How to design a REST API - api

Question

I am building a metadata platform for a broadcaster. It exposes a REST API to write data to the DB.
I struggle with the design of the endpoint boundaries. There is a wide range of options to split the API controller model so that the API producers can write data in either normalized or denormalized form. I am not sure what a widely accepted design is so that the platform provides the best possible developer experience.
I have modeled both extremes with a PUT request for updates. Which one is more usable?
Normalised
PUT /episodes
{
"season" : 1,
"episodeInSeason" : 2,
"texts" : [
"#ref TextId"
],
"variants" : [
"#ref VariantId"
]
}
PUT /variants
{
"texts" : [
"#ref TextId"
],
"assets" : [
"#ref AssetId"
]
}
PUT /assets
{
"texts" : [
"#ref TextId"
],
"renditions" : [{
"fps": 60.0
}],
"markIn" : "#ref MarkerId",
"markOut" : "#ref MarkerId"
}
PUT /texts
{
"title": "foo",
"lang" : "en"
}
PUT /renditions
{
"fps": 60.0
}
PUT /markers
{
"position" : 2.2
}
Denormalised
PUT /episodes
{
"season" : 1,
"episodeInSeason" : 2,
"texts" : [{
"title": "foo",
"lang" : "en"
}],
"variants" : [{
"texts" : [{
"title": "foo",
"lang" : "en"
}],
"assets" : [{
"texts" : [{
"title": "foo",
"lang" : "en"
}],
"renditions" : [{
"fps": 60.0
}],
"markIn" : {
"position" : 1.0
},
"markOut" : {
"position" : 2.2
}
}]
}]
}