Some hard truths about data sourcing and the future of LTX
Hello, LTX!
Despite the release of the strong model, I'm still rooting for you and for the Western model ecosystem (LTX, Kandinsky, Flux) to have the competition against the high-resource Chinese side.
However, there is a thing you might need to consider.
LTX will not take off without comprehensive anatomical knowledge.
This means the training has to include at least explicit softcore and fetish content.
Without it, no way it can beat Minimax and similar competitors.
Just as an artist studies anatomy drawing nudes, the AI must go through the same path too to grasp it. Not speaking about the appeal to the majority (maybe loud minority, which even more influences the popularity?) of the public users which uses the model.
All the major successful open-weight video-generation models have this knowledge, even from the regions where pornographic materials are prohibited:
Wan has it (China), Hunyuan has it (China), Minimax has it (China), Kandinsky has it (Russia!).
Even in the space of Western Large Language Models the models have it inside them: Gemma 4, newer models such as Laguna, online models like Claude can write erotica in appropriate contexts such as roleplay.
No world comprehension and no "world model" is possible without basic human anatomy and its physics.
Sorry for the "truth nuke". It sucks, but this is a step one has to take to be successful in the AIGC space.