Hi, this is the Carat team. 🤗
Today we have a story about a 2-person company that hit $1.8B in revenue using AI tools, Grok's official prompt guide, and the news shaking up the AI image market.

Matthew Gallagher, based in LA, launched telehealth startup Medvi with just $20,000 and two months of prep. His entire team? Himself and his brother — just 2 people.

Coding was handled by ChatGPT and Claude. Ad images were made with Midjourney, videos with Runway. Even customer support runs on AI agents. First-year revenue hit roughly $400M, and this year they're on pace for $1.8 billion.

The New York Times covered the story, calling it "Sam Altman's prediction of a one-person billion-dollar company come true." The key detail: all marketing visuals were AI-generated, and the quality was nearly indistinguishable from agency work.

The Grok team (@XCreators) published an official prompt guide. Here are the key takeaways for getting better image results.

The guide says to stack five layers in your prompt: specific subject, mood reference, lighting, composition, and finishing details. For example, instead of "a flower seller," try "a street vendor arranging dried flowers on a cobblestone alley in Lisbon, warm golden light from the right, eye-level medium shot, vintage color grade, shallow depth of field." The difference is night and day.

Another useful tip: if writing prompts feels hard, let AI write them for you. Describe your idea in a chat, and it'll convert it into an optimized image prompt. If you want consistent results every time, try Carat's style consistency feature — set your style once, and every generation matches it.

Two major image AI updates landed in the same week.

Microsoft released its in-house image model MAI-Image-2 on April 2. After relying on OpenAI, MS is going independent — with 2x faster generation speed. WPP, the world's largest ad agency, adopted it immediately, and it's rolling out to Copilot and PowerPoint next.

Midjourney announced V8.1 for release this week. Cost optimization is the focus, and V8 will be deprecated just two weeks after launch.

Netflix's AI team released VOID on April 3 — their first-ever public model. It removes objects from video.

What sets it apart is that it handles physics too. Remove a person, and the cup they were holding falls to the ground. Shadows and reflections disappear naturally. In user evaluations, 64.8% preferred VOID — Runway scored just 18.4%.

This is Netflix's first public release of its own video technology. The concept of "editing scenes without reshooting" could reshape post-production workflows.

You've probably seen those "slime face" videos trending on TikTok — a character turns into glossy jelly, and hands squish it like dough. Oddly satisfying, and people can't stop watching.

All you need is a character image and a video prompt. Upload your image and use this prompt:

Two hands entering frame and gently kneading the entire jelly face like soft dough. Deep compression, folding and stretching. Cheeks spreading across the surface, facial features distorted then slowly bouncing back. Oddly satisfying ASMR motion, ultra realistic physics

Here's what we made on Carat with Kling v3. ⬇️

The key terms in the video prompt are kneading like soft dough for the squish motion, bouncing back for the elastic return, and oddly satisfying ASMR which tells the AI to keep it slow and smooth.

Make your image on Carat with Nano Banana 2, then generate the video with Kling v3. Keep it 5–10 seconds for the best performance on Reels and Shorts.

If the prompt looks fun, copy it and paste it into Carat right now.

See you next week with more AI updates. Have a great day! 😊