Tips and tricks

venv\Scripts\activate.bat

python -c "import torch; print(torch.__version__); print(torch.version.cuda); print(torch.cuda.is_available()); print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'NO CUDA')"

ComfyUI doesn’t automatically see models from AUTOMATIC1111 (stable-diffusion-webui) unless you tell it to.

✅ Option 1 (BEST) — Reuse your existing models via extra_model_paths.yaml

This avoids duplicating 20GB+ models.

Step 1 — Locate the config file

Go to:

F:\ai\ComfyUI\extra_model_paths.yaml.example

Rename it to:

extra_model_paths.yaml

Step 2 — Edit it

Open it and add your paths:

a111:
  base_path: F:/ai/stable-diffusion-webui/models/
  checkpoints: Stable-diffusion
  vae: VAE
  loras: Lora
  embeddings: embeddings
  controlnet: ControlNet

Step 3 — Restart ComfyUI

python main.py

Now ComfyUI will see your existing models.

python main.py --force-fp32 --enable-manager

Wan 2.1 is a family of video models.

For Wan 2.2 see: Wan 2.2

Files to Download

You will first need:

Text encoder and VAE:

umt5_xxl_fp8_e4m3fn_scaled.safetensors goes in: ComfyUI/models/text_encoders/

wan_2.1_vae.safetensors goes in: ComfyUI/models/vae/

Video Models

The diffusion models can be found here

Note: The fp16 versions are recommended over the bf16 versions as they will give better results.

Quality rank (highest to lowest): fp16 > bf16 > fp8_scaled > fp8_e4m3fn

These files go in: ComfyUI/models/diffusion_models/

These examples use the 16 bit files but you can use the fp8 ones instead if you don’t have enough memory.

Workflows

Text to Video

This workflow requires the wan2.1_t2v_1.3B_fp16.safetensors file (put it in: ComfyUI/models/diffusion_models/). You can also use it with the 14B model.

Workflow in Json format

Image to Video

This workflow requires the wan2.1_i2v_480p_14B_fp16.safetensors file (put it in: ComfyUI/models/diffusion_models/) and clip_vision_h.safetensors which goes in: ComfyUI/models/clip_vision/

Note this example only generates 33 frames at 512×512 because I wanted it to be accessible, the model can do more than that. The 720p model is pretty good if you have the hardware/patience to run it.

Workflow in Json format

The input image can be found on the flux page.

Here’s the same example with the 720p model:

VACE reference Image to Video

This workflow requires the wan2.1_vace_14B_fp16.safetensors file (put it in: ComfyUI/models/diffusion_models/)

This example generates a video from a reference image, this is different from generating a video from a start image. You’ll notice that the video does not actually contain the reference image but is clearly derived from it.

Workflow in Json format

You can find the input image here that image contains a Chroma workflow if you are interested how it was generated.

Image Camera to Video

This workflow requires the wan2.1_fun_camera_v1.1_1.3B_bf16.safetensors file (put it in: ComfyUI/models/diffusion_models/) and clip_vision_h.safetensors which goes in: ComfyUI/models/clip_vision/ if you don’t have it already.

Workflow in Json format

The input image can be found on the flux page.

3. Basic AnimateDiff Workflow (copy this JSON)

Go to ComfyUI → Load → Paste JSON

{
  "last_node_id": 8,
  "last_link_id": 10,
  "nodes": [
    {
      "id": 1,
      "type": "CheckpointLoaderSimple",
      "pos": [100, 100],
      "widgets_values": ["realisticVision.safetensors"]
    },
    {
      "id": 2,
      "type": "CLIPTextEncode",
      "pos": [100, 250],
      "widgets_values": ["a cinematic shot of a fit young woman dancing in a modern gym, realistic lighting, detailed skin texture, 4k"]
    },
    {
      "id": 3,
      "type": "CLIPTextEncode",
      "pos": [100, 350],
      "widgets_values": ["blurry, low quality, deformed, bad anatomy"]
    },
    {
      "id": 4,
      "type": "AnimateDiffLoader",
      "pos": [300, 100],
      "widgets_values": ["mm_sd_v15_v2.ckpt"]
    },
    {
      "id": 5,
      "type": "EmptyLatentImage",
      "pos": [300, 250],
      "widgets_values": [512, 768, 16]
    },
    {
      "id": 6,
      "type": "KSampler",
      "pos": [500, 200],
      "widgets_values": [12345, 20, 6, "euler", "normal", 1]
    },
    {
      "id": 7,
      "type": "VAEDecode",
      "pos": [700, 200]
    },
    {
      "id": 8,
      "type": "SaveAnimatedWEBP",
      "pos": [900, 200],
      "widgets_values": ["AnimateDiff"]
    }
  ],
  "links": [
    [1, 0, 4, 0],
    [1, 1, 2, 0],
    [1, 1, 3, 0],
    [2, 0, 6, 1],
    [3, 0, 6, 2],
    [4, 0, 6, 3],
    [5, 0, 6, 0],
    [6, 0, 7, 0],
    [7, 0, 8, 0]
  ]
}