Unhandled exceptions softlocking comfy when attempting to upscale images > 1280px

[BiiirdPrograms]

Whenever attempting to upscale images with any dimension > 1280px, inference fails with the following stack trace:

Exception in thread Thread-14 (prompt_worker):
Traceback (most recent call last):
  File "D:\AI\ComfyUI\execution.py", line 345, in execute
    output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\execution.py", line 220, in get_output_data
    return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\execution.py", line 192, in _map_node_over_list
    process_inputs(input_dict, i)
  File "D:\AI\ComfyUI\execution.py", line 181, in process_inputs
    results.append(getattr(obj, func)(**inputs))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\custom_nodes\ComfyUI-Upscaler-Tensorrt\__init__.py", line 52, in upscaler_tensorrt
    result = upscaler_trt_model.infer({"input": img}, cudaStream)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\custom_nodes\ComfyUI-Upscaler-Tensorrt\trt_utilities.py", line 264, in infer
    raise ValueError("ERROR: inference failed.")
ValueError: ERROR: inference failed.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "C:\Users\Luna\miniconda3\envs\comfy\Lib\threading.py", line 1075, in _bootstrap_inner
    self.run()
  File "C:\Users\Luna\miniconda3\envs\comfy\Lib\threading.py", line 1012, in run
    self._target(*self._args, **self._kwargs)
  File "D:\AI\ComfyUI\main.py", line 183, in prompt_worker
    e.execute(item[2], prompt_id, item[3], item[4])
  File "D:\AI\ComfyUI\execution.py", line 523, in execute
    result, error, ex = execute(self.server, dynamic_prompt, self.caches, node_id, extra_data, executed, prompt_id, execution_list, pending_subgraph_results)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\execution.py", line 412, in execute
    input_data_formatted[name] = [format_value(x) for x in inputs]
                                  ^^^^^^^^^^^^^^^
  File "D:\AI\ComfyUI\execution.py", line 264, in format_value
    return str(x)
           ^^^^^^
  File "D:\AI\ComfyUI\custom_nodes\ComfyUI-Upscaler-Tensorrt\trt_utilities.py", line 271, in __str__
    for binding_idx in range(self.engine.num_bindings):
                             ^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'tensorrt_bindings.tensorrt.ICudaEngine' object has no attribute 'num_bindings'

I added a try-except to trt_utilities.py that at least fixes the soft-lock issue:

def __str__(self):
    out = ""
    try:
        for opt_profile in range(self.engine.num_optimization_profiles):
            for binding_idx in range(self.engine.num_bindings):
                name = self.engine.get_binding_name(binding_idx)
                shape = self.engine.get_profile_shape(opt_profile, name)
                out += f"\t{name} = {shape}\n"
    except:
        out = ""
    return out

This seems to be the exact hard limit, giving it an image with a width or height of 1281px triggers the error, going 1px below doesn't. Is this a limitation of tensorrt, not being able to work with tensors > 5160 wide (4x1280)? Model upscales work well way beyond these dimensions, if slowly. I'm running a 3090 with tensorrt 10.9.0.34, torch 2.7.0+cu126.

[yuvraj108c]

ComfyUI-Upscaler-Tensorrt/__init__.py

Line 100 in 9a178df

engine_min_h, engine_opt_h, engine_max_h = 256, 512, 1280

ComfyUI-Upscaler-Tensorrt/__init__.py

Line 101 in 9a178df

engine_min_w, engine_opt_w, engine_max_w = 256, 512, 1280

You can go beyond 1280 by changing engine_max_w and engine_max_h accordingly above

[BiiirdPrograms]

Thanks, I'll try that.
Might I recommend doing a check on the input image and throwing an exception including a meaningful error message when attempting to upscale images that fall outside of the hard coded range instead of softlocking the current comfy session?
I can create a pr later this week if you'd prefer.

[yuvraj108c]

yeah good idea
A PR would be highly appreciated, thanks!

[BiiirdPrograms]

PR is alive (: