What problem does it solve? Choosing the right Flux 2 Klein variant, step count, and prompt structure is non-obvious, and naive prompting against the model produces weak results. This Skill bundles the model's documented prompting patterns, endpoint schemas, and routing guidance so image generation requests produce sharper output on the first attempt. ## Core Features & Use Cases - Variant routing: Picks between the 4B variant (sub-second latency for live iteration) and the 9B variant (~25 steps for polished final output), and redirects to Flux 2 Pro, Seedream 5, or GPT Image 2 when the task exceeds Klein's limits. - Model-specific prompting guidance: Applies subject-first declarative grammar, step-count-by-phase strategy, and multi-reference alignment rules that empirically improve Flux 2 Klein output. - CLI-based generation: Invokes runcomfy run blackforestlabs/flux-2-klein/<variant>/text-to-image with validated JSON input, async polling, and automatic download of results. - Use Case: During a live art-direction session, iterate on product visualizations at 4-8 steps on the 4B variant, then re-run the same prompt at 25 steps on the 9B variant for the final asset. ## Quick Start Ask the AI to generate an image with Flux 2 Klein, for example: "Use flux 2 klein to generate a product photo of a matte ceramic mug on a wooden table with soft window light."
