How to Write Better Prompts for Flux in ComfyUI

Flux does not behave like older Stable Diffusion models. Keyword spam and a wall of quality tags do little, while clear natural-language sentences do a lot. This guide covers how to phrase prompts, where the guidance value fits, and how to get Flux to spell words correctly in the image.

What you need

• A working Flux Dev or Schnell workflow in ComfyUI
• The positive prompt node and, for Dev, the FluxGuidance node

Step 1: Describe the scene in full sentences

Flux was trained on detailed captions, so it rewards descriptive sentences. State the subject, the setting, the lighting, the camera angle, and the mood as if you were describing a photo to someone. Order matters less than clarity.

weak:  girl, forest, masterpiece, 8k, best quality, detailed

strong: A young woman in a green raincoat stands in a misty pine
forest at dawn. Soft golden light filters through the trees.
Shot on a 50mm lens, shallow depth of field, calm mood.

Step 2: Skip the legacy quality tags

Tags like masterpiece, best quality, and 8k were crutches for SD 1.5. Flux mostly ignores them and they can crowd out real description. Drop them and spend those words on the actual content of the image.

Step 3: Set guidance, not CFG

Flux Dev uses a FluxGuidance node rather than classic CFG. A value around 3.5 is the sweet spot. Lower values look softer and more natural; higher values increase contrast and prompt adherence but can look harsh.

Step 4: Render readable text in the image

Flux is unusually good at text. To put words on a sign or label, name the exact text in quotes and keep it short. Long passages still break, but a few words usually come out clean.

A rustic wooden cafe sign hanging from an iron bracket, the sign
reads "Open Daily", warm afternoon light, shallow depth of field

Result: prompts that match how Flux was trained, giving sharper composition, fewer wasted tokens, and clean in-image text when you ask for it.