As another user said below, the biggest problem is making sure you're using a good prng. There are really simple algorithms for generating "ok" random numbers, that's the kind of thing you might get when you use a rand() function in a programming language. Those rand functions are plenty good enough for deciding dice rolls in a video game, but they aren't robust enough for cryptography.

Outside of that, you're correct that the next problem is the seed, and how much total entropy is in that seed. In this context, entropy means the total search space for the seed ("hello" would be a low entropy seed, "RPSTl6HF7mPo3g" would be a much higher entropy seed).

Like you suggested, sometimes your seed is a long timestamp combined with other factors, that makes for a great high entropy seed! But sometimes, like with a Bitcoin key, you want your seed to be recreatable, should you lose it. That's when you tend to use a password as a seed. So the way that actually works, is that your key is a long string of binary bits, say 256 bits, and that string of bits is generated using your password as a starting seed. This process is the hashing algorithm. And this is where there tends to be a problem. Your hashing algorithm may produce 256 bit keys (high entropy), but whether it's actually secure will depend on whether your seed phrase (password) is high entropy. You can create a key based on anything, it could be (should be) a phrase with more entropy than the key itself, for instance "my special long password with 1 secret word Brathrok" that's a very long seed and it includes a word that won't be in a dictionary, that'll be hard to brute force, but easy to remember. Or... it could be a single word with low entropy, like "fool", which is a good way to lose your bitcoin, so please don't do that.