The strength of an encryption algorithm is not how noisy the image is, but how hard it is to reverse the process without knowing the key.

If you have a 32-bit seed, you only have 232 different possible random number sequences to evaluate. If you know or assume, for example, that the image is a sprite sheet with black background, you can just decrypt the first few pixels and see if they are all black. If they are, you can assume you have found a key that can decrypt the whole image and proceed to decrypt the entire image.

Sounds like an extremely weak cryptographic algorithm if you can make just a tiny assumption about the data it contains.

Actually, now when I think about it, your algorithm is much worse than that. Since you only encrypt a byte at a time, and the sequence of random numbers is fully known, you effectively only have a 8-bit XOR cipher. Not only is brute force viable, but even manual brute force BY HAND is viable since you effectively only have an 8-bit key and thus 256 combinations to manually examine.

Your algorithm to generate the random numbers is trivially known. Not only did you say which one you use in your first post, but one can just examine your code and figure it out.

Making several passes with an XOR algorithm is useless. If you make an XOR pass with key A, and then second XOR pass with key B, and a third XOR pass with key C, and so on for how many times you want, then that would be the same as a single pass with the key (A XOR B XOR C XOR... ). No matter how many XOR passes you do, you can decrypt, and consequently crack it, it with a single XOR pass.

If you use different passwords for the different passes, then that would be equivalent to having a single pass with a longer password (or rather, each additional password adds another 32-bits to the seed; not that the combined passwords would generate a stronger single 32-bit initial seed). But, that also multiple passwords.

If you have different keys based on different passwords, then yes, that is what I mentioned in the paragraph below the one you quoted. But my quote still remains a valid concern in that you can combine the keys and decrypt once, so you still have another point of attack; you can focus on the random sequences instead of, or as well as, on the data.

But your additional passes based on the individual letters are not very strong. If you have a text password, each letter only contributes with roughly 6 bits of entropy assuming upper and lower letters and some common symbols. In your case, 8 XOR passes with a 6-bit key is not a lot.

I think i see what you mean. so 6 bits * 8 chars in this case would give 48 bit of entropy for the key?

Effectively, yes. You also have to remember that you lose entropy because of the fact that the first 32-bit key uses the same password source as the subsequent passes. In fact, it may even render the initial pass useless if I'm thinking correctly about this, since, given a set of 8-bit keys, there can only be one initial 32-bit key.