Title: Buffer too small: needs 33554432 bytes, but only has 0 — LTX2_NAG regression in ComfyUI 0.24.1

### Custom Node Testing

- [ ] 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

Generation completes successfully and saves the output video, as it did before the 0.24.1 update.

### Actual Behavior

Diffusion completes (8/8 steps, ~9 minutes 40 seconds) but crashes immediately after in the LTX2_NAG node at video_embeddings_connector with: ValueError: Buffer too small: needs 33554432 bytes, but only has 0. Secondary error: HostBuffer.truncate failed in comfy_aimdo/host_buffer.py.

### Steps to Reproduce

1. Load LTX-V AV model (LTXAV) with LTX2_NAG node from comfyui-kjnodes
2. Set up image-to-video workflow with audio input (singing/music)
3. Settings: 720x1080, 29 seconds duration, 24fps, 8 steps
4. Run the workflow
5. Diffusion completes successfully (8/8 steps, ~9 minutes 40 seconds)
6. Crash occurs immediately after in video_embeddings_connector

### Debug Logs

(** ComfyUI startup time: 2026-06-10 00:04:41.052
[2026-06-10 00:04:41.053] ** Platform: Windows
[2026-06-10 00:04:41.053] ** Python version: 3.12.11 (main, Aug 18 2025, 19:17:54) [MSC v.1944 64 bit (AMD64)]
[2026-06-10 00:04:41.053] ** Python executable: C:\Users\gille\Documents\ComfyUI\.venv\Scripts\python.exe
[2026-06-10 00:04:41.054] ** ComfyUI Path: C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI
[2026-06-10 00:04:41.054] ** ComfyUI Base Folder Path: C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI
[2026-06-10 00:04:41.054] ** User directory: C:\Users\gille\Documents\ComfyUI\user
[2026-06-10 00:04:41.054] ** ComfyUI-Manager config path: C:\Users\gille\Documents\ComfyUI\user\__manager\config.ini
[2026-06-10 00:04:41.055] ** Log path: C:\Users\gille\Documents\ComfyUI\user\comfyui.log
[PRE] ComfyUI-Manager
[2026-06-10 00:04:43.533] C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\cuda\__init__.py:65: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
[2026-06-10 00:04:46.424] Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}
[2026-06-10 00:04:46.424] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}
[2026-06-10 00:04:46.424] Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
[2026-06-10 00:04:46.431] Checkpoint files will always be loaded safely.
[2026-06-10 00:04:46.515] Total VRAM 12287 MB, total RAM 16281 MB
[2026-06-10 00:04:46.515] pytorch version: 2.10.0+cu130
[2026-06-10 00:04:46.517] Set vram state to: NORMAL_VRAM
[2026-06-10 00:04:46.517] Device: cuda:0 NVIDIA GeForce RTX 3060 : cudaMallocAsync
[2026-06-10 00:04:46.528] Using async weight offloading with 2 streams
[2026-06-10 00:04:46.529] Enabled pinned memory 6512.0
[2026-06-10 00:04:46.534] working around nvidia conv3d memory bug.
[2026-06-10 00:04:46.771] Using pytorch attention
[2026-06-10 00:04:49.300] DynamicVRAM support detected and enabled
[2026-06-10 00:04:49.300] Python version: 3.12.11 (main, Aug 18 2025, 19:17:54) [MSC v.1944 64 bit (AMD64)]
[2026-06-10 00:04:49.301] ComfyUI version: 0.24.1
[2026-06-10 00:04:49.301] comfy-aimdo version: 0.4.9
[2026-06-10 00:04:49.302] comfy-kitchen version: 0.2.10
[2026-06-10 00:04:49.306] comfyui-frontend-package version: 1.44.19
[2026-06-10 00:04:49.306] comfyui-workflow-templates version: 0.9.98
[2026-06-10 00:04:49.306] comfyui-embedded-docs version: 0.5.2
[2026-06-10 00:04:49.306] comfy-kitchen version: 0.2.10
[2026-06-10 00:04:49.306] comfy-aimdo version: 0.4.9
[2026-06-10 00:04:49.308] [Prompt Server] web root: C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\comfyui_frontend_package\static
[2026-06-10 00:04:49.310] Asset seeder disabled
[2026-06-10 00:04:49.310] [START] ComfyUI-Manager
[2026-06-10 00:04:49.444] [ComfyUI-Manager] Using GitPython backend
[2026-06-10 00:04:49.499] [ComfyUI-Manager] network_mode: public
[2026-06-10 00:04:49.506] [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\gille\Documents\ComfyUI\.venv\Scripts\python.exe -m pip install matrix-nio

[2026-06-10 00:04:52.268] [Crystools INFO] Crystools version: 1.27.4
[2026-06-10 00:04:52.297] [Crystools INFO] Platform release: 11
[2026-06-10 00:04:52.297] [Crystools INFO] JETSON: Not detected.
[2026-06-10 00:04:52.298] [Crystools INFO] CPU: Intel(R) Core(TM) i5-10600K CPU @ 4.10GHz - Arch: AMD64 - OS: Windows 11
[2026-06-10 00:04:52.310] [Crystools INFO] pynvml (NVIDIA) initialized.
[2026-06-10 00:04:52.310] [Crystools INFO] GPU/s:
[2026-06-10 00:04:52.317] [Crystools INFO] 0) NVIDIA GeForce RTX 3060
[2026-06-10 00:04:52.318] [Crystools INFO] NVIDIA Driver: 596.49
[2026-06-10 00:04:52.359] Base directory: C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip
[2026-06-10 00:04:52.361] Checkpoints path: C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\Wav2Lip\checkpoints
[2026-06-10 00:04:52.361] Model path: C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\Wav2Lip\checkpoints\wav2lip_gan.pth
[2026-06-10 00:04:52.363] Traceback (most recent call last):
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\nodes.py", line 2212, in load_custom_node
module_spec.loader.exec_module(module)
File "", line 999, in exec_module
File "", line 488, in _call_with_frames_removed
File "C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\__init__.py", line 1, in
from .wav2lip import Wav2Lip, LoadAudio
File "C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\wav2lip.py", line 36, in
assert model_exists, f"Model {wav2lip_model_file} not found in {checkpoints_path}"
^^^^^^^^^^^^
AssertionError: Model wav2lip_gan.pth not found in C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\Wav2Lip\checkpoints

[2026-06-10 00:04:52.363] Cannot import C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip module for custom nodes: Model wav2lip_gan.pth not found in C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip\Wav2Lip\checkpoints
[2026-06-10 00:04:52.364] 🎵 [InfiniteTalk Native Sampler] Loaded: InfiniteTalkAutoSampler, InfiniteTalkAutoSamplerAdvanced
[2026-06-10 00:04:52.367]
Import times for custom nodes:
[2026-06-10 00:04:52.367] 0.0 seconds: C:\Users\gille\Documents\ComfyUI\custom_nodes\infinitetalk-native-sampler
[2026-06-10 00:04:52.367] 0.0 seconds (IMPORT FAILED): C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI_wav2lip
[2026-06-10 00:04:52.367] 0.0 seconds: C:\Users\gille\Documents\ComfyUI\custom_nodes\comfyui-frame-interpolation
[2026-06-10 00:04:52.367] 0.0 seconds: C:\Users\gille\Documents\ComfyUI\custom_nodes\comfyui-kjnodes
[2026-06-10 00:04:52.367] 0.1 seconds: C:\Users\gille\Documents\ComfyUI\custom_nodes\ComfyUI-Crystools
[2026-06-10 00:04:52.368] 0.7 seconds: C:\Users\gille\Documents\ComfyUI\custom_nodes\comfyui-clip-interrogator
[2026-06-10 00:04:52.368]
[2026-06-10 00:04:52.377] Context impl SQLiteImpl.
[2026-06-10 00:04:52.378] Will assume non-transactional DDL.
[2026-06-10 00:04:52.494] Using RAM pressure cache.
[2026-06-10 00:04:52.497] Starting server

[2026-06-10 00:04:52.498] To see the GUI go to: http://127.0.0.1:8000
[2026-06-10 00:05:45.515] [DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/buttonGroup.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[2026-06-10 00:05:45.877] [ComfyUI-Manager] The ComfyRegistry cache update is still in progress, so an outdated cache is being used.
[2026-06-10 00:05:45.962] FETCH DATA from: C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\comfyui_manager\custom-node-list.json [DONE]
[2026-06-10 00:05:54.474] got prompt
[2026-06-10 00:05:55.251] VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
[2026-06-10 00:05:55.389] Requested to load AudioVAE
[2026-06-10 00:05:55.710] loaded completely; 693.46 MB loaded, full load: True
[2026-06-10 00:05:57.484] Found quantization metadata version 1
[2026-06-10 00:05:57.484] Detected mixed precision quantization
[2026-06-10 00:05:57.484] Using mixed precision operations
[2026-06-10 00:05:57.484] Native ops: , emulated ops: mxfp8, nvfp4, float8_e4m3fn, float8_e5m2
[2026-06-10 00:05:57.588] model weight dtype torch.bfloat16, manual cast: torch.bfloat16
[2026-06-10 00:05:57.606] model_type FLUX
[2026-06-10 00:05:58.945] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[2026-06-10 00:05:58.949] no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.
[2026-06-10 00:05:58.963] Requested to load VideoVAE
[2026-06-10 00:05:58.984] Model VideoVAE prepared for dynamic VRAM loading. 1384MB Staged. 0 patches attached.
[2026-06-10 00:05:59.828] Found quantization metadata version 1
[2026-06-10 00:05:59.832] Using MixedPrecisionOps for text encoder
[2026-06-10 00:06:00.997] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-06-10 00:06:01.009] Requested to load LTXAVTEModel_
[2026-06-10 00:06:01.473] Model LTXAVTEModel_ prepared for dynamic VRAM loading. 11200MB Staged. 0 patches attached. Force pre-loaded 400 weights: 1749 KB.
[2026-06-10 00:06:08.013] 0 models unloaded.
[2026-06-10 00:06:08.160] Model LTXAVTEModel_ prepared for dynamic VRAM loading. 11200MB Staged. 0 patches attached. Force pre-loaded 400 weights: 1749 KB.
[2026-06-10 00:06:15.421] Requested to load LTXAV
[2026-06-10 00:06:15.423] 0 models unloaded.
[2026-06-10 00:06:16.295] Model LTXAV prepared for dynamic VRAM loading. 23835MB Staged. 1660 patches attached. Force pre-loaded 2104 weights: 3308 KB.
[2026-06-10 00:06:28.702] FETCH ComfyRegistry Data [DONE]
[2026-06-10 00:06:28.961] [ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
[2026-06-10 00:06:29.000] FETCH DATA from: C:\Users\gille\Documents\ComfyUI\user\__manager\cache\1514988643_custom-node-list.json [DONE]
[2026-06-10 00:06:29.371] [ComfyUI-Manager] All startup tasks have been completed.
[2026-06-10 00:15:25.333]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [09:03<00:00, 67.15s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [09:03<00:00, 67.97s/it]
[2026-06-10 00:15:31.613] Model VideoVAE prepared for dynamic VRAM loading. 1384MB Staged. 0 patches attached.
[2026-06-10 00:15:35.943] !!! Exception during processing !!! Buffer too small: needs 33554432 bytes, but only has 0.
[2026-06-10 00:15:36.276] Traceback (most recent call last):
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\execution.py", line 536, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\execution.py", line 336, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\execution.py", line 310, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\execution.py", line 298, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy_api\latest\_io.py", line 1880, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\custom_nodes\comfyui-kjnodes\nodes\ltxv_nodes.py", line 523, in execute
context_video = diffusion_model.video_embeddings_connector(context_video)[0]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ldm\lightricks\embeddings_connector.py", line 297, in forward
hidden_states = block(
^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ldm\lightricks\embeddings_connector.py", line 93, in forward
attn_output = self.attn1(norm_hidden_states, mask=attention_mask, pe=pe)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ldm\lightricks\model.py", line 456, in forward
q = self.to_q(x)
^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 1244, in forward
output = self.forward_comfy_cast_weights(input, compute_dtype, want_requant=isinstance(input, QuantizedTensor))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 1187, in forward_comfy_cast_weights
weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype, want_requant=want_requant)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 321, in cast_bias_weight
offload_stream = cast_modules_with_vbar([s], dtype, device, bias_dtype, non_blocking)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 188, in cast_modules_with_vbar
handle_pin(s, pin, xfer_source, xfer_dest, size=dest_size)
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 186, in handle_pin
cast_maybe_lowvram_patch(source, pin, offload_stream, xfer_dest2=dest)
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\ops.py", line 177, in cast_maybe_lowvram_patch
comfy.model_management.cast_to_gathered(xfer_source, xfer_dest, non_blocking=non_blocking, stream=stream, r2=xfer_dest2)
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\model_management.py", line 1431, in cast_to_gathered
dest_views = comfy.memory_management.interpret_gathered_like(tensors, r) if r is not None else [None] * len(tensors)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\memory_management.py", line 162, in interpret_gathered_like
raise ValueError(f"Buffer too small: needs {offset + size} bytes, but only has {gathered.numel()}. ")
ValueError: Buffer too small: needs 33554432 bytes, but only has 0.

[2026-06-10 00:15:36.285] Prompt executed in 581.78 seconds
[2026-06-10 00:15:36.829] Exception ignored in:
[2026-06-10 00:15:36.831] Traceback (most recent call last):
[2026-06-10 00:15:36.831] File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\model_patcher.py", line 1692, in __del__
[2026-06-10 00:15:36.855] self.unpin_all_weights()
[2026-06-10 00:15:36.855] File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\model_patcher.py", line 1759, in unpin_all_weights
[2026-06-10 00:15:36.856] self.partially_unload_ram(1e32)
[2026-06-10 00:15:36.857] File "C:\Users\gille\ComfyUI-Installs\ComfyUI\ComfyUI\comfy\model_patcher.py", line 1986, in partially_unload_ram
[2026-06-10 00:15:36.858] hostbuf.truncate(offset, do_unregister=module._pin_registered)
[2026-06-10 00:15:36.858] File "C:\Users\gille\Documents\ComfyUI\.venv\Lib\site-packages\comfy_aimdo\host_buffer.py", line 122, in truncate
[2026-06-10 00:15:36.871] raise RuntimeError("HostBuffer.truncate failed")
[2026-06-10 00:15:36.872] RuntimeError: HostBuffer.truncate failed
)
```

### Other

This worked correctly before updating to ComfyUI 0.24.1. The crash is reproducible 100% of the time. The crash location is in ComfyUI core files (memory_management.py, ops.py, model_management.py), not in the custom node itself. System: RTX 3060 12GB, 16GB RAM, Windows 11, PyTorch 2.10.0+cu130, comfy-aimdo 0.4.9.