arkennemasis fal · MiniMax Speech 02 Turbo · $0.06/1k chars

Six cents per 1,000 characters, same input surface as the HD node next to it. Turbo is the version you run while the script is still moving - cheap enough that changing one sentence and re-rendering doesn't sting, which is the whole reason it exists.

What it is

The fal wrapper for minimax/speech-02-turbo, MiniMax's fast TTS tier. MiniMax's voice line is the piece of their catalogue they position directly against ElevenLabs, and this is the cheap rung: same predefined voices, same language list, less polish. In ComfyUI terms it's a text-in, audio-out node with no local weights, and it's the natural companion to a video pipeline - narration, then the pack's Narration Length node to make each shot outlast its own voice-over.

How it works

FAL_KEY from the .env beside your launcher, validation, a cost estimate computed from your text's character count, one billed request to fal, polling, download into output/fal/fal-ai_minimax_speech-02-turbo/. Text is the only input type; nothing gets uploaded, so the only thing you can lose on a bad configuration is time.

Inputs worth setting

• text - the field is named text on this generation of nodes, capped at 5000 characters. Your bill is a function of this box: roughly 500 English words is 2,600 characters, about 16 cents at Turbo rates.
• voice_setting_voice_id - free text, defaulting to Wise_Woman. Paste any MiniMax predefined voice ID; an empty value leaves it to the model.
• voice_setting_emotion - (not set), happy, sad, angry, fearful, disgusted, surprised, neutral. Leave it at (not set) unless you need a specific read; the model picks something sensible from your text.
• language_boost - set the language explicitly rather than leaving it at (not set). It's the difference between a German sentence read as German and a German sentence read by an English speaker guessing.
• voice_setting_speed - 0.5 to 2.0. Nudging to 1.05 is a common fix for a read that feels slow in a short-form cut.
• Also here: voice_setting_pitch (−12 to 12), voice_setting_vol (0.01–10), voice_setting_english_normalization (off by default - turn on for number-heavy scripts).
• Output plumbing: audio_setting_format (mp3, pcm, flac), audio_setting_bitrate (32k–256k), audio_setting_sample_rate (8k–44.1k), audio_setting_channel (1 = mono, 2 = stereo), pronunciation_dict (JSON text replacements for names the model keeps getting wrong), and output_format (url or hex, non-streaming only - leave it at hex).

Optional pack controls: max_cost_usd (default $20 - effectively unreachable on TTS unless you're pasting a novel, and 0 disables it), max_concurrent (default 1; one paid fal call at a time across a run), reuse_identical_run (unchanged text returns the already-paid file), extra_json (raw JSON merged into fal's request last).

Outputs: audio (a real waveform with a player on the node), audio_path (file on disk), duration_ms (the measured read length - feed it into Narration Length or Word Timings so captions and shots follow the voice instead of a guess), and info with the request id, estimate and fal's answer.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis
pip install httpx

Or Manager → Install via Git URL. Then create .env beside run_nvidia_gpu.bat:

FAL_KEY=your-key-from-fal.ai/dashboard/keys

Restart and it's under arkennemasis/fal/Audio/MiniMax.

When Turbo is enough, and when it isn't

For anything where the voice is functional - a tutorial, a product demo, a rough cut you'll re-record - Turbo is done. The honest reason to step up to HD at $0.10 is the final delivery on a narration people will sit through for ten minutes: that's where the extra warmth and the better handling of long sentences shows.

One habit that will save you money and confusion: because reuse_identical_run is on by default, tweaking the speed but not the text still counts as a change, while re-running with nothing changed returns your old file. If you're doing a five-variant read-aloud comparison, that's six cents a take - cheaper than the coffee you're drinking while you decide.