Skip to main content

​
Camera & Lens Cheat Sheet

​
Lighting Keywords

Describe the light source, its direction, and whether it is soft or harsh. Example lighting phrases:

​
Style Keywords

​
Composition Techniques

​
Model-Specific Quick Reference

  • Start with a short description. FLUX 3 Image expands it before generating, so add detail only for what you want to control: positions, appearance, light, or text.
  • Refer to references by position: “image 1” is the first entry in images. Send up to 10.
  • For layout control, follow the caption with an element table: {"id", "bbox", "desc"} rows, boxes as [top, left, bottom, right] on a 0 to 1000 grid, each id cited in the caption as <id>. Set aspect_ratio to the shape you drew the boxes for.
In this example, [150, 150, 850, 850] puts the runner in the middle 70% of the frame. Hover over the caption to see each box, or move the pointer over the image to read its coordinates:
  • Write edits as a record: Modify <ref_image_0> by replacing … with … Keep the rest of the image exactly unchanged, preserving …
  • Use a hex code to specify a target color: change the color of only one bird in the middle to #e01075
  • Quote the exact words for text in the image and describe the type: weight, case, color, alignment. Multi-line posters and Japanese text render in the launch examples.
  • There are no negative_prompt, seed, prompt_upsampling, or guidance fields. Unknown fields return 422.
See: FLUX 3 Text to Image → · Layout prompts → · FLUX 3 Image Editing → · Precise editing →
Other languagesFLUX.2 can be prompted in multiple languages, so you don’t need to write in English. English prompts tend to produce the most precise results, since most of the training data is in English.Reference images per API requestFLUX.2 [pro] has a 9 MP total limit for input plus output. At 1 MP output you can use up to 8 reference images, at 2 MP output up to 7, and so on.See: Multi-Language → · Which FLUX.2 model to choose →
  • No negative prompts supported
  • Excellent typography: use quotation marks for exact text
  • Hex color codes in text-to-image prompts for brand colors, text, and gradients: “The logo text ‘ACME’ in color #FF5733”
  • JSON structured prompts supported for production workflows
  • Reference specific camera models for authentic photorealistic looks
Text tips
  • Front-load text descriptions for better accuracy
  • Keep text short: long strings are harder to render accurately
See: JSON Prompting → · Typography & Design → · HEX Color Prompting →
  • What you write is exactly what the model receives
  • Write in prose, not keyword lists: describe scenes like a novelist
  • Describe the light source, direction, and softness
  • Supports image editing with single and multi-reference inputs: up to 4 reference images (input_image … input_image_4)
  • Add Style: [style]. Mood: [mood]. at the end for consistent aesthetics
See: Single-Reference Editing → · Multi-Reference Editing →
  • Specify what should change: the input image provides all other visual context
  • Use quotation marks for text editing: Replace 'joy' with 'BFL'
  • Be explicit about preservation: “while maintaining the same facial features and hairstyle”
  • Prefer specific verbs: “change the clothes” over “transform the person”
  • For multiple changes, add as many explicit details as possible