DiffTextNode

DiffTextNode is the least glamorous node in ComfyUI-DiffSynth-Studio, and it's also the one you can't avoid. It's a text box. You type a prompt, it hands you a TEXT value. That's the entire job. It feels like a prank if you're used to ComfyUI's conditioning nodes, but it isn't - it's a type bridge, and the shipped Diffutoon workflow literally will not connect without it.

Why a node that does nothing exists

DiffSynth's nodes don't speak ComfyUI's standard types. DiffutoonNode expects its prompts as a custom TEXT type - not CONDITIONING, not a plain STRING. DiffTextNode is what produces that type. In the bundled Diffutoon workflow it sits there as the negative-prompt box, pre-filled with verybadimagenegative_v1.3, which is the negative embedding the pipeline auto-loads and resolves during prompt encoding. If you've seen this node's name in a workflow and wondered why someone built a node just to echo text, that's the answer: the echo is the feature, because the downstream node needs the right wire type.

The one input, the one output

There's a single input, text - a multiline STRING. And it's flagged for dynamic prompts in the source, so if you run the DynamicPrompts extension you can use {castle|ruins|village} style syntax here instead of babysitting one prompt at a time. The output is TEXT, and it goes straight into DiffutoonNode's postive_prompt and negative_prompt slots (yes, "postive" - the typo is in the author's own schema, so that's the spelling you'll see in the node).

One thing to keep straight: this is raw text for DiffSynth's own tokenizer, not ComfyUI conditioning. Don't go looking for CLIP-text-encode behavior on this node. If you want a second prompt source, you can just add another DiffTextNode and feed its TEXT output where a prompt is expected - there's no magic here, and that's the point.

Installing it

You don't install DiffTextNode on its own; it ships with the pack. That install is the heavier part: either search "ComfyUI-DiffSynth-Studio" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AIFSH/ComfyUI-DiffSynth-Studio.git
cd ComfyUI-DiffSynth-Studio
pip install -r requirements.txt

The requirements pull in some serious weight - cupy-cuda12x, transformers, controlnet-aux, imageio[ffmpeg] - and the author tested on Python 3.10 with torch 2.3.0+cu121. You'll also want ffmpeg on your PATH, because the video nodes depend on it. Restart ComfyUI and the whole AIFSH_DiffSynth-Studio category appears.

When you'd reach for it

Honestly? Almost never by itself. It earns its keep the moment you're building a Diffutoon workflow: grab it, type your positive and negative prompts, and wire both into the Diffutoon node. It's not a model, it won't slow you down, and there's nothing to tune. For a utility node this thin, "it just works" is the whole review.