Audio To Image Draft (Creepybits)

Ignore the intimidating name. The display name on this node - "Audio To Image Draft (Creepybits)" - tells you exactly what it's for: it's the first step in a pipeline that converts an audio clip into an image prompt. You feed it text describing an audio recording, and it wraps that text in a system prompt that tells an LLM to translate audio content into a visual scene description. What comes out is a draft of an image prompt, ready for an LLM to turn into the real thing.

What it's actually doing

Like several nodes in the Creepybits pack, this one is a "system prompt in a box." At init it loads a text file - assets/prompts/audio_keywords.txt, which does ship with the pack - and concatenates it with your text_2 input. That file is a long, well-written instruction set for an LLM: analyze the speech, music, and atmosphere in an audio clip and translate it into visual terms - setting, characters, actions, mood, color palette, artistic style - then output it as a single continuous string. No bullet points, no chit-chat. One string, ready for a prompt box.

One input (text_2, multiline), one output (text). Pure string concatenation, no API call, no key. All the "AI" happens downstream.

The pipeline it's meant for

On its own this node does nothing but prepend instructions. It earns its keep inside the pack's audio-to-image chain, which looks like this:

1. RandomAudioSegment (also in this pack) clips a segment out of a longer audio file.
2. GeminiAudioAnalyzer sends that clip to Gemini and gets back a text description of what the audio contains.
3. Audio To Image Draft takes that description, wraps it in the "translate audio into a visual scene" instructions, and hands the bundle to the Gemini API node.
4. Gemini produces the final image prompt, which you pipe into a CLIP Text Encode.

If you're not running the pack's Gemini nodes, you can still use this as a manual drafting aid: type your own notes about an audio clip into text_2 and paste the output into whatever LLM you use.

Installing it and what to watch out for

The node is part of the Creepybits pack - install the pack and it's there:

cd ComfyUI/custom_nodes
git clone https://github.com/Creepybits/ComfyUI-Creepy_nodes.git

Restart ComfyUI afterward (or find "Creepy" in ComfyUI Manager). The requirements pull in google-genai and friends for the Gemini side of the pack; the pure prompt nodes like this one don't strictly need them, but pip installs them anyway.

The one real setup task for the full audio pipeline is the Gemini API key: edit assets/scripts/api_keys_config.json so the "Gemini Free" / "Gemini Paid" entries point at text files containing your actual keys. Without that, the analyzer and API nodes can't run - but this node itself will happily output its draft regardless, because it never touches the network.