SUCCESS" Hunyuan with TEACACHE from 90s/it to 17s/it making Halo Strix and Hunyuan 1.5 fly. What you need to do.

[bkpaine1]

Here are the two fixes I made to
/home/brent/ComfyUI/custom_nodes/teacache/nodes.py:

---

Fix 1: Remove broken import (line 12)

Changed:
from comfy.ldm.flux.layers import timestep_embedding, apply_mod
from comfy.ldm.lightricks.model import precompute_freqs_cis
from comfy.ldm.lightricks.symmetric_patchifier import latent_to_pixel_coords

To:
from comfy.ldm.flux.layers import timestep_embedding, apply_mod
from comfy.ldm.lightricks.symmetric_patchifier import latent_to_pixel_coords

And updated the call around line 631 from:
pe = precompute_freqs_cis(fractional_coords, dim=self.inner_dim,
out_dtype=x.dtype)

To:
pe = self._precompute_freqs_cis(
fractional_coords,
dim=self.inner_dim,
out_dtype=x.dtype,
max_pos=self.positional_embedding_max_pos,
use_middle_indices_grid=self.use_middle_indices_grid,
num_attention_heads=self.num_attention_heads,
)

---

Fix 2: Update Hunyuan Video function signature for 1.5 compatibility

The teacache_hunyuanvideo_forward function signature was outdated. Changed
from:
def teacache_hunyuanvideo_forward(
self,
img: Tensor,
img_ids: Tensor,
txt: Tensor,
txt_ids: Tensor,
txt_mask: Tensor,
timesteps: Tensor,
y: Tensor,
guidance: Tensor = None,
guiding_frame_index=None,
ref_latent=None,
control=None,
transformer_options={},
) -> Tensor:

To:
def teacache_hunyuanvideo_forward(
self,
img: Tensor,
img_ids: Tensor,
txt: Tensor,
txt_ids: Tensor,
txt_mask: Tensor,
timesteps: Tensor,
y: Tensor = None,
txt_byt5=None,
clip_fea=None,
guidance: Tensor = None,
guiding_frame_index=None,
ref_latent=None,
disable_time_r=False,
control=None,
transformer_options={},
) -> Tensor:

Plus updated the function body to handle the new Hunyuan 1.5 features
(time_r_in, byt5_in, vision_in, cond_type_embedding) and pass
transformer_options to block calls.