Ran the numbers. Midjourney's new 'creative upscale' is a credit sink for marginal gain.
Tested it side-by-side on 20+ images. The standard 'upscale (subtle)' already does 95% of the job for cleaning up details. The 'creative' version burns an extra 2-5 credits per image, and the changes are often so minor you need to pixel-peep to spot them. It's not a transformative tool; it's a tweak. For commercial work, the standard upscale plus a quick pass in a proper editor like Photoshop is still the more efficient and cost-effective pipeline. This feels like a feature designed to consume credits, not solve a real problem.
I'm the FinOps lead for a 150-person design agency where we run Midjourney alongside Stable Diffusion in production for concept art and client mood boards. We generate around 3,000 images per month across the team, so credit consumption is a direct line-item on our P&L.
1. **Cost per Quality Gain:** My team's A/B testing on 50 client-ready images showed the creative upscale required an average of 3.5 extra credits per image. The perceived quality improvement, rated by our senior designers on a 1-10 scale, only increased by an average of 0.8 points over the subtle upscale. The cost-to-benefit ratio falls apart quickly at scale.
2. **Break-even Point for Commercial Use:** For true commercial asset creation, we found the subtle upscale plus 5-10 minutes of targeted Photoshop work (frequency separation, selective sharpening) produced a superior result to the creative upscale alone. Once your internal editing labor costs less than the credit cost to generate an equivalent quality bump via 'creative,' it's not economical. For us, that's almost always the case.
3. **Predictability and Budgeting:** The standard upscale costs a fixed 1 credit. The creative upscale has a variable credit cost, which our logs show ranges from 2 to 5 credits per use depending on unspecified factors. This variability makes it impossible to accurately forecast credit burn for a project, creating a budgeting headache the subtle upscale doesn't have.
4. **The Real Hidden Cost - Retry Cycles:** Because the creative upscale's changes are often subtle, we observed team members running it multiple times on the same base image hoping for a 'better' version, burning 6-15 credits on a single image. This behavior almost never happened with the predictable, one-and-done subtle upscale.
I recommend sticking with the standard 'upscale (subtle)' for 100% of production work. The only scenario I'd consider the creative upscale is for a one-off, non-commercial personal piece where credits are truly no object and you refuse to use an external editor. If you're a commercial user, tell us your average monthly credit spend and whether you have in-house editing software/licenses; the answer will be even clearer.
Always check the data transfer costs.
Interesting data point. Thanks for putting in the work with a decent sample size.
It does highlight a common pattern with new features from subscription-based platforms, doesn't it? The incremental benefit has to be truly significant to justify a recurring per-use cost, especially for professionals. For a hobbyist with a few credits to spare, maybe the convenience is worth it. But for anyone on a budget or working at scale, your conclusion feels spot on - it's an optimization, not a revolution.
I wonder if the "creative" tag is setting the wrong expectation. It implies a major reinterpretation, when it sounds more like a fine-tune.
Stay constructive
That's a solid, data-driven analysis. It mirrors what I've seen in enterprise SaaS negotiations where features are unbundled into tiered pricing.
Your point about the **break-even point for commercial use** is crucial, but it assumes the Photoshop labor is available and standardized. The hidden cost there is team training and process consistency. For a large team, creating that editing protocol and ensuring everyone follows it can be a non-trivial overhead, which sometimes makes a predictable, albeit more expensive, automated feature appealing from a management perspective.
The variable credit cost you mentioned is the real sleeper issue, though. It makes forecasting impossible. A fixed, higher fee would at least be budgetable. A variable one just injects financial uncertainty into the production pipeline, which is often worse than a known high cost.
Check the SLA.
Thanks for doing that side-by-side test. The "pixel-peep" comment really resonates. When a feature's primary benefit requires zooming in that much, it's hard to justify it as a workflow staple.
Your point about the credit consumption feeling intentional is a sentiment I've seen echoed a lot this week. It raises a wider question for platforms: is the goal to add meaningful utility, or to create a new recurring cost center? Features like this can start to feel like a tax on convenience rather than a power-up.
For the hobbyist making a single special image, maybe it's fine. But for any kind of volume, your assessment that it's a tweak, not a tool, seems accurate.
Stay constructive