Meta launches LlamaVision-XL: high-resolution multimodal model for creatives

AI · 4 min read

LlamaVision-XL handles inputs up to several thousand pixels and generates layered outputs that preserve separate elements like foreground subjects, backgrounds, and masks. This is designed to reduce manual extraction work during compositing.

Meta emphasizes creative control: the model returns editable layer maps, suggested blend modes, and masking recommendations, plus annotations for edge refinement and artifact handling. Outputs are optimized for common design tools and high-DPI print formats.

The company offers both cloud and limited on-prem deployment options for enterprise creative teams concerned about IP. Meta also published guidelines for anti-hallucination tuning to ensure outputs remain faithful to prompts and references.

Creative agencies testing the beta reported strong results for rapid concepting and print-ready mockups, though they still recommend artist oversight for final-grade assets. Meta plans future collaboration plugins with major creative suites.