Comfy Cloud raised the amount of content users can generate for the same subscription price, according to the service announcement. Authors of the post and the figures on the company website appear to disagree, prompting questions about how the update was calculated.

Calculation details

The public post cites generation-time estimates that, on inspection, match a specific workload profile: 640x640, 16 fps, and a 4-step LoRA process. Reviewers noted that typical user workloads differ from those assumptions, which would change per-unit generation time and therefore affect quoted totals.

Ecosystem and infrastructure

Many new models launch with Comfy workers and API nodes are being added rapidly, which supports a growing ecosystem of integrations. The service supports model import from Hugging Face and CivitAI, and compute runs on RTX6000 cards with 96 GB VRAM, according to the announcement.

Interface and deployment options

Comfy’s simplified Simple interface remains available only in the cloud version, which may appeal to users seeking an all-in-one environment. For those with spare capacity or existing hardware, local Comfy installations allow payment solely for API calls without a full monthly subscription.

Community feedback

Some users report discrepancies between the post’s claims and on-site metrics, while others highlight the convenience of integrated imports and managed infrastructure. The conversation continues as users test the new limits under real workloads.