CivArchive
    ← All articles
    Published March 1, 2025by The_frizzy1

    Wan 2.1 Low vram Comfy UI Workflow (GGUF) 4gb Vram

    1.8K views6 reactions1 comments on CivitAI16 collected
    workflow with lorawan2.1quantizationworkflowsupscallerimage generation4gb vrami2vggufhunyuan videot2vlow vram

    This is a workflow i made in comfy ui it can run

    1. Wan2.1 T2V

    2. Wan2.1 I2V

    On just a RTX3050 Laptop edition 4gb Vram

    I am a beginner but here is what I did:

    Used the GGUF custom nodes and models

    Nodes: (or use comfyui manager to install custom nodes)

    https://github.com/city96/ComfyUI-GGUF

    https://github.com/kijai/ComfyUI-WanVideoWrapper

    https://github.com/BlenderNeko/ComfyUI_TiledKSampler

    only models

    Models:https://huggingface.co/calcuis/wan-gguf

    install the vae,clip,clip vision,Text encoder from here

    https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files

    You should also use tools like Xformer,sageatten,triton or settings to speed it up a bit.

    Specs:

    It uses Tiled Ksampler tiled vae decoder everything tiled and about 480p quality but then using an upscaller i can get a pretty good result.

    Now to the results:

    Generation for 53 frames at 480x848 plus an upscaller to 1080p at 25 steps took 5.32 hours

    yes it takes forever and forever however it does not give you a oom error thats good enough for me.

    Notes:

    If you want to do just text use a hunyuan empty latent video I liked these custom nodes for that

    I also like there sampler as you can use it aswell with custom gguf nodes.

    Attachments