A simple I2V workflow that I've been using recently. It features basic up-scaling and frame interpolation - nothing fancy, just designed to be easy to work with.
Keep reading for more details!
It'll zip zap zoop your images into only the highest quality of slop - that is, if you prompt it just right and remember to make the occasional ritual sacrifice to the WAN gods.
Yoink the WAN model from here - get the HIGH AND LOW models, both are required.
Pillage huggingface and CivitAi and grab the VAE and CLIP
Because darksidewalker decided to bless upon us mortals the DaSiWa line of WAN checkpoints, you too can animate your waifu or husbando! Check their profile for new versions and other cool schtuff.
Read the notes scattered around the workflow for useful tidbits of knowledge.
Description
o7
FAQ
Comments (13)
Hey, chief. Some tips for you. (Feel free to peep my workflow for some references.)
- You rgthree Fast Bypasser can use color coding so the first major group does toggle. There's no reasons to toggle it and kind of breaks the point of the toggle.
- You can collapse and hide the "real" negative prompt boxes for NAG support. There's not a great reason to leave them exposed
- In general, I've found RIFE-VFI to be much slower than GIMM-VFI with no appreciable quality difference. You should also check out FILM-VFI, for sake of comparison.
- I'm also a big fan of the rgthree Seed node, for managing the random seed. I'd consider adding it.
- I'd also suggest making the full workflow a little narrower. This could be done with a little reorganizing. Just a personal preference, but I'd urge folks to minimize the need to pan left and right.
Otherwise, seems pretty tidy. Thanks for sharing.
(I know I've already replied the same to other WF but its just for the sake of sharing it here as well) Regarding RIFE, you should check this massive speedup hack that makes GIMM not worth it. https://github.com/Fannovel16/ComfyUI-Frame-Interpolation/pull/102
@rollerinot All good. I appreciate the call out, testing it right now. Will update my workflow if it's as stated!
@rollerinot Hold please, found an issue in my testing.
@rollerinot Okay. I do see the correct speeds now:
GIMM : 23.5 24.5
FILM : 39.8 42.9
RIFE49 : 6.8 x
RIFE47 : 7.5 x
(rife patch applied above)
RIFE49 : x 39.3
RIFE47 : x 44.7
(no rife patch)
So yes, you are correct! There are definitely motion artifacts in the videos I tested compared to GIMM, so it's not a crystal clear "Always RIFE" for me, but it will make for a nice performance option. I'll consider how it might fit in my workflow.
Thanks for the suggestion, I'll yoink it from your workflow and stitch it in, test it out and ship it :)
Also thanks for the suggestion on collapsing things - I knew how to do this, but I'm kind of a tinker-with-everything kind of person so it doesn't always occur to me to make things smaller or obfuscated.
As for scrolling, I'm personally a fan of left-to-right for workflows, but I see no reason to not post one of each orientation.
Added a new version that has a vertical layout, and includes GIMM-VFI, but doesn't make use of it. I found it actually slowed things down substantially (from ~80s to ~120s) - but I am using a pretty fragile manually compiled python environment under Debian Trixie, so... probably ghosts in the machine :)
@PassiveImp Yeah, it's machine dependant. I think GIMM requires fairly recent CUDA. (Not sure what the minimum is.) It's been faster on the cards I've been renting, but like most of is this stuff, it's not universal.
@boobkake22 👀 I’m using a 5090
@PassiveImp lol. Yeah, should have no problem then, but it could certainly be an installation issue. It's been made clear to me that getting the right set of things installed can be a bear.
I’m using the same checkpoint you recommended, but the workflow still asks for a VAE.
I also noticed the VAE node isn’t connected—just wondering if that’s normal.
The non-gguf models usually have VAE packed into the checkpoint, so it's normally routed from the load checkpoint node. If you're using the GGUF versions, or a checkpoint that doesn't have the VAE baked in, then you'll need to download and load the VAE separately, then connect it up.
The other stuff is slop. Except for the furry one, all of the Missionary POV stuff you've uploaded is great.