AIMarketCap | AI Trends & Prompts

Trending prompt 🖼️ Upload a photo and use the prompt below:

A surreal mid-century editorial illustration depicting an eccentric version of the man in the photograph and a [TIGER] inside a stylish [luxury restaurant], elongated proportions, exaggeratedly narrow faces, deadpan expressions, oversized accessories, bizarre fashionable atmosphere, picturesque brush textures, distorted anatomy, subtle low-grade surrealism, sophisticated absurdity, vibrant retro color palette expressive brushstrokes of oil painting, fashionable vintage clothes, spectacular jewelry and glasses, whimsical characters in the background, softly blurred in the distance, elegant cafe culture energy, cinematic composition, textured background, contemporary art gallery combined with an illustration of the 1960s Riviera, highly stylized editorial aesthetics, 4:5 vertical composition

Figma launches its own AI agent Figma has introduced a built-in design agent that works directly inside the file with no third-party tools and no switching between windows. It understands the full project context: components, tokens, libraries, and the design system. That’s the key difference from external tools, which only read the file from the outside. What it can do:

— generate multiple visual directions from a single prompt — compare different approaches to the same screen — update typography, component states, and content across the entire file — create component documentation — turn team comments into actual design changes

You can also run multiple prompts in parallel and iterate with the agent inside the same file. It’s currently in beta, and doesn’t consume credits yet. Available for Professional, Organization, and Enterprise plans.

Google releases Gemini 3.5 Flash At Google I/O, the company unveiled Gemini 3.5 Flash. The model outperforms Gemini 3.1 Pro on most benchmarks and runs 4× faster than other flagship models. Key numbers:

— Finance Agent v2 — 57.9%, best result among all compared models — MMMU-Pro — 83.6% — CharXiv Reasoning — 84.2% — GDPval-AA Elo — 1656 vs 1314 for 3.1 Pro — MCP Atlas — 83.6% vs 78.2% for 3.1 Pro

Gemini 3.5 will be built directly into Google Search: AI dialog in the search bar, file uploads, agent creation for specific tasks, and delegating routine work product search, price comparison, news monitoring without switching tabs.

Manus now supports scheduled tasks with context retention Manus has released Scheduled Tasks 2.0. The key difference from the first version is that recurring tasks now continue within the same conversation instead of starting from scratch. Previously, each scheduled run created a separate new task, and the agent lost previous instructions, files, and decisions. Now Manus returns to the same conversation with accumulated context. Three new scenarios:

— Tasks with history — daily digests or weekly reports continue in an already configured thread — Projects — scheduled tasks use shared files, connectors, and instructions from the project — Web apps — apps built in Manus can now update data on schedule without user involvement

Also added: a calendar view and run history to see what's completed and what's planned.

GitHub and Microsoft have launched a free course on AI agent development The first part of the official course, “Developing in Agentic AI Systems,” is now available on Microsoft Learn. It’s free, includes a certificate, and is designed for beginner to intermediate learners. What you’ll learn: — how to design, implement, and manage agentic AI systems across the full development lifecycle — the fundamentals of agentic AI in GitHub — agent architecture and integration into the development workflow — tools, MCP servers, and runtime environments What you need to get started: — a GitHub account — a basic understanding of AI — familiarity with repositories, branches, and pull requests An important point: AI agent management has officially been recognized for the first time as a separate engineering role. The course is aimed at DevOps engineers, developers, and solution architects. This is the first of two parts. The second part will be released later via the same link.

ChatGPT can now analyze your bank accounts OpenAI has launched a built-in finance tool that lets users connect their accounts and ask ChatGPT where their money is going, which subscriptions are active, and what upcoming payments they have. The integration works through Plaid and supports more than 12,000 financial institutions, including Chase, Fidelity, Robinhood, and American Express. ChatGPT only has read-only access it can see transactions, balances, and liabilities, but it cannot transfer money or manage your accounts. The feature is currently available only to Pro subscribers in the US on web and iOS. OpenAI plans to expand access to Plus users later.

OpenAI Codex now works on your phone 📱 OpenAI has added Codex to the ChatGPT mobile app. Now you can connect to any machine running Codex whether it’s a laptop, a Mac mini, or a remote server and manage it right from your phone. What’s available on mobile:

— view all active tasks — approve commands — switch models — start new tasks — track results in real time, including screenshots, terminal output, diffs, and test results

Files, access, and settings stay on the machine where Codex is running, while updates are synced to your phone live. For teams using corporate remote environments, Remote SSH is now available too. Codex can connect directly to those servers, and the desktop app automatically detects hosts from your SSH config. Device sync works through a secure relay layer, so machines are never exposed directly to the internet 🔒 Codex is available on ChatGPT Plus, Pro, Business, Edu, and Enterprise.

Selfie with anime characters Generate a cool selfie with your favorite anime characters using this prompt:

Ultra-realistic 9:16 vertical format fisheye selfie of me with [Shinchan, Doraemon, Naruto, Nobita]. We’re all making silly, exaggerated faces. Set in a small, bright living room with white tones. High camera angle. Extreme fisheye distortion. Realistic, cinematic lighting, anime characters integrated with stylized realism.

The “kiss before the disaster” trend is everywhere right now 🔥 We’ve prepared a ready-to-use template where you can add your faces and create the same viral video in just a couple of minutes. You can try it and create your own video прямо on the website 👀

Claude Code can now work until the result is reached without your input Claude Code now includes the goal command. You define the completion condition once, and the model keeps working for as many steps as needed until that condition is met. After each step, a separate fast model Haiku by default checks whether the goal has been completed. If not, Claude automatically starts the next step. If yes, the goal is marked as done. It works together with auto mode. Auto mode removes confirmation prompts for each tool call, while goal removes confirmation for each step. Together, they enable a fully autonomous session.

Scare things up in ChatGPT with this prompt:

Generate an image of the scariest thing you could possibly imagine scribbled into a child's coloring book

It will create a creepy image in the style of a children’s coloring book, straight out of a horror movie.

Open Design is an open alternative to Claude Design that runs on your computer Open Design has appeared on GitHub an open-source alternative to Anthropic’s Claude Design. It runs locally, can be deployed on Vercel, and isn’t tied to any single agent or provider. How it works:

— it doesn’t include its own AI, but connects to tools you already use: Claude Code, Codex, Cursor, Gemini CLI, OpenCode, or Qwen — the agent is detected automatically at launch — inside are 19 ready-made skills for landing pages, dashboards, mobile screens, presentations, and documents — it also includes 71 design systems from companies like Linear, Stripe, Vercel, Notion, Apple, Anthropic, and Tesla — before generation, the agent asks clarifying questions and can suggest 5 visual directions if no brand is provided — the result is rendered in an isolated iframe and can be exported to HTML, PDF, PPTX, or ZIP

Higgsfield can now predict a video’s performance before release Higgsfield has launched a new tool that helps estimate the potential of videos up to 15 seconds long before they go live. What it does:

— estimates the probability of virality — analyzes the strength of the opening hook — predicts attention retention — shows a heatmap of viewer response

If you enable Ad Reference, the model can automatically rework the clip for stronger performance metrics. The tool is already available via web, MCP, and CLI. For now, it runs in demo mode and doesn’t consume credits.

Ideogram launches a background remover Ideogram has launched BG Remover, a tool for removing backgrounds from images. Its main difference from most alternatives is that it handles transparent areas correctly. Typical background removers replace transparent parts with white or gray. BG Remover preserves transparency in the alpha channel, so glass stays glass, liquid stays liquid, and fine logo details and text don’t get blurred. The tool is designed for three main use cases:

— logos with fine details — product photos with glass or plastic packaging — images with typography

It’s available on all Ideogram plans, including the free tier. It works via the web interface and API for integration into your own services.

ЮAnthropic teaches agents to learn while you sleep Anthropic has added three new modes to Claude Managed Agents:

— Dreaming: the agent analyzes past sessions in the background, finds repeated mistakes, and saves the conclusions to memory. You can enable automatic updates or require manual approval. — Outcomes: a separate evaluator agent compares the result against a target outcome and triggers revisions if something doesn’t match. According to Anthropic, this can improve task success rates by up to 10 percentage points. — Multi-agent orchestration: a lead agent breaks a complex task into parts and assigns them to specialized agents, with all steps visible in Claude Console.

Dreaming is currently available by request only.

Microsoft launches Copilot Cowork mobile app and expands system integrations Microsoft has introduced the Copilot Cowork mobile app and added new features for working with business systems. Users can now start tasks on their smartphone and continue them on desktop without losing context. What else is new: — new connectors for integrating with enterprise systems — access to data and workflows across different business tools — a more seamless experience between mobile and desktop

Updated Clarity A great image upscaler that can gently “humanize” both generated and real people without changing their appearance. It also supports 8K. You can use it in Weavy AI through the Import model node by pasting this link there.

Grok has been taught to do everything: text, images, and video on a single canvas xAI has launched Agent Mode — a workflow where you don’t need to switch between different tools. In one window, you can sketch an idea in text, generate images, edit them, and instantly turn them into video. All of this happens without switching to other services or losing context. For now, it’s only available on the website.

Logo for your business Generate a logo for any business niche using this prompt:

Create a logo for [brand / project / blog / product] named [name]. Come up with the visual concept, symbol, typography, color palette, and 2–3 usage variations yourself. The logo should be simple, memorable, and easy to read at small sizes: on an avatar, website, packaging, and business card. Avoid complex illustrations — we need a clean symbol or a neat combination of symbol + name.