Keying the split off upscale_model.input_channels was wrong for any model that does not take 3 channels: a single channel model fed a plain RGB image took only the red channel and returned 1 - green as the alpha mask, where it used to fail loudly. Alpha is channel 3 of an RGBA image, nothing else. |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| compositor_blend_fixture_gen.py | ||
| compositor_blend_golden.json | ||
| compositor_blend_test.py | ||
| compositor_node_test.py | ||
| image_stitch_test.py | ||
| nodes_math_test.py | ||
| nodes_number_convert_test.py | ||
| test_seedvr2_conditioning.py | ||
| test_seedvr2_nodes.py | ||
| test_seedvr2_post_processing.py | ||
| test_seedvr2_temporal_chunk.py | ||
| upscale_model_alpha_test.py | ||