text2speech

text2speech turns typed text into a spoken audio file inside ComfyUI. It's the older name for what Mana Nodes now calls Generate Audio, and it runs on Bark, Suno's generative TTS model - which means it does the expressive stuff (laughs, sighs, singing, mid-sentence language switches) that plain read-aloud nodes can't.

The honest take first

This is the experimental part of the pack - the author flags the text-to-speech node as such and keeps "Improve Text to Speech" on the to-do list. Bark is expressive and fun, but it's slow, non-deterministic (same text, different result each run), and happy to wander off your script. Meanwhile the ComfyUI audio ecosystem has grown a bunch of other TTS options. So: reach for text2speech when you want characterful audio and can tolerate a few rerolls, not when you need clean, predictable narration.

Steering Bark with tags

Bark doesn't do "voice settings" so much as respond to cues you write into the text:

• Bracketed non-speech sounds: [laughter], [sighs], [gasps], [clears throat], [music].
• ♪ ... ♪ around lyrics to push it toward singing.
• CAPS to emphasize a word.
• MAN: / WOMAN: / NARRATOR: speaker hints (not always obeyed).
• Mixed languages in one prompt - it'll attempt each with an appropriate accent.

Most of the behavior lives in how you write the prompt, which is why the node itself has so few controls.

The inputs and output

• text - multiline; your script plus any Bark tags.
• filename_prefix - output path for the .wav, relative to the output folder (default audio_files\audio).

The output is audio_file, a string path to the generated WAV. Wire it into an audio2video / Combine Video node and the generated speech becomes the soundtrack of a rendered clip.

Installing it

Part of Mana Nodes. ComfyUI Manager → search Mana Nodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ForeignGods/ComfyUI-Mana-Nodes.git

Install requirements and restart. Expect a large one-time download: Bark pulls several gigabytes of weights on first use. The first generation will hang on that download before it produces anything - that's normal - and you'll want enough disk and VRAM for it.

Common issues

Naming, first: a current install probably shows Generate Audio rather than text2speech. Same node, newer label; point old workflows at it.

Then the Bark realities. It's slow and it won't reproduce the same take twice, so treat rerunning as the primary "fix" for a bad result - there's not much to tweak. Long inputs are where it drifts hardest; give it a sentence or two at a time and join the clips afterward instead of one long block. If the first run looks frozen, it's the multi-GB model download finishing. And keep expectations calibrated: TTS is the pack's rough, experimental corner - the text-animation nodes are the reason Mana Nodes has a following, with this one very much a bonus.