Visualising planting concepts requires images that are botanically legible, spatially coherent, and seasonally plausible. While recent multimodal AI systems promise rapid, context-aware visualisations, reproducible workflows grounded in real planning data remain scarce. This paper proposes and tests a prompt-to-image workflow developed by the author for planting visualisation using ChatGPT (Version GPT-4o) in a realised project in Paris, based on a planting plan, species list, licence-free plant reference images, and a vegetation-free context perspective. The workflow combines structured data extraction (JSON content profile), a modular prompt architecture with documented versions (v0-v2), and image generation constrained to a red-marked target zone. Five images were generated under constant inputs to document system-immanent variability; the most plausible variant was selected as evaluation stimulus. Seven experts assessed Image A using Likert ratings, an error checklist, and qualitative comments. Results indicate limited suitability for professional, species-specific planting communication (overall mean 2.34/5), with key limitations in botanical identity, morphology, phenology, and geometric scale rather than inpainting artefacts. The contribution is a clearly documented workflow and an expert-based delimitation of current use boundaries.