TemporalNet is a ControlNet model designed to enhance the temporal consistency of generated outputs, as demonstrated in this example:
https://twitter.com/CiaraRowles1/status/1637486561917906944. While it does not eliminate all flickering, it significantly reduces it, particularly at higher denoise levels. For optimal results, it is recommended to use TemporalNet in combination with other methods.
-
Add the model "diff_control_sd15_temporalnet_fp16.safetensors" to your models folder in the ControlNet extension in Automatic1111's Web UI.
-
Create a folder that contains:
-
Customize the "temporalvideo.py" script according to your preferences, such as the image resolution, prompt, and control net settings.
-
Launch Automatic1111's Web UI with the --api setting enabled.
-
Execute the Python script.
Also, I highly recommend you use this in conjunction with the hed model, the settings are already in the script.
Write an Extension for the web ui.
Write a feature that automatically generates an "init.png" image if none is provided.
̶C̶h̶a̶n̶g̶e̶ ̶t̶h̶e̶ ̶e̶x̶t̶e̶n̶s̶i̶o̶n̶ ̶t̶o̶ ̶.̶s̶a̶f̶e̶t̶e̶n̶s̶o̶r̶s̶ ̶a̶n̶d̶ ̶i̶n̶v̶e̶s̶t̶i̶g̶a̶t̶e̶ ̶c̶o̶m̶p̶r̶e̶s̶s̶i̶o̶n̶.̶
