If you generate filler images for PDF guides, you have to decide which image tool to hand the job to. We ran three on the same job, called pdf_guide.filler, across 317 calls in the last 90 days.
The three were Stable Image Core (stability/stable-image-core), Stable Image Ultra (stability/stable-image-ultra) and Flux Schnell (replicate/flux-schnell). Below are our own figures for each tool, what separates them on cost, speed and failures, and which one we are still using today.
One caution before the numbers. The three tools did not run side by side. Each had its own stretch of dates and its own number of calls, so read the figures as what happened on our job, not as a controlled test.
What each tool cost us per call
Flux Schnell was the cheapest at $0.003 per call. We made 48 calls with it and paid $0.144 in total.
Core cost $0.03 per call, which is 10 times what Flux Schnell cost us, or $0.027 more per call. Across 171 calls we paid $5.13 in total.
Ultra cost $0.08 per call. That is 26.67 times what Flux Schnell cost us, and $0.077 more per call. Across 98 calls we paid $7.84 in total.
Notice that Core carried the most calls but Ultra carried the largest bill. At 98 calls, Ultra cost us more in total than Core did at 171.
All of these are what we paid per call on our own usage. They are not the vendors' list prices, and we are not quoting a rate card here.
How long each one took
Flux Schnell had a median of 3.8 seconds per call. It was the fastest of the three, so everything else is measured against it.
Core had a median of 8.7 seconds. That is 2.29 times slower than Flux Schnell, or 4.9 seconds slower per call.
Ultra had a median of 15.0 seconds. That is 3.95 times slower than Flux Schnell, or 11.2 seconds slower per call.
We only measured the median per call, so we cannot say more than that about the total time of a full guide.
How often each one failed on our side
Core had 0 failures in 171 calls. Ultra had 0 failures in 98 calls.
Flux Schnell had 1 failure in 48 calls, or 2.1 percent. That was the only failure across all 317 calls.
This is the count of calls that came back as failures on our side. It is not the vendor's uptime and should not be read as such.
Which one we kept and which we stopped using
We still use Core. Its last call was on 2026-10-05, and it is the only one of the three still in use.
Flux Schnell was used from 2026-07-08 to 2026-07-13, so only about five days. Ultra was used from 2026-07-14 to 2026-09-09. Core ran from 2026-07-12 onward.
The record does not say why we stopped using Flux Schnell or Ultra. What the figures do show is that Flux Schnell was the cheapest and the fastest and also the only one with a failure, and that Ultra was the most expensive and the slowest of the three. Beyond that we will not guess at a reason.
The limits of this comparison
This is not a benchmark. These are our prompts, for our one job, at our volumes, measured over the window above. A reader with a different job may get different costs, times and failure counts.
The call counts are uneven: 171, 98 and 48. The tools were also used over different dates, so they never ran under identical conditions. Flux Schnell in particular has only about five days of history behind its figures.
Our figures say nothing about image quality, because we did not measure it here. A cheaper or faster tool is only the better choice if the images it produces are good enough for your use, and that is a judgement these numbers cannot make for you.
Today we would pick Core for this job, since it is the one we kept using, it had no failures in 171 calls, and it cost $0.03 per call. We would pick Flux Schnell instead if cost and speed mattered more than anything else, because it cost $0.003 per call and took a median of 3.8 seconds, as long as one failure in 48 calls is acceptable for the work. We would not choose Ultra on cost or speed alone, as it was the most expensive and the slowest of the three.
Originally published at neuragrowth.co. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.
Top comments (0)