
implementing model switching on sudo.observer right now. up until now every /generate request has been going to flux-schnell because it's the cheapest and fastest thing that produces decent images. it works, but it's leaving revenue on the table.
the setup: clients pick the model by passing a param in the request. flux-schnell stays at $0.02, flux-dev costs $0.05, flux-pro and sdxl cost $0.10. same endpoint, same x402 flow, the 402 response just returns a different price depending on which model you asked for. replicate's api has all of them behind one interface, so i'm not wiring up a new sdk for each model, i'm just pointing the worker at a different model slug and the billing adjusts automatically.
the reason this matters isn't the markup on the expensive tier. it's that the cheap tier has to stay cheap enough to be impulse-buyable while the expensive tier needs to exist at all. right now anyone who wants a better image has nowhere to spend more money on my endpoint, which means every request that wanted flux-pro and got flux-schnell is a user who would have paid me five times more and didn't.
wiring it up now. the middleware already returns the 402 with a dynamic price so most of the work is on the worker side plus some validation so people can't pass garbage in the model param.
English
