QUESTION: Why on T2V the result will be upscaled 2x, but on I2V the upscaled result is the same as input image?
On ANY workflow text to video: If you choose the video output for example 544x960 The result will be 1088 x 1920 if using the upscaler
But if you use the same output for image to video the workflow Will reduce the image by half and then upscale again to the original size. And the output would be 544x960 or the same as input image
Why on the workflow for text to video you will get an upscaled result, but on the workflow image to video, you will not get an upscale result, more like the same resolution as the original image???
There is a node on IMAGE TO VIDEO that cut the resolution in half
This doesn’t make sense.
Default workflows in ComfyUI - both T2V and I2V - use 2 stages: first make a video in half size then using a spatial upscaler 2x model.
You can disable the 0.5 multiplier and the upscale pass to make a video only with 1 pass.
Or disable the multiplier and make a normal sized video then upscale again 2x, if your PC can handle.
Default workflows in ComfyUI - both T2V and I2V - use 2 stages: first make a video in half size then using a spatial upscaler 2x model.
You can disable the 0.5 multiplier and the upscale pass to make a video only with 1 pass.
Or disable the multiplier and make a normal sized video then upscale again 2x, if your PC can handle.
yep, i tried to disable the node but got OOM right at the end after the 3 steps VAE encode, :(
got oom at the end even with these arguments --reserve-vram 2.0 --lowvram --disable-smart-memory --disable-pinned-memory
Default workflows in ComfyUI - both T2V and I2V - use 2 stages: first make a video in half size then using a spatial upscaler 2x model.
You can disable the 0.5 multiplier and the upscale pass to make a video only with 1 pass.
Or disable the multiplier and make a normal sized video then upscale again 2x, if your PC can handle.yep, i tried to disable the node but got OOM right at the end after the 3 steps VAE encode, :(
got oom at the end even with these arguments --reserve-vram 2.0 --lowvram --disable-smart-memory --disable-pinned-memory
If the OOM occurs at the VAE decode node, you can trey replacing it with the tiled vae decode node instead - it will take longer but that should not cause any OOM's when decoding at least.