ComfyUI Mochi Edit Enhance Local AI Video Generated Result With Unsampling

[@VladislavKusmin]

nice, just tried it yesterday

[@TheEconoVision]

So CogvideoX or Mochi 1 , which one you see is better? In terms of running in local.

[@davidblenkinsopp9597]

Thanks for this, this is amazing. While the second image is amazing it looks like the contrast has been turned up to 11 - is there any way of toning this down?

[@TheFutureThinker]

Hi, yes we can adjust. Because the node in Unsampling and Resampling were using default values for most of the setting. We can lower the steps or CFG in Resampling, and it lower the contrast. Another way to do is let it run as it is, and we add another node after VAE Decoding for the images output, to do auto contrast. I remember one called Auto Layer custom node which cover most of the color tuning and layering features.

[@crazyleafdesignweb]

So the Mochi Edit , in the Github page mentioned. A similar strategy as RF-Inversion

[@TheFutureThinker]

Yes. Unsampling RF Inversion. There are Comfyui custom node for this as well.

[@Guus]

Yoo. Do you think proper controllnets will be on the way soon?

[@TheFutureThinker]

I think the road map for an AI Video model to become a solid model. Stage 1- Txt2vid, 2- img2vid , 3- start /end frames , 4- then controlnet for pose or v2v Runway Gen3 as a good example.

[@kalakala4803]

Awesome thanks! Will try it in my PC. My 4070 can generate 6.8 seconds txt2vid 😉

[@TheFutureThinker]

😎👍

[@skyebrows]

Now if you can figure out how to un deep fry the image. Maybe if someone makes a lora that understands contrast.

[@TheFutureThinker]

Lower the CFG and see