A practical guide to prompt engineering for Flux Pro and Flux Dev — covering aspect ratios, negative prompts, style anchors, and batch generation via the API.
Flux Pro is currently the best commercially available image generation model for photorealistic outputs. But like all diffusion models, getting great results consistently requires understanding a few key mechanics.
Prompt structure
Flux responds well to structured prompts. Start with the subject, then add style descriptors, then lighting, then technical camera details. A prompt like portrait of a woman, cinematic, golden hour lighting, 85mm f/1.4, shallow depth of field will consistently outperform an unstructured description.
Aspect ratios
Flux Pro supports arbitrary aspect ratios. For social content, 9:16 (1080×1920) works natively without cropping. For landscape photography, 16:9 or 3:2. For product shots, 1:1. Always specify dimensions explicitly in the API call rather than relying on defaults.
Using the Nova API
Calling Flux through Nova is identical to any other image model. Set the model to black-forest-labs/flux-pro, pass your prompt, and specify width and height. Output is returned as a base64 PNG or a signed URL depending on your preference.
Batch generation
Need to generate 100 variations? Use the n parameter. Nova handles the concurrency automatically and returns all images in a single response object. At $0.055 per image for Flux Pro, generating 100 variants costs $5.50 — far less than a stock photo license for a single image.
Lena Müller
Head of Design at Nova