The results using the BF16 model with the full text encoder are pretty good.

#10
by fengjunwei - opened

Don't use the lightweight text encoder with this model.
minimax_h3_image_t2i_00066_

You mean this one encoder: qwen3vl_32b_minimax_h3_bf16.safetensors ? What about the diffusion model? Does it also have to be the b16 version to get the effect you show, or Is it enough to just change the bf16 text encoder???

Sign up or log in to comment