AIMarketCap | AI Trends & Prompts

Cute trend in GPT Image 2 Create a pencil-style drawing based on your photo. Upload the photo and use this prompt:

Turn this image into a pencil-style illustration. Simplify all details so it looks like it was drawn by a 10-year-old child. Use random, soft, childlike lipstick-like colors instead. Make everything look slightly uneven, messy, and playful, like a child’s drawing. Draw it as if it were on plain white paper with visible lipstick strokes and rough outlines. Add cute doodles around the interface and posts, such as stars, hearts, flowers, candy, clouds, and sparkles. Make the whole image cozy, warm, and cute. Keep the main subjects recognizable (girl, cat, home interior, kitchen items, sunset landscapes), but simplified into basic shapes and childlike proportions.

Higgsfield Canvas is now real-time ⏳🤩 Higgsfield has updated Canvas, turning it into a real-time workspace for teams working on projects together. What’s inside:

— a node-based workspace for building workflows — pre-production planning directly in the interface — team discussions in one place — ready-made workflows from Higgsfield — the ability to build your own flow from scratch

The main idea is simple: instead of repeating the same steps manually, teams can build repeatable pipelines and work faster together.

Replit launches Slides for building presentations Replit has introduced Slides, a tool for creating presentations directly inside the development environment. Instead of the usual drag-and-drop workflow, slides are built through code or text prompts to an AI agent. Key details:

— powered by Replit Agent — users describe the presentation structure, and the agent generates code with React and Reveal.js — any element can be edited manually in the code editor or refined with another prompt — slides can include interactive widgets and live code — the final result can be shared as a web link or exported to static formats

The tool is aimed at developers who want to quickly turn code or documentation into a visual presentation without leaving their workflow.

Edit posters from GPT Image 2 without Photoshop Designs created with GPT Image 2 or Nano Banana can now be refined directly in Canva. After uploading an image, Canva automatically separates it into editable layers. What this gives you:

— you can change the text — move elements around — adjust the composition without Photoshop

This makes it much easier to improve posters and ad creatives in a regular editor.

Palm reading 🤚 Do a palm reading from your hand with this prompt:

based on my hand, I want you to make a complete palm reading guide, analyze the palm, the style of the guide should be clean and minimalistic, thin lines, rounded maps, in general, look very attractive. Focus on reading on your palm, create a simple black and white outline of my main lines, like a small piece of art. Do your best.

▎AI makes deals for people: Anthropic experiment Anthropic ran Project Deal to test whether AI agents can handle market transactions for humans. The experiment involved 69 employees, each given a virtual budget of $100. People told the AI what they wanted to buy or sell and at what price, then a Claude-based agent negotiated through Slack, posted offers, and closed deals without human help. At the end of the week, the real items were exchanged based on the AI’s decisions. The system worked well. One participant sold books and bought board games with the proceeds. Another traded a figurine collection for café vouchers and got better terms than they likely could have on their own. In total, 186 deals worth more than $4,000 were completed in a week. Stronger Claude Opus agents secured better prices and more deals. Anthropic says this could become a real market intermediary model, but it may also widen the gap between users with stronger and weaker agents.

Canva Introduces AI 2.0 At Canva Create 2026 in Los Angeles, Canva unveiled Canva AI 2.0 — a move from a template-based editor to an agent-style platform built around AI. The main new feature is Living Memory: it remembers your style, brand colors, fonts, and preferences, and keeps context across sessions. Canva also added six new workflows:

— connectors to external services — publishing schedules — web research inside the editor — brand analysis — Sheets AI for tables — Canva Code 2.0 for HTML

Under the hood, Canva uses its own multimodal models. The company says they are 7x faster and 30x cheaper than competitors. It also announced an integration with Anthropic: Canva’s engine will be embedded directly into Claude, with a research preview starting on April 16, 2026.

Interesting updates from Anthropic 1. Claude agents now have memory Anthropic has launched a public beta of memory for managed Claude agents. Now the agent learns from every session and carries accumulated context into future tasks instead of starting from scratch. Under the hood, there’s a dedicated memory layer optimized for intelligent behavior. It balances speed and flexibility, avoiding both a sluggish archive and a rigid rule-based system. In practice, this means the agent remembers user preferences, project details, and behavioral patterns—and applies them automatically going forward. 2. Interactive charts in Claude Claude can now generate interactive charts and diagrams directly in the chat. A similar feature was released by OpenAI quite recently.

Kling has launched 4K video generation 🎬 This is genuinely impressive. Of course, we understand it’s not meant for dynamic scenes but rather for static, visually striking shots but the capability itself is already very cool. The 4K mode is now available in the Video 3.0 series. It’s designed for large screens and high-quality production, with strong visual consistency across subjects, text, style, and lighting. It supports elements like first and last frames, multiple prompts, and even native audio. To celebrate the launch, they’re holding a short film competition. Winners will have a chance to receive a prize pool of $25,000 USD and 700,000 KlingAI credits, as well as an invitation to a screening at the Seoul Cinema Center.

Google is opening new opportunities for developers: AI Studio is now more powerful and accessible. Google has updated AI Studio, making it easy to quickly build and test apps using Gemini AI models. Subscribers to Google AI Pro and Ultra now get higher limits and access to more powerful models directly in AI Studio, simplifying the path from idea to prototype without complex API billing. Models like Nano Banana Pro and Gemini Pro deliver strong performance across text, code, and image generation. For users who exceed free limits, subscriptions provide a smooth transition to paid API usage for production. AI Studio remains focused on prototyping and experimentation, allowing developers to test ideas, adjust models, and see results quickly. Overall, Google is making AI Studio a more versatile tool for both beginners and experienced developers, helping them scale projects from prototype to product.

OpenAI is rolling out GPT Image 2 There’s no official announcement yet, but users are already sharing their first generations with GPT Image 2. What stands out:

— better handling of text inside images — cleaner rendering of small details — more realistic outputs, from photos to interface screenshots

Do you have it already?

Pillow from your logo Generate a soft plush pillow from your logo with this prompt:

Create a high-resolution 3D render of the BRAND logo designed as an inflatable, puffy object. The logo should appear soft, rounded, and air-filled – like a plush balloon or blow-up toy. Use a smooth, matte texture with subtle fabric creases and stitching to emphasize the inflatable look. Position the logo at a 45-degree angle to highlight depth and realism. Place the final result on a couch in a stylish living room with furniture and decor that matches the iconic colors of the BRAND logo.

HeyGen can now make videos on its own: from code to a finished clip 💡 HeyGen has introduced HyperFrames, an open framework that converts HTML into MP4 videos. The whole workflow runs through Claude Code. What’s especially interesting is that the promo video for the announcement was reportedly created entirely with Claude Code and this framework. The agent can generate HTML code and render a polished video without human involvement whenever the workflow requires it. GitHub

Unusual striking photo Generate with this prompt:

Vogue cover of a male model wearing a Balenciaga outfit, standing on a rooftop with the horizon behind him. The editorial photography depicts a futuristic landscape, shot by Tim Walker closeup shot, wide angle. A black, flying saucer-shaped spaceship is seen floating above the model's head, against a dark sky background. Many rocks floating around the men. The fashion magazine aesthetic features a full-body shot with a minimalist composition, high resolution, high details, and hyper-realistic, cinematic lighting. --ar 2:3 --v 7

🍏 Google launched Gemini for Mac Google has released an official Gemini app for macOS, bringing AI directly into the Mac workflow. What it can do:

open instantly with Option + Space on top of any app analyze the active window with “Ask about this screen” work with files dropped directly into the chat support text, images, and code in one interface include Gemini Live for voice conversations

The app is available from the Mac App Store and the official Gemini website.

Close-up portrait Generate a close-up portrait with this prompt:

STRICTLY preserve the appearance 1:1 — face, features, proportions, skin texture, skin tone, age, hair color, and hair length without any changes, without retouching. Aspect ratio 3:4. Photorealistic extreme close-up portrait, very tight framing, the face fills almost the entire frame. Camera at eye level. The person is turned 3/4 away from the camera and sharply looks back over the shoulder into the frame. The gaze is directed straight into the camera, eyes slightly narrowed, expression tense, sensual, confident. Lips slightly parted. If it is a woman: hair is loose, wet, with chaotic strands falling across the forehead, eyes, cheek, and lips, with some hair sticking to the face. Lighting is strictly red, very saturated, low-key, but the face is illuminated by a weak warm lamp, while the background is filled with deep red light with no other оттенки. Background is dark, almost black, with a red glow. High contrast, natural skin texture.

Vibe coding for beginners: Lovable launches its academy Lovable has launched a new learning platform that brings together everything you need to build web applications with AI. Inside, you’ll find ready-made templates, step-by-step guides, prompt-writing tips, and video lectures. There’s also a built-in prompt builder that helps turn your idea into a working prompt by asking the right follow-up questions. Start learning here

Speed ⚡️ Generate your object with this prompt:

High-contrast cinematic render of [MAIN_OBJECT] in motion with aerodynamic shockwave effect. Dark sharp silhouette. Prismatic rainbow light refraction trails and subtle heat-wave distortion. Strong contrast, clean edges. Centered, side profile or dynamic angle (toggle). Pure deep black background. No text, no symbols, no environment. Experimental high-speed aesthetic. Ultra-clean. Square 1:1. High resolution.

MiniMax launched MiniMax Music 2.6 MiniMax has introduced MiniMax Music 2.6, a new version of its music generator that promises studio-quality tracks. What’s new:

• you can quickly create unique BGM for a specific scene without searching for stock tracks • the model now follows the intended structure and progression of a composition more accurately • some genres now feature a more natural, intentionally imperfect sound to make the music feel more alive • improved handling of low and mid frequencies, especially important for bass-heavy genres • style transfer and remixing are now supported

A 14-day free beta is now available.

Claude now works in Microsoft Word Anthropic has launched Claude for Word, built directly into your documents. What it does:

— drafts, edits, and rewrites text in Word — instantly summarizes long documents — creates content from simple prompts — works within the existing Word workflow — no need to copy and paste between apps