This study proposes a valuation framework and determination model to estimate the appropriate royalty rate for the technology transfer of generative AI models. As complex technological assets, generative AI models comprise model architecture, trained weights, training and fine-tuning data, prompt assets, retrieval-augmented generation structures, safety modules, operational know-how, retraining rights, output usage rights, and liability arrangements. Therefore, simply applying conventional royalty rates used for patented technologies or general software does not adequately reflect the economic value and risks associated with generative AI models. To address this limitation, this study proposes a royalty rate determination procedure that involves decomposing technological assets, defining the scope of rights, estimating base value, allocating technology contribution, adjusting for risk, incorporating real option value, and deriving a negotiable royalty range. In particular, this study identifies the economic useful life of training data, data contribution, reusability, rights and contractual conditions, copyright and privacy risks, and model obsolescence as key adjustment factors. The results suggest that the appropriate royalty rate for generative AI models should be determined not as a single fixed rate, but as a risk-adjusted negotiation range reflecting data rights, model performance, technological life, substitutability, exclusivity, liability risk, update rights, and expansion potential.