Save Image / Preview Image (Node 2.0) stops updating after opening & closing a new empty workflow tab (UI shows stale image)

### 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

After each run, Save Image / Preview Image nodes should refresh and display the latest output image.

### Actual Behavior

After creating and closing an empty workflow/tab, subsequent runs do not refresh Save Image / Preview Image node preview thumbnails; they remain stuck on an older image.

### Steps to Reproduce

1. Start ComfyUI normally. (I tried ComfyUI Desktop / Windows Portable / Manually Build This Github Python Project)

2. Load a simple workflow:

Load Checkpoint → (CLIP Text Encode pos/neg) → KSampler → VAE Decode → Save Image
(Optionally also test with Preview Image node.)

3. Run once → node preview updates correctly (shows Image A). Screenshot 1 attached (normal behavior).

4. [IMPORTANT] Create a new empty workflow/tab (blank canvas), then close that empty workflow/tab. Screenshot 2 attached shows the intermediate “new empty workflow created” state.

5. Go back to the original workflow and run again (different seed / run again) → a new image is generated (Image B).

6. Observe that Save Image / Preview Image node preview thumbnail still shows Image A (stale) and does not refresh. Screenshot 3 attached (stuck preview).

### Debug Logs

```powershell
C:\Users\xayah\Desktop\ComfyUI\.venv\Scripts\python.exe C:\Users\xayah\Desktop\ComfyUI\main.py --enable-manager
[START] Security scan
[DONE] Security scan
** ComfyUI startup time: 2026-02-13 13:14:53.643
** Platform: Windows
** Python version: 3.13.5 (tags/v3.13.5:6cb20a2, Jun 11 2025, 16:15:46) [MSC v.1943 64 bit (AMD64)]
** Python executable: C:\Users\xayah\Desktop\ComfyUI\.venv\Scripts\python.exe
** ComfyUI Path: C:\Users\xayah\Desktop\ComfyUI
** ComfyUI Base Folder Path: C:\Users\xayah\Desktop\ComfyUI
** User directory: C:\Users\xayah\Desktop\ComfyUI\user
** ComfyUI-Manager config path: C:\Users\xayah\Desktop\ComfyUI\user\__manager\config.ini
** Log path: C:\Users\xayah\Desktop\ComfyUI\user\comfyui.log
[PRE] ComfyUI-Manager
Checkpoint files will always be loaded safely.
Found comfy_kitchen backend cuda: {'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 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': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Total VRAM 32607 MB, total RAM 128568 MB
pytorch version: 2.10.0+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 57855.0
working around nvidia conv3d memory bug.
Using pytorch attention
Python version: 3.13.5 (tags/v3.13.5:6cb20a2, Jun 11 2025, 16:15:46) [MSC v.1943 64 bit (AMD64)]
ComfyUI version: 0.13.0
ComfyUI frontend version: 1.38.13
[Prompt Server] web root: C:\Users\xayah\Desktop\ComfyUI\.venv\Lib\site-packages\comfyui_frontend_package\static
[START] ComfyUI-Manager
[ComfyUI-Manager] network_mode: public
[ComfyUI-Manager] The matrix sharing feature has been disabled because the `matrix-nio` dependency is not installed.
To use this feature, please run the following command:
C:\Users\xayah\Desktop\ComfyUI\.venv\Scripts\python.exe -m pip install matrix-nio

Import times for custom nodes:
0.0 seconds: C:\Users\xayah\Desktop\ComfyUI\custom_nodes\websocket_image_save.py
0.0 seconds: C:\Users\xayah\Desktop\ComfyUI\custom_nodes\comfyui-promptdrafter

Context impl SQLiteImpl.
Will assume non-transactional DDL.
[PromptDrafter] Loaded v1.0.0
Assets scan(roots=['models']) completed in 0.014s (created=0, skipped_existing=19, orphans_pruned=0, total_seen=19)
Starting server

To see the GUI go to: http://127.0.0.1:8188
[ComfyUI-Manager] The ComfyRegistry cache update is still in progress, so an outdated cache is being used.
FETCH DATA from: C:\Users\xayah\Desktop\ComfyUI\.venv\Lib\site-packages\comfyui_manager\custom-node-list.json [DONE]
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load SDXLClipModel
loaded completely; 29407.80 MB usable, 1560.80 MB loaded, full load: True
Requested to load SDXL
loaded completely; 27713.87 MB usable, 4897.05 MB loaded, full load: True
100%|██████████| 20/20 [00:03<00:00, 6.29it/s]
Requested to load AutoencoderKL
loaded completely; 17099.98 MB usable, 159.56 MB loaded, full load: True
Prompt executed in 7.89 seconds
got prompt
100%|██████████| 20/20 [00:02<00:00, 6.95it/s]
Prompt executed in 3.30 seconds

Stopped server
Cannot connect to comfyregistry.
FETCH DATA from: C:\Users\xayah\Desktop\ComfyUI\user\__manager\cache\1514988643_custom-node-list.json [DONE]
[ComfyUI-Manager] All startup tasks have been completed.

Process finished with exit code 0
```

### Other

Not only `Save Image` and `Preview Image`, It seems `Load Images (From Output)` (Node 2.0) also has similar bugs. Please help fix this.

Start by locating the Node 2.0 frontend implementations for Save Image, Preview Image, and Load Images (From Output), then reproduce the stale-thumbnail behavior using the listed workflow and empty-tab sequence. Trace how previews update after each run and after a tab is created and closed. Done means the latest output appears in all affected node previews after the reproduction steps.

Written by the indexing model from the issue text.