Can I unload WANTE during the denosing?

[high-velocity-rotating-pigeon]

I am using 5060TI 16GB. I noticed that the wan2.2 text encoder only load once in the first run, and then it seems to keep in the vram. Does it results in a loss of usable memory for the diffusion model? If it does, is there a way to unload it once the clip encoding finished?

[high-velocity-rotating-pigeon]

I just found the parameter --lowvram can run text encoder on CPU? So I tried starting with both --lowvram and --use-sage-attention, and got this:
[INFO] got prompt
[INFO] Using pytorch attention in VAE
[INFO] Using pytorch attention in VAE
[INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[INFO] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[INFO] Requested to load WanTEModel
[INFO] Model WanTEModel prepared for dynamic VRAM loading. 10835MB Staged. 0 patches attached. Force pre-loaded 73 weights: 488 KB.
[INFO] Model WanTEModel prepared for dynamic VRAM loading. 10835MB Staged. 0 patches attached. Force pre-loaded 73 weights: 488 KB.
[INFO] Requested to load WanVAE
[INFO] 0 models unloaded.
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
FETCH ComfyRegistry Data: 10/160
[INFO] Found quantization metadata version 1
[INFO] Detected mixed precision quantization
[INFO] Using mixed precision operations
[INFO] Native ops: int8_tensorwise, mxfp8, float8_e5m2, nvfp4, float8_e4m3fn
[INFO] model weight dtype torch.float16, manual cast: torch.float16
[INFO] model_type FLOW
[INFO] Requested to load WAN21
[INFO] 0 models unloaded.
[INFO] Model WAN21 prepared for dynamic VRAM loading. 13630MB Staged. 0 patches attached. Force pre-loaded 160 weights: 1603 KB.
FETCH ComfyRegistry Data: 15/160
0%| | 0/10 [00:00<?, ?it/s, Model Initializing ... ]FETCH ComfyRegistry Data: 20/160
FETCH ComfyRegistry Data: 25/160
FETCH ComfyRegistry Data: 30/160
FETCH ComfyRegistry Data: 35/160
FETCH ComfyRegistry Data: 40/160
FETCH ComfyRegistry Data: 45/160
10%|▉ | 1/10 [00:35<05:15, 35.02s/it, Model Initialization complete! ]FETCH ComfyRegistry Data: 50/160
FETCH ComfyRegistry Data: 55/160
FETCH ComfyRegistry Data: 60/160
FETCH ComfyRegistry Data: 65/160
FETCH ComfyRegistry Data: 70/160
FETCH ComfyRegistry Data: 75/160
FETCH ComfyRegistry Data: 80/160
FETCH ComfyRegistry Data: 85/160
20%|████████▊ | 2/10 [01:23<05:33, 41.73s/it]FETCH ComfyRegistry Data: 90/160
FETCH ComfyRegistry Data: 95/160
FETCH ComfyRegistry Data: 100/160
FETCH ComfyRegistry Data: 105/160
FETCH ComfyRegistry Data: 110/160
FETCH ComfyRegistry Data: 115/160
FETCH ComfyRegistry Data: 120/160
FETCH ComfyRegistry Data: 125/160
30%|█████████████▏ | 3/10 [02:05<04:52, 41.75s/it]FETCH ComfyRegistry Data: 130/160
FETCH ComfyRegistry Data: 135/160
FETCH ComfyRegistry Data: 140/160
FETCH ComfyRegistry Data: 145/160
FETCH ComfyRegistry Data: 150/160
FETCH ComfyRegistry Data: 155/160
FETCH ComfyRegistry Data: 160/160
FETCH ComfyRegistry Data [DONE]
40%|█████████████████▌ | 4/10 [02:46<04:10, 41.79s/it][INFO] [ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[INFO] [ComfyUI-Manager] All startup tasks have been completed.
100%|███████████████████████████████████████████| 10/10 [06:56<00:00, 41.66s/it]
[INFO] Found quantization metadata version 1
[INFO] Detected mixed precision quantization
[INFO] Using mixed precision operations
[INFO] Native ops: int8_tensorwise, mxfp8, float8_e5m2, nvfp4, float8_e4m3fn
[INFO] model weight dtype torch.float16, manual cast: torch.float16
[INFO] model_type FLOW
[INFO] Requested to load WAN21
[INFO] Model WAN21 prepared for dynamic VRAM loading. 13630MB Staged. 0 patches attached. Force pre-loaded 160 weights: 1603 KB.
100%|███████████████████████████████████████████| 10/10 [06:52<00:00, 41.24s/it]
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] Prompt executed in 00:14:18

(Please ignore the Comfy Manager's info)
This is crazy, at least for me, generating a 480x832 5s 16fps video.

[high-velocity-rotating-pigeon]

I know its not the default loading behavior cause I perhaps used --disable-dynamic-vram or --disable-smart-memory in earlier tests, but i do not know how to restore it and i think this result still make sense. Maybe others can reproduce.

[high-velocity-rotating-pigeon]

Also I have found the power usage is different: ~110W with no or single parameter on v.s. ~145W with both parameters on.