Requesting an R256 version of LTX 2.5

#1
by liuxingsyue - opened

The 768 model in LTX 2.5 is too exaggerated. Is there a 256 model? This version of LTX 2.3 is closer to normal ,version 2.5 should be the same.

Yeah ill give it a shot!

Yeah it worked much better cheers, this was not the case with 2.3 i think.

I tested most of the models you made, and only LTX2.3_DISTILLED_BAKED_LTX_SULPHUR_STYLE_IS_10Eros_v12_r256 and LTX2.3_DISTILLED_BAKED_LTX_SULPHUR_STYLE_IS_10Eros_v15_r256 have more realistic effects, models 13 and 14 are messy, and 768 is too exaggerated, LTX2.5 has much stronger fluid and physics simulation capabilities than LTX2.3, I hope you can provide LTX2.5 versions of these two models. Thank you in advance.

Should be much better again now did only STYLE_LAYERS = [
"to_q",
"to_k",
"to_v",
"to_out",
"ff.net.0.proj",
"ff.net.2"
]

This R256 version is good, much more realistic. Version 768 is too exaggerated, liquid is splattered everywhere, like an explosion, and the character is covered in liquid, turning it into a hilarious comedy.

I used a hybrid of Sulphur and 10eros version 768 of v15, which was the most accurate model for expressing the prompts in the LTX 2.3 workflow. Using only 10eros version 768 of v15 resulted in content variation and corruption, but it was still more accurate than R256 and other models. However, in the LTX 2.5 workflow, it produced excessively exaggerated errors.

I found the 768 in two stage followed prompts better

Yeah na... the 256 is better i think I have made it for the dev. However i think really just depends how you want to use it

Sign up or log in to comment