@jboogx_creative you're awesome!

Pretty sure "Save Image" is the standard node, but "Image Save" sounds like WAS node suite, which I use to save in different formats and control the filenames.

@albamuth It is! You're right! Sadly, I'm still missing:

Apply ControlNet Stack

IPAdapterModelLoader

Control Net Stacker

ImageDuplicator

IPAdapterApply

@albamuth Installed "WAS node Suite" and the "Images Save" error message went away. Thanks!

Error(s) in loading state_dict for Resampler: size mismatch for proj_in.weight: copying a param with shape torch.Size([768, 1280]) from checkpoint, the shape in current model is torch.Size([768, 1024]).

Ip adapter problems idk

if i cut them off it worked with some warnings

@poweraipower0957 You will need to have everything up to date. It works for me flawlessly

@poweraipower0957 Check in Custom-Nodes There are mostlikely two IP-Adapter folders, copy the models in both folders that resolved this issue for me

@hwangbill0645 still dont work idk

this isnt a broken workflow... just download the missing nodes. This is honestly one of the best comfy workflows out there right now. The creator isnt holding anything back

@syntaxdiffusion for me it is idk, i reinstalled the custom nodes, created a ipadapter dict in comfyui, i tryed all ip adapter models that are available... tryed about 7 8 different ckpts or safetensors i still got ipadapter apply error xd

@poweraipower0957


I have these:


cubiq 
ComfyUI_IPAdapter_plus

ComfyUI reference implementation for IPAdapter models. The code is mostly taken from the original IPAdapter repository and laksjdjf's implementation, all credit goes to them. I just made the extension closer to ComfyUI philosophy.

laksjdjf
 IPAdapter-ComfyUI

Nodes:Load IPAdapter. This custom nodes provides loader of the IP-Adapter model.

NOTE: To use this extension node, you need to download the ip-adapter_sd15.bin file and place it in the custom_nodes/IPAdapter-ComfyUI/models directory. Additionally, you need to download the 'Clip vision model' from the 'Install models' menu as well.

They can be found through the manager typing IPAdapter and you ll see them in the custom node folder

I just don't know which one worked, but they do not interfere with eachother

it did. Thank you!

@jboogx_creative Please, don't be sorry. I learn a lot using your workflow, it's really helpful. I'm just switching to comfyui, so spending more time reading docs and readmes is not bad for me ! :)

AnimateDiff produces uses / produces batches of latents which is the problem, when you process a batch through ksampler every image is running in parallel so the maximum is going to depend on vram and size (and probably not work at all in directml).

I haven't tried this but it should probably work.
The built in node latents->batch->rebatch latents can split them up and process in whatever batch size is optimal or just one at a time. I would do one at a time in your case (the maximum you can do before OOM probably isn't the fastest way to go, but it depends on the card and image size, and there's still VAE decode). Stick one of these before the latent upscaler in that section so there aren't upscaled latents sitting around eating memory before they're being processed.

If the KSampler in that group relies on an internal random seed that isn't supposed to be changing across the full animation length and was relying on the fact that it was processing a batch to stay constant, it'll need to be set externally so it doesn't change after each image.

The problem is if you're hitting OOM during the ksampler, doing them one-by-one will probably get you through the VAE decode but you might still need to save an image sequence instead of a video depending on how the video output is implemented.

Edit: Fixed node -> group in the ksampler / seed info.

@jboogx_creative i dont know never use the prompt scheduler before , change it clip text enc it work

@lihaoyunlu930 I avoided the 'batch prompt scheduler' entirely, used a regular ClipTextEnc as a positive prompt and it fixed the 'index out of range' error.