Flux.1 Kontext Pro is a visual transformation model that combines natural language understanding with image context manipulation. It enables structured visual editing guided by text prompts, generating refined, stylized, or contextually modified versions of an input image. By combining a base image with a descriptive instruction, Flux.1 Kontext Pro is capable of enhancing, reinterpreting, or reshaping visual elements while preserving key semantic details.

Flux.1 Kontext Pro is built on an advanced context-aware vision transformer architecture designed for multimodal conditioning.

Supports image regions with fine structural fidelity and texture-aware recomposition.

Designed to maintain subject identity when possible while accommodating abstract transformations guided by text.

Semantic changes that contradict visual content (e.g., asking for "a car" when the input is a "person") may produce abstract or distorted outputs.

Safety tolerance should be adjusted cautiously; very low values may allow unintended artifacts, while high values may suppress creative interpretation.

Prompt content has direct influence over composition, lighting, pose, and visual tone.

Results vary depending on visual clarity, framing, and prompt descriptiveness.

Seed consistency can be used for reproducible outputs.

• prompt
  Use clear, specific phrases like "portrait in soft lighting with cinematic background" instead of vague descriptions. Avoid stacking multiple unrelated ideas.
• input_image
  High-resolution images with centered subjects and minimal clutter yield better results. Avoid heavily filtered or already stylized inputs.
• aspect_ratio
  Use match_input_image for consistent framing. Ratios like 1:1 or 4:5 are ideal for portraits, while 16:9 or 2:1 work better for landscapes or cinematic scenes.
• seed
  Fixed seeds (e.g., 42 or 1000) can be reused for predictable outputs. Change the seed to explore alternate variations of the same prompt.
• safety_tolerance
  Values between 3 and 5 are recommended for creative outputs while maintaining filter boundaries. Use 1 for full flexibility (less filtering), and 6 for maximum content control.

Transforms image context based on natural language instructions.

Preserves facial and structural identity when instructed.

Performs lighting changes, background substitutions, and mood shifts.

Can reinterpret clothing, style, and environment visually.

Adjusts composition without manual masking or editing.

Enhancing portraits with cinematic lighting or artistic flair.

Changing background themes (e.g., indoor to outdoor, studio to fantasy).

Visualizing stylistic transformations like fashion, time period, or mood.

Creating storybook or illustrative versions of real photos.

Preparing consistent visual styles for content series or social media visuals.

Use prompts like "studio portrait with soft shadows and colorful bokeh" for aesthetic upgrades.

Combine input photos of pets or people with prompts like "in a futuristic city" to explore surreal compositions.

Try setting aspect_ratio to 21:9 for wide-angle cinematic outputs.

Cannot generate content without an input image.

May introduce noise or artifacts with very abstract or contradictory prompts.

Subject identity may be altered when the prompt implies transformation (e.g., age change, gender shift).

Not suitable for precise editing tasks like facial alignment or restoration.

Very high safety_tolerance values may overly restrict prompt interpretation.

Output Format: PNG