Fun stuff, thanks for the video. One small thing, you're getting a cropped video image because you didnt change the width and height in the SVD_img2vid_Conditioning node ^-^ you can do 1024x1024
I tried to replicate your error but wasn't able to, but it seems to me that the image is causing the error. When it gets taken in be the img2vid node it can't read any of the embeddings or data from the image, so I would verify that the image you are using is a compatible type, like png or jpg. If that doesn't work I would re-install the model & nodes.
It's kinda unfair how great your videos are turning out, while I have the same settings and just get those strange messy videos with stuff blurring out all the time, people diffusing like I was using animatediff with a shatter lora 😂
I am sorry to hear that. Different images and subjects will usually need some tweaking with the settings. I recommend just make sure your CFG is relatively low, maybe 2.5. I used the default sampler "Euler", but perhaps try "dpmpp_3m_sde" or any other samplers to help. There are also some controlnet models that can be beneficial such as the "SDXL-controlnet: OpenPose (v2)" to detect poses for any complicated images you are using. If you want us to do a second video with more complex images and fine tuning let us know!
@@compute-grid oh, so you could use sdxl controllers with svd? 🤔 that sounds like a great idea! Well anything that could help me get better results with my rubbish 8gb vram is welcome 😂🙏🏼
Fun stuff, thanks for the video. One small thing, you're getting a cropped video image because you didnt change the width and height in the SVD_img2vid_Conditioning node ^-^ you can do 1024x1024
I keep getting 'NoneType' object has no attribute 'encode_image when it gets to svd img2vid conditioning any thoughts?
I tried to replicate your error but wasn't able to, but it seems to me that the image is causing the error. When it gets taken in be the img2vid node it can't read any of the embeddings or data from the image, so I would verify that the image you are using is a compatible type, like png or jpg. If that doesn't work I would re-install the model & nodes.
@@compute-grid no worries I finally figured it out, thanks
It's kinda unfair how great your videos are turning out, while I have the same settings and just get those strange messy videos with stuff blurring out all the time, people diffusing like I was using animatediff with a shatter lora 😂
I am sorry to hear that. Different images and subjects will usually need some tweaking with the settings. I recommend just make sure your CFG is relatively low, maybe 2.5. I used the default sampler "Euler", but perhaps try "dpmpp_3m_sde" or any other samplers to help. There are also some controlnet models that can be beneficial such as the "SDXL-controlnet: OpenPose (v2)" to detect poses for any complicated images you are using.
If you want us to do a second video with more complex images and fine tuning let us know!
@@compute-grid oh, so you could use sdxl controllers with svd? 🤔 that sounds like a great idea!
Well anything that could help me get better results with my rubbish 8gb vram is welcome 😂🙏🏼