance 2.0 and got a full war movie scene. Thousands of warriors, a hazy sky, a commander on horseback — all from one prompt.

The trick: be specific about three things. ① Camera angle (what perspective), ② Time of day (golden hour, night), ③ Atmosphere (smoke, ash, dust). Breaking it down like this helps AI create much richer scenes.

Epic wide-angle shot of a vast ancient battlefield at golden hour, thousands of warriors clashing with swords and shields under a hazy amber sky thick with smoke and ash. A lone commander on horseback raises a blade, backlit by the dying sun, while siege towers burn in the distance and arrows arc across the sky in slow motion.

Seedance 2.0 is coming soon to Carat. If you want to create epic scenes like this, remember to include camera angle, time of day, and atmosphere keywords in your prompts.

See you tomorrow with more. ☺️