Torsten's Wan 2.2 14B i2V Low-VRAM Workflow with Added Features - v1.0 Showcase
Torsten's first publicly-released Workflow!
Drag and Drop the image below into ComfyUI to access the workflow. This is also fully capable of NSFW generation!
Using my RTX 5060Ti (16GB) GPU, I have been testing a handful of Image-To-Video workflow methods with Wan2.2. Mainly using this workflow as a base, found in AIdea Lab's video (show your support, give him a like and subscribe) I was able to simplify some of the process while adding a couple extra features. Remember to use Wan2.1 VAE with the Wan2.2 i2v 14B Quantization models! You can drag and drop the embedded image into your ComfyUI to load the Workflow Metadata. This uses a few types of Custom Nodes that you may have to install using your Comfy Manager.
Simplified Processes
Who needs a complicated flow anyway? Work smarter, not harder. You can add Sage-ATTN and Model Block Swapping if you would like, but that had a negative impact on the quality and prompt adherence in my testing. Wan2.2 is efficient and advanced enough where even Low-VRAM PCs like mine can run a Quantized Model on its own with very little intervention from other N.A.G.s
Added Optional Features - LoRa Support and RIFE VFI
This workflow adds LoRa model-only loaders in a wrap-around sequential order. You can add up to a total of 4 LoRa models (backward compatible with tons of Wan2.1 Video LoRa). Load up to 4 for High-Noise and the same 4 in the same order for Low-Noise. Depending what LoRa is loaded, you may experience "LoRa Key Not Loaded" errors. This could mean that the LoRa you loaded is not backward-compatible for the new Wan2.2 model, or that the LoRa models were added incorrectly to either High-Noise or Low-Noise section.
The workflow also has an optional RIFE 47/49 Video Frame Interpolation node with an additional Video Combine Node to save the interpolated output. This only adds approximately 1 minute to the entire render process for a 2x or 4x interpolation. You can increase the multiplier value several times (8x for example) if you want to add more frames which could be useful for slow-motion. Just be mindful that more VFI could produce more artifacts and/or compression banding, so you may want to follow-up with a separate video upscale workflow afterwards.
TL;DR
It's a great workflow, some have said it's the best they've ever seen. I didn't say that, but other people have. I think it's the best workflow anyone has ever uploaded. You know what we need on this platform? We need to Make Workflows Great Again!