Inputs and outputs

iNodes only reads what you expose in ComfyUI's app builder. This page describes what the bridge accepts today: it evolves with it.

Settings

• Prompt: the text field of a CLIP Text Encode, or a field named prompt.
• Seed, whole and decimal numbers: numeric settings.
• Switches: checkboxes.
• Choice lists, such as the sampler or a format: drop-down menus.

Images

An IMAGE socket, or the Load Image and Load Image (as Mask) nodes. Without a specific label, their order in the builder sets their role:

1. the first image becomes the Source input;
2. the second becomes Reference;
3. the next ones become Reference 2, Reference 3…

A label containing “mask” or “reference” sets the image's role. For an edit or reference workflow, order the images the way the model expects them.

Videos

A VIDEO socket, or the Load Video, VHS Load Video and VHS Load Video Path nodes. The first video becomes the Source input, the second Source 2. With a video, images become references.

What's left out

• text fields other than the prompt;
• MASK and AUDIO sockets, and types not listed on this page;
• file lists from the ComfyUI machine, such as media;
• model selectors (checkpoint, UNet, VAE, CLIP, LoRA);
• other video load nodes.

iNodes lists them for you at conversion. The workflow keeps the values saved in the file for them.

The output

One output node per workflow. Its type decides what your node produces:

• an image, with Save Image or Preview Image;
• a video, with a node whose name contains “video”;
• a 3D model, with a mesh or 3D node, Save GLB or Save OBJ. Like the other 3D models, it works in Blender and Houdini.

Technical reference

iNodes recognizes a node by its technical name: the one in the node's properties in ComfyUI (Node name for S&R), and the class_type in the API file.

• Accepted: the input becomes a setting or an input of your node.
• Ignored: the workflow still converts, and the input keeps the value saved in the file.
• Refused: the conversion stops, with a message.

Accepted inputs

You expose	Technical name	In your node
The prompt	text of CLIPTextEncode or TextEncode, value of PrimitiveString(Multiline), or a prompt field	Prompt field
The seed	a seed field	Seed setting
A number	INT, FLOAT	Numeric setting
A switch	BOOLEAN	Checkbox
A choice list	COMBO (sampler, scheduler…)	Menu, 64 choices at most
An image	IMAGE socket, LoadImage, LoadImageMask	Source input, then Reference…
A video	VIDEO socket, LoadVideo, VHS_LoadVideo, VHS_LoadVideoPath	Source input, then Source 2…

Ignored inputs

You expose	Technical name
Another text field	STRING
A mask, a sound	MASK, AUDIO, and any type missing from the table above
A model selector	ckpt_name, unet_name, vae_name, clip_name, lora_name, model_name
A file list	a COMBO of files from the ComfyUI machine (models, media)

Accepted output

One output node per workflow:

Your output node	Technical name	Your node produces
An image	SaveImage, PreviewImage	An image
A video	a name containing video: SaveVideo, VHS_VideoCombine…	A video
A 3D model	a name containing mesh or 3d, SaveGLB, SaveOBJ	A 3D model

Refused

• Another output node, such as SaveAnimatedWEBP or SaveAudio.
• Several output nodes in the same workflow.

An example: LTX-2.5 AlphaGen

Our LTX-2.5 AlphaGen and LTX-2.5 SDR to HDR workflows run through Comfy Bridge. A video goes in, an EXR sequence comes out, with a video preview.

123

SDR to HDR follows the same path: an SDR video goes in, an ACEScg EXR sequence comes out.