
ComfyUI 17 min read
How to install and set up ComfyUI in 2026
Desktop, portable, manual or cloud. Which one to pick, and how to get your first image.
ResourcesVideo 01
Everything from the video, free: the two workflows it walks through, the prompt that wrote the custom node, and the node's code.
01Image workflow
The image workflow from the video. Nine nodes turn a written prompt into a 1024 × 1024 photograph, and every one of them has a note beside it saying what it does. Every node ships with ComfyUI, so there's nothing extra to install.

| File | Folder | Size | Link |
|---|---|---|---|
flux1-krea-dev_fp8_scaled.safetensorsThe image model. FP8 version of FLUX.1 Krea [dev], 12B parameters. | models/diffusion_models/ | 11.90 GB | Hugging Face (download flux1-krea-dev_fp8_scaled.safetensors) |
clip_l.safetensorsFirst text encoder. | models/text_encoders/ | 246 MB | Hugging Face (download clip_l.safetensors) |
t5xxl_fp16.safetensorsSecond text encoder. Use t5xxl_fp8_e4m3fn.safetensors instead if VRAM is tight. | models/text_encoders/ | 9.79 GB | Hugging Face (download t5xxl_fp16.safetensors) |
ae.safetensorsThe VAE that turns the finished latent into a visible image. | models/vae/ | 335 MB | Hugging Face (download ae.safetensors) |
02Video workflow
The video workflow from the video. It's the same skeleton as the image one with a frame count added, and it makes five seconds of 704 × 1216 video at 24 fps. Every node has a note beside it, and every node ships with ComfyUI.

length counts frames, not seconds: 121 frames at 24 fps is about five seconds.| File | Folder | Size | Link |
|---|---|---|---|
wan2.2_ti2v_5B_fp16.safetensorsThe WAN 2.2 TI2V 5B video model. | models/diffusion_models/ | 10.00 GB | Hugging Face (download wan2.2_ti2v_5B_fp16.safetensors) |
umt5_xxl_fp8_e4m3fn_scaled.safetensorsThe UMT5 text encoder WAN reads prompts with. | models/text_encoders/ | 6.74 GB | Hugging Face (download umt5_xxl_fp8_e4m3fn_scaled.safetensors) |
wan2.2_vae.safetensorsThe VAE that turns generated latent frames into visible ones. | models/vae/ | 1.41 GB | Hugging Face (download wan2.2_vae.safetensors) |
03Node prompt
The exact prompt from the video. It went to Claude Code and came back as one complete, working file. The trick is being specific: the node format, every input, the output and the category are all spelled out.
Create a simple ComfyUI V3 custom node called Image Picker. Requirements: - use comfy_api.latest - 3 IMAGE inputs: image_1, image_2, image_3 - dropdown with Image 1, Image 2, Image 3 - 1 IMAGE output - return whichever image is selected - category: Model Chief - provide one complete __init__.py file
comfy_api.latest keeps the answer on it.04Node code
The complete node from the video. Three image inputs, a dropdown and one output: it passes on whichever image you pick. It's written in ComfyUI's V3 node format, so it needs a recent version of ComfyUI.
from typing_extensions import override
from comfy_api.latest import ComfyExtension, io
class ImagePicker(io.ComfyNode):
@classmethod
def define_schema(cls) -> io.Schema:
return io.Schema(
node_id="ModelChiefImagePicker",
display_name="Image Picker",
category="Model Chief",
inputs=[
io.Image.Input("image_1"),
io.Image.Input("image_2"),
io.Image.Input("image_3"),
io.Combo.Input(
"pick",
options=["Image 1", "Image 2", "Image 3"],
),
],
outputs=[
io.Image.Output(),
],
)
@classmethod
def execute(
cls,
image_1,
image_2,
image_3,
pick,
) -> io.NodeOutput:
images = {
"Image 1": image_1,
"Image 2": image_2,
"Image 3": image_3,
}
return io.NodeOutput(images[pick])
class ModelChiefExtension(ComfyExtension):
@override
async def get_node_list(self) -> list[type[io.ComfyNode]]:
return [
ImagePicker,
]
async def comfy_entrypoint() -> ModelChiefExtension:
return ModelChiefExtension()

custom_nodes folder inside your ComfyUI folder.model_chief_image_picker.__init__.py. Or unzip the folder download straight into custom_nodes.
ComfyUI 17 min read
Desktop, portable, manual or cloud. Which one to pick, and how to get your first image.

ComfyUI 21 min read
One real text-to-image workflow, taken apart in the order the data moves.

ComfyUI 13 min read
One Python file, three image inputs and a dropdown, built line by line.