Agent Skills: klein9b-prompting

Use when prompting FLUX.2 [klein] 9B for image generation or editing, when keyword lists, vague lighting, or unclear reference-image roles are producing weak results, resulting in scene-first prompts with explicit visual priorities.

UncategorizedID: zenobi-us/dotfiles/klein9b-prompting

Install this agent skill to your local

pnpm dlx add-skill https://github.com/zenobi-us/dotfiles/tree/HEAD/devtools/files/pi/agent/bundles/creator/skills/imagegen/klein9b-prompting

Skill Files

Browse the full folder contents for klein9b-prompting.

Download Skill

Loading file tree…

devtools/files/pi/agent/bundles/creator/skills/imagegen/klein9b-prompting/SKILL.md

Skill Metadata

Name
klein9b-prompting
Description
Use when prompting FLUX.2 [klein] 9B for image generation or editing, when keyword lists, vague lighting, or unclear reference-image roles are producing weak results, resulting in scene-first prompts with explicit visual priorities.

klein9b-prompting

Overview

FLUX.2 [klein] 9B responds best to scene-first prose, not keyword soup. It does not upsample prompts, so the wording you give it is the wording it gets.

Reference guide: references/flux2-klein-prompting-guide.md.

Quick Reference

| Task | Prompt shape | |---|---| | Text-to-image | Subject → setting → details → lighting → atmosphere | | Lighting control | Name source, quality, direction, temperature, and surface interaction | | Style control | Add a short style or mood tag at the end | | Single-image edit | State the transformation, not a generic improvement | | Multi-reference edit | Assign roles: image 1 = identity, image 2 = style or environment |

Prompt Rules

  • Write as flowing prose.
  • Put the main subject first.
  • Make lighting explicit; “good lighting” is too vague.
  • Keep every sentence visually useful.
  • Use style or mood tags only when they help consistency.
  • For edits, describe the target change clearly and let the input image provide the base.

Good Prompt Shape

An elderly fisherman in a salt-stained wool sweater stands at the bow of a small wooden boat, calm water around him, morning mist on the horizon. Soft, diffused dawn light comes from camera-left, catching on wet rope and the boat’s chipped paint. Quiet, cinematic, photoreal.

Editing Patterns

  • Style transfer: “Turn this portrait into a moody winter editorial.”
  • Object swap: “Replace the bicycle with a black horse.”
  • Environmental change: “Change the scene to winter with falling snow.”
  • Multi-reference: “Keep image 1’s person and pose; use image 2 for neon city mood and reflective lighting.”

Common Mistakes

  • Keyword lists instead of prose
  • Buried subject or buried action
  • Vague lighting like “nice light”
  • “Make it better” / “fix the image” style edits
  • Mixing up what each reference image should control
  • Stuffing prompts with filler like “ultra-detailed” when it adds no visual meaning