Deepseek Design

I want to try to design some product images of monitors for Amazon using Deepseek as main model. I'm going to do it via Figma so both agent and me can edit images. But Deepseek can't see so I need to give it eyes. What is the best way? I'm using Deepseek via ClinePass subscription and I have Google Pro subscription so it seems like I should somehow tell Deepseek to go to Gemini Flash 4.6 and ask is to describe an image? Is it the best way? What harness would be the best? I have experience with OpenCodex and its vision sidecar but it uses Google AI studio model and limits there are not enough for design work where I assume model would want to look at the result at every turn