flux-2-klein

Generate images with Flux 2 Klein variants on RunComfy.

5|2|Updated May 18, 2026
One-click install
npx skills add https://github.com/doany-ai/skills --skill flux-2-klein-doany-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: flux-2-klein
Source: https://github.com/doany-ai/skills/tree/main/flux-2-klein
Command: npx skills add https://github.com/doany-ai/skills --skill flux-2-klein-doany-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Flux 2 Klein on RunComfy enables latency-first image generation workflows for real-time art-direction and rapid concepting, reducing iteration time from minutes to seconds.

Core Features & Use Cases

  • Latency-optimized inference using 4B variant for fast concepting and 9B variant for polishing detail.
  • Multi-reference styling with consistent branding across iterations.
  • Clear prompt grammar and model-specific guidance for subject-first composition and edge-case handling.

Quick Start

Provide a subject-first prompt and run the 4b endpoint to generate a draft image for rapid concepting.

Frequently Asked Questions about flux-2-klein

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate images rapidly for real-time art direction?

To generate images rapidly for real-time art direction, use the latency-optimized 4B model variant for concepting and the 9B variant for polishing detail. This approach reduces iteration time from minutes to seconds during sessions.

Can I maintain brand consistency across multiple image generation iterations?

Yes, you can maintain brand consistency across image generation iterations using multi-reference styling. This feature ensures consistent branding and visual coherence across multiple generated outputs and prompt workflows.

Do I need a RunComfy account to use latency-focused image generation workflows?

Yes, you need a RunComfy CLI account and proper model access to execute latency-focused image generation workflows. These prerequisites ensure reliable inference and access to the required 4B and 9B model variants.

What is the recommended step count for fast concepting versus final renders?

The recommended step count for fast concepting is 4 to 8 steps using the 4B variant, while final render polishing requires approximately 25 steps using the 9B variant for high quality detailed outputs.

What prompt engineering structure works best for deterministic image generation?

For deterministic image generation, use a subject-first prompt structure. Clear prompt grammar and model-specific guidance handle edge cases and ensure proper subject composition during the rendering process.