How this workflow（LTX-2.3_-_I2V_T2V_Basic.json） changes the spoken language
Hello, Master. I've encountered an issue: I wrote my prompt in Chinese, but after passing through the "Enhanced Prompt" node, it was converted into English. How can I resolve this? Thank you.
Change this, but may get incorrect results if the model was trained in English.
You can try bypass the prompt enhancer (it as a toggle switch bottom of workflow) or change the word "English" there. As Sundeveloper pointed out.
The prompt instruction is from LTX themselves by the way.
Its a 12 billion parameter Google Gemma model, so I bet it can understand chinese very well. You prompting in chinese (for sure this if fine), but I am a little unsure if you should let the prompt enhancer just stay on English or use Chinese there as well. You can try. You will see the output result (as in your first post), and can check if the chinese version looks ok.
I am not sure what languages LTX support best for prompting, but they do mention more than english ...
When i am searching for it, i find this info:
LTX (specifically LTX-2 and LTX-2.3) primarily supports English for prompting, but the model has shown capability in handling other languages due to the training data. Users have reported success with prompting in other languages. Based on the provided search results, LTX (specifically LTX-2) does not natively or reliably support Chinese for prompting, although it has some limited, inconsistent multilingual capabilities.
(from Gemini)
But how accurate that info is I am not sure ;-) Just try a few with the "prompt instruction" changed to chinese, and see if the results looks good or not.
thanks, master
Okay, I tested this method, and it indeed produced output in Chinese. It worked quite well—thank you both, masters!