Gemini 3.8 Flash-Lite TTS · Text to Speech

The cheap voice in the family

PopNow ships three Gemini speech nodes and this is the bottom of the ladder: 0.03 credits per second of audio, thirty prebuilt voices, and the least expressive of the three. If you're voicing a rough cut, generating scratch narration for a video you'll re-do, or need fifty lines of placeholder dialogue to time an edit, the arithmetic here is hard to argue with - a minute of speech costs about two credits.

The catch is right there in the author's tooltip for text: "Spell out sounds such as sighs or laughter; bracketed cues may be read aloud." On the Pro and Flash nodes, [sighs] becomes a sigh. Here it might become the word "sighs". Write the sound out - "hah", "mmm", a long dash for a pause - and treat square brackets as decorative at best.

How the call works

Nothing runs on your machine. The node posts your text (or your scripted dialogue) to PopNow with a price quote first, submits the task under an idempotency key so a retry can't bill twice, polls until it's done, and decodes the returned audio into ComfyUI's AUDIO format. When the model returns several takes they arrive as a batch, resampled and padded to the longest so a single Save Audio node writes one file each.

The pack has no Python dependencies outside ComfyUI's own stack, which means no TTS project to babysit - no transformers version roulette, no per-engine runtime isolation, none of the tax that makes local audio a chore. You're paying credits in exchange for skipping that.

The inputs a beginner actually touches

• text - the words to be spoken. For a single voice, this is the whole job.
• dialogue + speakers - leave both empty unless you want a two-person scene. If you do, dialogue is a JSON list of {"text", "speaker"} objects and speakers is a JSON list of one or two voices, each with a name (referenced by the dialogue lines), a required voice_id, and optional pace, style, accent and profile.
• voice_id - combo, default Kore. Thirty options; each is a fixed character, not a dial.
• style, pace, accent - all default to model default. That's the honest default: the field is omitted from the request and the voice does its natural thing. accent here defaults to model default rather than a fixed neutral, unlike the 2.5 Pro node - worth knowing if you're copying settings between the two.
• instructions, scene - free text that shapes delivery without being spoken. Mood, pace, who's listening; where the words are spoken. Cheap to use, occasionally surprising.
• temperature - -1 to 2, step 0.05, where -1 hands the choice back to the model. Bump it if you want three genuinely different takes of the same line.

Output is audio (AUDIO, possibly a batch) and task id (STRING).

Install

Manager → search PopNow → Install, or:

cd ComfyUI/custom_nodes && git clone https://github.com/popnow-ai/popnow-comfyui

Restart, then create a key at popnow.ai/api/console/keys and paste it under Settings > PopNow, or export POPNOW_API_KEY if the box is headless. Restart again - the node list is rebuilt from your account's catalog on startup, and "the node vanished" is nearly always a missing restart. Delete the old standalone popnow_api.py if you have one.

Fixing the usual failures

No key or a revoked one gives you No PopNow API key is set or PopNow 401. 40310 means your email isn't verified. 40201 means no credits - at this node's rate that's harder to hit than on the Pro model, but a long script still adds up. A 400 that names dialogue or speakers is almost always invalid JSON; the fields are strings, not parsed objects, so one stray comma kills the run. Models that decline are reported as content_rejected and refunded.

Beginner move worth unlearning: hitting cancel. It stops you waiting; the task keeps running on PopNow and is charged if it finishes. If you want to stop spending, stop queueing.

PopNow is a new vendor with no evident community history, so take the low-stakes route - small credit pack, key in the environment, and a workflow you've actually run before you point it at a deadline.