FLUX.2 [max] is the top-tier image generation and editing model developed by Black Forest Labs, representing the highest quality offering in the FLUX.2 family. It excels in producing professional-grade outputs with unmatched photorealism, precision, and visual consistency, surpassing FLUX.2 [pro] in image quality, prompt understanding, and editing reliability across all dimensions. Released as an advancement over previous models, it supports complex workflows requiring superior hit/miss ratios for intricate edits involving multiple references.

Key features include state-of-the-art multi-reference editing for maintaining character, product, and style consistency; exact color matching via hex codes; production-ready text rendering; reliable spatial reasoning with accurate physics, lighting, and perspective; and support for up to 4MP resolutions in any aspect ratio. The model handles long prompts up to 32K tokens and generates images in under 10 seconds, making it suitable for high-volume production while offering maximum creative control.

What makes FLUX.2 [max] unique is its hybrid architecture combining flow matching with advanced transformer and diffusion techniques, enabling unprecedented detail in textures like fabric and skin, faithful style representation, and precise editing that preserves geometry, lighting, and identities even in complex scenes with up to 8 reference images. It bridges the gap between generated and photographed content, with enhanced world knowledge for grounded visualizations of trends, events, and styles.

• Architecture: Hybrid flow-matching transformer-diffusion model
• Parameters: Not publicly disclosed
• Resolution: Up to 4MP, any aspect ratio; works from low-res inputs (400x400px minimum)
• Input/Output formats: Text-to-image, image-to-image, multi-reference (up to 8 images via API), JSON-based control, hex color codes, up to 9MP total input capacity
• Performance metrics: Sub-10-second generation speeds, 32K text input tokens, up to 3x faster than competing models of similar quality

• Use for scenarios demanding highest consistency in complex multi-reference edits or superior prompt adherence, where lower-tier models like pro or flex may fall short
• Best practices: Provide detailed prompts with specific instructions for positioning, lighting, and styles; use multiple high-quality references for character/product consistency; incorporate hex codes for exact brand colors
• Common pitfalls: Overly vague prompts can reduce precision; exceeding recommended reference limits (e.g., more than 8) may degrade quality
• Quality vs speed trade-offs: Delivers maximum quality with near-pro speeds, but prioritizes precision over flex's typography specialization
• Prompt engineering tips: Structure prompts with clear actions (e.g., "Replace X with Y from image Z, matching lighting"); leverage long context for complex scenes; reference images explicitly for multi-element combinations

• Optimal parameter settings: Use full 32K token context for intricate descriptions; set aspect ratios dynamically; enable generative expand/shrink for scene adjustments
• Prompt structuring advice: Start with subject/action, add style/lighting details, end with references (e.g., "The person from image 1 pets the cat from image 2 in a sunny park, exact hex #FF5733 for accents")
• How to achieve specific results: For product retexturing, prompt "Redesign surface material to leather while preserving shape and lighting, reference image 1"; for character consistency, use repeated multi-references across batches
• Iterative refinement strategies: Generate low-res drafts first (400px), then upscale/edit; refine via targeted inpainting on specific areas
• Advanced techniques: Combine hex color refs with spatial prompts for UI mockups (e.g., "Typography in brand font, color #001F3F"); multi-ref for scene population (e.g., "Insert animals from images 2-6 into scene from image 1, adjust proportions naturally")

• Exceptional photorealism closing the gap with real photography, especially in skin, hair, fabric textures, hands, and architectural details
• State-of-the-art image editing with highest consistency in preserving colors, lighting, faces, text, objects, and identities across complex changes
• Best-in-class prompt following for short/long instructions, vast world knowledge for grounded current events/styles
• Multi-reference control supporting up to 8 images for seamless element combination (characters, products, environments)
• Production-ready features: Exact hex color matching, complex typography/UI, reliable spatial reasoning with physics/perspective
• High versatility: Text-to-image, image-to-image, any aspect ratio up to 4MP, low-res to high-res workflows

• Professional product marketing and e-commerce: Generating consistent, marketplace-ready photos with precise retexturing and variations
• Interior design and 3D reconstruction: Combining furniture/elements from multiple refs into coherent scenes
• Character design for video production: Maintaining facial features/expressions across poses, environments, styles
• Food imagery and branding: Exact color-matched visuals with realistic textures/lighting
• Creative asset batches: Hundreds of consistent images for games/ads, using multi-ref for style preservation
• UI/typography mockups: Reliable text rendering and brand color accuracy in complex layouts

• Experimental multi-reference shines in complex edits but performs best with clear prompts distinguishing elements
• Known quirks: May require prompt tweaks for extreme deformations; excels more in photorealism than abstract art
• Performance: Sub-10s generations scale well, but high-res/multi-ref increases compute needs
• Resource requirements: Handles 9MP total inputs efficiently on enterprise hardware
• Consistency: Superior hit rate for pro-level tasks, praised for "production-grade" reliability in user tests
• Positive feedback: Users highlight "unmatched quality" in realism/prompt adherence; "game-changer for editing" in community shares
• Common concerns: Slightly higher cost for max quality tier; occasional minor artifacts in very crowded scenes

• Optimized for photorealism and precision editing, less ideal for highly stylized/abstract art or speed-critical low-quality drafts
• Parameter count and full training details not disclosed, limiting custom fine-tuning to dev variants
• Dependent on prompt/reference quality for peak performance; can underperform with ambiguous inputs compared to specialized flex for typography