I've been using Leonardo for a few months now, primarily for generating product scene mockups and marketing assets. Over the last 4-6 weeks, I've noticed a tangible inconsistency in output that seems to correlate with their reported user growth.
My main gripe isn't about the occasional bad image—it's about the degradation of *reliable* parameters. I run the same prompts with the same models (like Photoreal) and the same settings to maintain brand consistency. Here's a recent benchmark I logged:
**Prompt (used consistently):** "A modern smartphone sitting on a polished marble surface, next to a small potted succulent, studio lighting, clean background"
* **2 months ago:** 8 out of 10 generations were usable. High detail on the smartphone edges, realistic marble texture.
* **Last week:** 3 out of 10 are usable. Increased frequency of: blurred/misformed phone edges, unnatural plastic-like sheen on the marble, the succulent looking like a vague green blob.
I'm seeing more "overcooked" artifacts, especially in fine details, and a higher rate of generations that just miss the core object composition.
I want to separate anecdote from trend. Before I invest more time in tweaking my workflow or looking at alternatives, I need data.
Has anyone else done systematic comparisons with older outputs?
Specifically:
* Are you experiencing a drop in fidelity with previously reliable models?
* Is there an increase in "default" or generic-looking outputs, even with detailed prompting?
* Could this be server-load related (slower processing leading to quality cuts), or a deliberate change in the underlying model weights?
If you have side-by-side comparisons (same seed, same prompt, different time periods), that's the evidence I'm looking for. Vendor demos always use cherry-picked outputs; real user workflows tell the true story.
Show me the query.