I’m not sure if this is a widespread problem or if I’m doing something that’s specifically causing this bug to happen, but when I set a clip’s transparency to 0 it shows up opaque (same as if I set it to 1). However, any values between 0 and 1 have the expected opacity (or at least it looks like it—I’m not going to pretend I can tell the difference between 0.3 and 0.4 
). To remedy this for now, I’ve just been replacing any 0 values with 0.01 instead and that’s close enough for my purposes. That being said, I thought you should be aware of this since it is not the intended outcome (according to the API docs).
I’ve included JSON below for a video (weird AI content warning again) that demonstrates the problem for the first 31.25 seconds (because render fails if I use the full number of clips with no error so I’m trying to figure out what’s causing that to happen since it’s not file size and appears to be the number of clips).
The JSON is several times longer than the character limit for a forum post, so here’s a link: opacity-zero-is-opaque-ex.json
The JSON looks correct to me, plus it works just fine/as expected as long as I don’t use a zero for the opacity, so I think there’s a bug on your end. This isn’t the first weird behavior I’ve seen with float values but I was careful with the frame rate to ensure any fractional values weren’t excessively long. The only explanation that I can think of that makes sense is that the API is misinterpreting the opacity value as a boolean when it’s set to zero, doesn’t accept it, and defaults to opaque.
Of course, this is kind of a stupid issue to have because why would anyone ever need to set a clip’s opacity to zero since there’s not point in adding it if it can’t be seen. I mean, maybe there is a reason that I can’t think of but I just did it because I was trying to test if I could essentially create an animation via PNG sequence and wanted to verify that/discover any limitations before removing redundancies and not adding clips that can’t be seen. (But on that note, this is a reason to consider increasing the POST request limit above 391kb—which is a very specific number, so I’m curious why you picked it. 
)
Anyway, this may not even really need a fix if there’s never a reason someone would need to have an invisible clip—at least until keyframing for state tweens is a feature—but if this is the common behavior the API docs should reflect that you can’t actually have a zero opacity and should not add the clip if you don’t want it to show up. I haven’t tested clip opacity outside of this scenario, however, so there could be some other reason this is happening. In any case, I figure it’s better if you’re aware of it.