Use with mask.
The following error is encountered when using the model with the "Load Advanced Vision Model" with "ReduxAdvanced" node:
!!! Exception during processing !!! The shape of the mask [729] at index 0 does not match the shape of the indexed tensor [1024, 4096] at index 0
Traceback (most recent call last):
File "T:\ComfyUI\execution.py", line 327, 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 "T:\ComfyUI\execution.py", line 202, 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 "T:\ComfyUI\execution.py", line 174, in _map_node_over_list
process_inputs(input_dict, i)
File "T:\ComfyUI\execution.py", line 163, in process_inputs
results.append(getattr(obj, func)(**inputs))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "T:\ComfyUI\custom_nodes\ComfyUI_AdvancedRefluxControl\nodes.py", line 240, in apply_stylemodel
filtered = cond[i][mask[i]]
~~~~~~~^^^^^^^^^
IndexError: The shape of the mask [729] at index 0 does not match the shape of the indexed tensor [1024, 4096] at index 0
The workflow works with the original redux.
How to use mask with this model ?
It looks like the ComfyUI_AdvancedRefluxControl would need to be updated to handle the larger vision encoder.
Hi,
I fixed that. Unfortunately, ComfyUI makes it really hard to just support any model. We have to update the plugins for every new model that comes in... :/
However, I found that the downscaling approach doesn't work with Ostris' redux model. If you use area-downscaling it loses all its power. If you use something else, you get a lot of artifacts like deformed limbs and so on. You can still use the Ostris redux model with attention bias to get a similar effect as downscaling.
I also found that Ostris' redux model is less dominating the prompt. Maybe it is a bit underfitted, but this also means that the prompt following should be already better with this model.