New model versions ship often enough now that I lost track of which tool handled which specific task well, relying on vague memory of a demo I saw weeks earlier rather than actual tested knowledge. I started keeping a simple personal reference document a while back, and it's become one of the more genuinely useful pieces of my workflow, unglamorous as it sounds.
What actually goes in the document
Not a general review of each model, that information is available everywhere and goes stale fast. Specific, task-oriented notes: which model handles hands convincingly, which one respects detailed camera movement instructions, which one produces skin tones I don't have to fight in the grade afterward. Narrow, practical entries I can scan in ten seconds mid-project rather than a long essay I'd never actually reread.
Why memory alone wasn't working
Models update often enough that a note from three months ago might already be stale, and I've started dating every entry and revisiting it whenever I use a tool again rather than trusting an old note indefinitely. That habit alone has caught a few cases where a model genuinely improved at something I'd written off, and a couple of cases where a model I remembered fondly had actually been surpassed by something newer without me noticing.
- Task-specific notes, not general reviews, scannable in seconds
- Date every entry and revisit rather than trusting old notes indefinitely
- Include failure notes, not just wins, they're often more useful later
The unexpected benefit
Writing these notes down forces a level of specificity I wasn't applying when I was just forming vague impressions in my head. "This model is good at hands" is a much weaker and less useful note than "this model handles a hand holding a small object convincingly at close range but still struggles with two hands interacting." The act of writing the sharper version has made me a more precise evaluator of these tools generally, not just a better-organized one.