Griptape Audio Transcription Driver: OpenAI

The straightforward way to transcribe audio with OpenAI inside a Griptape workflow. This node builds an OpenAI audio transcription driver - Whisper or one of the newer gpt-4o transcription models - and hands it out as an AUDIO_TRANSCRIPTION_DRIVER that plugs into the "Griptape Audio Transcription Task" node. No local model downloads, no VRAM, just your OPENAI_API_KEY and an audio file.

How it works

It constructs an OpenAiAudioTranscriptionDriver with the model you pick and the key it resolves from settings/environment, and returns it on the DRIVER output. Everything else happens in whatever consumes the driver: the Audio Transcription Task node takes the driver on its driver input and transcribes the audio you hand it. (If you skip the driver entirely, that task falls back to the transcription driver inside the agent's config - but wiring this node explicitly is how you control which model and key get used.)

The inputs that matter

Two, both optional:

• audio_transcription_model (default whisper-1, with gpt-4o-transcribe and gpt-4o-mini-transcribe in the dropdown) - the classic Whisper model vs the newer gpt-4o transcription family, which tends to be better at punctuation, speaker turns, and messy audio.
• api_key_env_var (default OPENAI_API_KEY) - the env var / settings key name, not the key itself.

That's the entire node. Output is a single DRIVER of type AUDIO_TRANSCRIPTION_DRIVER.

Installing it

Part of the one-pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/griptape-ai/ComfyUI-Griptape

or ComfyUI Manager → search "Griptape". Restart, set your OPENAI_API_KEY in Settings → Griptape (or as an env var), and you're set. Shared gotcha: the pack pulls griptape[all] plus a torch build that can conflict with ComfyUI's own; the README's troubleshooting covers the recovery.

Common issues

• [WARNING]: OPENAI_API_KEY is not set - the pack's most common failure; verify the settings entry and env var name spelling.
• Transcription of long files - OpenAI's API has file-size limits and billing is per-minute. For very long audio, chunking or a local model is worth considering before you hit a big invoice.
• Bare driver doing nothing - connect it to an Audio Transcription Task (or a tool that consumes transcription); a driver alone in the graph is inert.

When to pick this over the Groq version: you already have an OpenAI key, you want the gpt-4o transcription models (Groq only serves Whisper flavors), or you want all your Griptape drivers under one provider/key. When to pick Groq: you want speed and a generous free tier. Both are two-input nodes - the choice is really about which provider you're already paying.