Terribly slow speeds with the TextEncodeAceStepAudio1.5 node

### Custom Node Testing

- [x] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

TextEncodeAceStepAudio1.5 uses regular ram and takes forever each time lyrics has changed
6GB vram used and 20GB ram used, tried a fresh install of comfyui, no arguments, no extra custom nodes, disabled offload to ram in nvidia settings..

Other stuff like Flux generate in breeze, nothing touches RAM
I have 3090

### Actual Behavior

.

### Steps to Reproduce

1.Fresh install of Comfyui, Windows11, 3090
2.Very slow each generation

### Debug Logs

```powershell
G:\AI-Apps\Comfy_Ace_test>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Checkpoint files will always be loaded safely.
WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.
Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
Total VRAM 24564 MB, total RAM 65303 MB
pytorch version: 2.10.0+cu128
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3090 Ti : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 29386.0
working around nvidia conv3d memory bug.
Using pytorch attention
Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
ComfyUI version: 0.12.2
ComfyUI frontend version: 1.37.11
[Prompt Server] web root: G:\AI-Apps\Comfy_Ace_test\python_embeded\Lib\site-packages\comfyui_frontend_package\static

Import times for custom nodes:
0.0 seconds: G:\AI-Apps\Comfy_Ace_test\ComfyUI\custom_nodes\websocket_image_save.py

Context impl SQLiteImpl.
Will assume non-transactional DDL.
Assets scan(roots=['models']) completed in 0.018s (created=0, skipped_existing=22, orphans_pruned=0, total_seen=22)
Starting server

To see the GUI go to: http://127.0.0.1:8188
got prompt
model weight dtype torch.bfloat16, manual cast: None
model_type FLOW
VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
Requested to load ACE15TEModel_
loaded completely; 4673.04 MB loaded, full load: True
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
Requested to load ACEStep15
loaded completely; 16946.63 MB usable, 4565.35 MB loaded, full load: True
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:03<00:00, 2.55it/s]
Requested to load AudioOobleckVAE
loaded completely; 321.70 MB loaded, full load: True
Prompt executed in 201.35 seconds
```

### Other

_No response_

Reproduce the TextEncodeAceStepAudio1.5 node with the fresh Windows 11 setup described, using the main.py command and debug logs as the starting point. Compare the reported 201.35-second prompt execution, VRAM use, and RAM use when lyrics change; done means the cause is identified and the node no longer incurs the reported excessive delay and memory usage.

Written by the indexing model from the issue text.