Users estimated ComfyUI Cloud's RTX 6000 tier at roughly $4.62 per GPU hour, while another thread pointed to €3000 workshops selling ComfyUI expertise. A subgraph failure post showed why complex graphs still create paid setup demand.
the cloud pricing math turned a credit system into per minute and per hour numbers, the workshop thread made the business market visible, and the subgraph PSA linked back to a public workflow repo. Those three posts sketch the same picture from different angles: ComfyUI is free, but reliability, setup, and hosted compute are where the money shows up.
The pricing complaint is unusually concrete. The post converts ComfyUI Cloud's 4,200 credits for $20 into roughly $0.0047 per credit, then multiplies that against observed burn of about 0.27 credits per second.
That yields about $0.077 per minute and $4.62 per hour for the RTX 6000 96GB tier, according to the original calculation. The same post says the $100 plan would cover only about 36 minutes of daily runtime over a month if that per second estimate holds.
The workshop thread lands on a different part of the stack. The screenshot shows ComfyUI services priced in the thousands of euros, and commenters did not treat that as surprising.
One reply inside the same thread said a local ComfyUI asset creator was producing around 50 game levels with four variants each per hour, while artists using another tool managed 5 to 6 a day. Another reply in that discussion argued the seller was monetizing time spent mastering the tool, not charging for ComfyUI itself.
The most useful counterpoint is not about money at all. It is about fragility. The post says subgraphs killed a favorite workflow and left the author wary of both subgraphs and get/set nodes.
That matters because the author also linked a public workflow collection, which makes the complaint less theoretical. In the same ecosystem where cloud GPU time is being priced to the second and training is being sold as a service, one broken abstraction is still enough to push experienced users back toward "simple and noodly" graphs.
