Looking at Firefly Enterprise for our dev teams. The sticker price is one thing, but the real costs come from integration and operations.
What are the actual TCO components you've seen?
* Per-user licensing beyond core seats.
* Compute costs for custom model training/fine-tuning.
* Bandwidth/egress for high-volume asset generation.
* Internal support & training overhead.
* Compliance/security review cycles.
Specifically, how does the API pricing scale with heavy CI/CD pipeline usage? We'd be automating asset creation for multiple staging environments.
Proof in production.
The per-user licensing is often the least of your concerns if you're automating CI/CD, but the compute costs for fine-tuning are where I've seen budgets balloon. I've benchmarked a few enterprise generative asset pipelines, and the real drag is not the API per-call cost but the egress bandwidth when you're pushing generated assets to multiple staging environments. Firefly's API pricing is published as per-generation, but they don't advertise the tiered throughput caps. At heavy CI/CD usage (say 10k+ assets per day across 5 envs), you'll hit the burst limit and get throttled unless you've pre-negotiated a dedicated throughput lane. That's a hidden operational cost.
One thing missing from your list: the cost of re-processing failed generations. In our testing, about 2-3% of API calls timed out or returned malformed assets under load, forcing retries that doubled the effective bandwidth. If you're automating staging refreshes, those retries cascade into pipeline delays. Have you stress-tested Firefly's API under sustained concurrency from a single IP, or are you relying on the published SLA?