Krea Agents Opens Beta to Replace Node-Wiring With Plain English Prompts

Krea just opened the beta waitlist for Krea Agents, a new platform built on top of its existing creative suite that lets AI agents drive the canvas instead of requiring users to wire nodes and pick models manually. Describe what you want, and an agent plans the pipeline, selects the models, and produces finished images or video.

Most agentic AI tools have targeted developers and data teams. Krea's bet is that the same pattern works when the output is visual and the operator is a designer. With a hosted MCP endpoint on one side and a canvas-aware planner on the other, the company is positioning itself as connective tissue between coding agents and creative tools, so a Claude Code or Cursor session can produce finished visual assets without dropping into a separate UI.

How the agent works

The core mechanic debuted in Krea's Node Agent: type what you want to make, and the system reads your canvas, plans the pipeline, wires the nodes, and runs the job. Before touching anything, it shows you a plan. For a futuristic city, that might be four stages: prompt refinement, image generation on K1 or Flux, enhancement, then video.

The agent is stateful, which separates it from most generation tools. Whatever is already on the canvas, including existing nodes, connections, and outputs from earlier runs, gets factored in. A style node from a previous session gets reused rather than rebuilt from scratch.

What's shipping alongside it

Krea Agents was previewed alongside Krea 3, the company's next foundation model. Recent infrastructure additions include a Slack integration and a hosted MCP server designed for agent-to-agent workflows.

The Krea MCP server lets an external AI agent call Krea directly to generate images and video, enhance and upscale assets, apply trained styles, and run multi-step creative workflows from a single prompt. It works with Claude, Codex, Cursor, and other MCP-compatible agents.

The model stack

Krea Agents orchestrates a large catalog of third-party and in-house models rather than shipping a single new one:

• 60+ aggregated models including Flux, Veo 3.1, Kling, Runway, Luma, and Ideogram, alongside Krea's own Krea 2.
• Krea 2 is a single-stream Diffusion Transformer at 12 billion parameters, built with Grouped-Query Attention, gated sigmoid attention, QK-Norm, SwiGLU MLPs at 4x expansion, zero-centered RMSNorm, and 3D axial RoPE positional encoding.
• Style personalization via LoRA fine-tuning on Krea 2, which trains a small set of adapter parameters instead of retraining the base model.
• MCP and API surfaces so external agents can call Krea as a tool.

Who's behind it and where it sits in the market

Krea has over 30 million users across 191 countries and is backed by a16z and Google Gradient. It competes in a crowded field alongside Midjourney, Runway, and Adobe's Firefly, but has staked out a distinct position as an orchestration layer, routing each step of a creative job to whichever model performs best rather than pushing a single house model.

Krea Agents extends that logic to the workflow itself. If Krea 2 was about pulling model quality in-house, Agents is about pulling the full production pipeline in-house, replacing the ComfyUI-style graph-wiring that has kept professional AI creative work confined to technically fluent users. A ten-node creative pipeline shouldn't require learning node graphs.

What developers should know

For teams already building on generative models, three things make Krea Agents worth tracking. It is one of the first serious attempts to bring the agent pattern into visual production at scale. It commits to interop through MCP and API, slotting into existing Claude Code or Cursor pipelines rather than replacing them. And it comes from a team that has shipped both an in-house foundation model and a polished product layer around dozens of external ones.

The beta is invite-only, distributed through the waitlist and social engagement on the launch announcement. Krea has not yet published benchmarks, Agents-specific pricing tiers, or a firm timeline for general availability.