CBV: Clean-label Backdoor Attacks on Vision Language Models via Diffusion Models | Read Paper on Bytez