The flagship of FLUX.2
FLUX.2 [max] combines the 32-billion parameter rectified flow transformer architecture with a Mistral-3 24B vision-language model, creating a 46,864-token context window that enables dramatically more nuanced prompt understanding than any previous FLUX model. The result is stronger adherence to complex multi-element prompts, more consistent character and style identity across references, and grounded generation that can visualize real-world trending subjects. FLUX.2 Max is the top tier in the FLUX.2 family — above FLUX.2 Pro — for workflows where maximum quality and prompt fidelity are the priority.Capabilities
Strongest prompt adherence
The highest-fidelity prompt following in the FLUX.2 family — complex multi-element scenes, specific spatial relationships, and detailed stylistic directives are executed reliably.
Up to 10 reference images
Accepts up to 10 reference images simultaneously for character consistency, product compositing, style anchoring, and multi-reference scene construction.
Grounded generation
Real-time web search integration enables accurate visualization of trending products, current events, and recently released subjects without detailed descriptions.
4-megapixel output
Generates up to 4MP (approximately 2752×1536) — print-ready resolution for commercial, editorial, and large-format work.
46,864-token context
Exceptionally large context window enables nuanced, multi-paragraph creative briefs and highly specific compositional instructions.
Advanced editing consistency
Reliable across retexturing, character consistency, spatial reasoning, and style transfer editing operations — maintains subject identity through complex transformations.
Specifications
| Feature | Details |
|---|---|
| Architecture | 32B rectified flow transformer + Mistral-3 24B VLM |
| Context window | 46,864 tokens |
| Resolution | Up to 4MP |
| Reference images | Up to 10 |
| Web grounding | Yes (real-time search) |
| Generation time | 8–10s (4MP), 4–6s (1MP) |
| Released | November 25, 2025 |
How to use
Write your prompt
Be as detailed as needed. With 46,864 tokens of context, Flux.2 Max can handle exhaustive creative briefs — describe spatial relationships, lighting, materials, mood, and style precisely.
Add reference images (optional)
Upload up to 10 reference images for character consistency or style anchoring.
Prompting tips
- Maximize the context window — Unlike smaller models, Flux.2 Max won’t drop details from long prompts. Use the space to be precise about lighting, materials, spatial composition, and style.
- Use reference images for consistency — For campaigns or series, passing reference images maintains visual identity far better than text descriptions alone.
- Web grounding works for current subjects — Reference recently launched products or current events by name; grounded generation handles the visual accuracy.
Example prompts
A high-end residential kitchen with Calacatta marble countertops, brushed brass fixtures, integrated appliances, warm pendant lighting over a central island, architectural photography style, shot at dusk.
A futuristic electric sports car (based on [reference image]) in a wind-tunnel test environment, dramatic directional lighting, photorealistic CGI quality.
Compare models
| Model | Parameters | Resolution | References | Best for |
|---|---|---|---|---|
| Flux.2 Max | 32B | Up to 4MP | 10 | Maximum quality, complex campaigns |
| Flux.2 Pro | 32B | Up to 4MP | 10 | Production-grade, text rendering |
| Flux 1.1 Ultra | — | 4MP | Image-to-image | Raw mode, high-res commercial |
| Flux Dev | 12B | ~1MP | — | Open-weight, creative exploration |
Flux.2 Max sits at the top of the FLUX.2 lineup. For production-grade output with strong text rendering where the absolute ceiling on prompt adherence isn’t needed, Flux.2 Pro is a slightly faster and more cost-efficient option.