CANVIL: Designerly Adaptation for LLM-Powered User Experiences introduces a practice for designers to engage with large language models (LLMs) as an adaptable design material. The study identifies three key characteristics of designerly adaptation: 1) low technical barriers, 2) leveraging designers' user-centered perspectives, and 3) encouraging iterative model tinkering. Based on these characteristics, the authors developed CANVIL, a Figma widget that enables designers to author system prompts, test adapted models, and integrate model outputs into interface designs. A group-based design study with 17 designers explored the implications of integrating designerly adaptation into design workflows. The results showed that designers could iteratively adapt LLMs to enhance user interaction and recognized collaborative opportunities for adaptation. The study also highlighted the need for tools that support designers' user-centered expertise in LLM-powered product development. CANVIL is presented as a technology probe that operationalizes designerly adaptation, enabling designers to experiment with LLMs within their design workflows. The study underscores the potential of collaborative AI tinkering tools and the importance of materiality in shaping social practices within product teams. The work contributes insights from a formative study with designers, the development of CANVIL, and empirical findings from a task-based design study. The authors propose a workflow for designerly adaptation to guide future research and practice. The study also discusses the implications of designerly adaptation for collaborative design and beyond, including a proposed workflow for designerly adaptation. The work highlights the need for tools that support designers' user-centered expertise in LLM-powered product development.CANVIL: Designerly Adaptation for LLM-Powered User Experiences introduces a practice for designers to engage with large language models (LLMs) as an adaptable design material. The study identifies three key characteristics of designerly adaptation: 1) low technical barriers, 2) leveraging designers' user-centered perspectives, and 3) encouraging iterative model tinkering. Based on these characteristics, the authors developed CANVIL, a Figma widget that enables designers to author system prompts, test adapted models, and integrate model outputs into interface designs. A group-based design study with 17 designers explored the implications of integrating designerly adaptation into design workflows. The results showed that designers could iteratively adapt LLMs to enhance user interaction and recognized collaborative opportunities for adaptation. The study also highlighted the need for tools that support designers' user-centered expertise in LLM-powered product development. CANVIL is presented as a technology probe that operationalizes designerly adaptation, enabling designers to experiment with LLMs within their design workflows. The study underscores the potential of collaborative AI tinkering tools and the importance of materiality in shaping social practices within product teams. The work contributes insights from a formative study with designers, the development of CANVIL, and empirical findings from a task-based design study. The authors propose a workflow for designerly adaptation to guide future research and practice. The study also discusses the implications of designerly adaptation for collaborative design and beyond, including a proposed workflow for designerly adaptation. The work highlights the need for tools that support designers' user-centered expertise in LLM-powered product development.