Open Duemellon opened 2 months ago
Not support yet.
As an NLP model(LLM), t5 uses the entire sentence to encode, so t5 can actually fully understand your semantics. I think the weight of a certain word can be strengthened semantically.
The "step" feature is a direct request to the program, not a semantics item, meaning, does ELLA understand when I say "alternate between an image of a dog, a cat, and a bear, on every other step?"
I'll give some phrasing a try for the others.
Do these work?
Such as the "weighted text" person changed to [person:1.1]
And the toggle between [cat|dog] or [blue | red | gold]
or the prompt steps: [x::number] Ignores x after number steps have finished. [x:x:number] Uses the first x until number steps have finished, then uses the second x.