NVIDIA RTX 3080 10GB I2V Workflow - Rollerinot v1

WARNING *28/11/2025: WanVideoBlockSwap has been deprecated by Comfyui latest updates and it is not anymore available from the custom model manager. The new alternative is a new node with the same name but is not yet compatible with the Lora stacker by rgthree and WanMoeKsampler (the connection is not compatible)

If you had the node previously downloaded there is a workaround. Go to ComfyUI_windows_portable\ComfyUI\comfy_extras\ and delete or put into a backup folder the file nodes_nop.py and when you restart it should work.

18/12/2025: ComfyUI now has a new memory management natively integrated. Just bypass the WanBlockSwap node, it works better now (Reminder: you must have plenty of system ram available)

This is my Workflow for the users who want to generate WAN video with the 10GB version of the RTX3080.

Optimizations:

With Wanvideoblockswap it is possible to run Q6 GGUF base model Now natively integrated in ComfyUI (update it)

Pre stacked accelerator lora loader: feel free to use lightning or any other

SageAttention

RIFE & GMFSS interpolation options. The latter is better for cartoon interpolations. *Apply this speedup hack for RIFE: KSampler

Image iterator and automatic resizer for batch processing*

*Needs manual installation: Here is the python file: post by the author here: this settings It is the maximum quality to ram usage ratio I achieved with 32GB of system ram. I did not try all the gguf base models available like RAPID or FASTMIX. Feel free to try and give feedback.

a 3KSampler WF will be available soon