(Update) benji_Wan_Vace-Native-V2V-CN Long Video (Unlimited Length) Ver 20250710

nope, that's what the AI model design architecture. Using context for referencing, so it need some prompt. I didn't go too detail prompt, still work for me. But for Wan VACE 2.1, there are still reference need to be enhance.

Benjamin Law

is it possible to make the reference image intact? The final looks that being generated in the video are not the one from reference image, unless i make a detailed prompt. But what is the use of reference image if a prompt is needed?

Creative Khim

only using last frame? let see what you have made. motion quality degrade, computer can't handle it, when try it high processing (not only about the VRam) it can do longer. "openpose controlnet sometimes bleeds through " cause you did wrong setting. ;) that's why you can't solve nothing as you already said.

Benjamin Law

I am really noticing that both the image and motion quality degrade after each loop. I feel like the controlnet is degraded after the first loop by a lot and the end motion does not resemble the original. It shows in your preview video that not only does the image quality degrade, but the audio isn't synced with the video. Honestly, I feel like this is no better than using the last frame. At least when using last frame, it at least keeps the motion. I've also noticed that the openpose controlnet sometimes bleeds through into the video, so you have that "stick figure" overlayed. I've also been messing around with both the native and Kijai Nodes for Wan, I've noticed that Kijai Nodes are better at managing memory. It's a really interesting attempt at a looping process, but in the end, it solves nothing.

nsfv nsfv

Thank you for your innovative and nice workflow! I am much more comfortable with video generation! My question is, when I generate a 20 second video, the movement is in slow motion from 10 seconds onwards. Sorry for the sketchy information, but is there any solution? We are using a translator here, so I apologize for any unnatural expressions.

y-ffff

I'm getting an error from cofyui-essentials, unfortunately that repo is in maintenance mode, so you can't create a ticket... custom_nodes\comfyui_essentials\image.py", line 1113, in execute reference = reference.permute([0, 3, 1, 2]).to(device) ^^^^^^^^^^^^^^^^^ AttributeError: 'NoneType' object has no attribute 'permute'

Lars Oeschey

I2v posted in the previous post.

Benjamin Law

Hi, I'd like to use the same workflow using I2V....in short to create longher videos with this neat overlaps. Any suggestions?

Oci vidi Usi slysi

I see Wan as a good solid base, any tips on how to upscale with some kind of extra diffusion?

WillieLonkya

Check your aspect ratio and resolution on input image

WillieLonkya

Al Go

thanks, you are man of culture

Rahul Kumar