Skip to main content
Whatcha need to Know?
Table of Contents
< All Topics
● Knowledge Base · Panels

AI Generator F4

Generate sound effects and speech inside basehead, and spin AI variations of your own sounds. SFX and Speech run on ElevenLabs… the Variator (and SFX on the Stability provider) run on Stability AI (Stable Audio). Available on all Premium editions.

AI Generator icon

Three modes — SFX, Speech, and Variator. Open the panel via the View menu or F4.

SFX

Generate sound effects from a text prompt with control over duration, seed, and prompt influence.

Speech

Generate speech using a selected voice model with stability, similarity, and expression controls.

Variator Stability AI

Make AI variations of a sound you already have — select it in Results or the Taglist and the Variator re-generates it through Stability Stable Audio (audio-to-audio).

SFX generator panel
SFX
Speech generator panel
Speech
Variator generator panel
Variator

API key

  • API Key Input — paste your ElevenLabs API key to enable generation.
  • Get API Key button — opens a free signup link (affiliate link, supports basehead).
  • Subscription Info button — shows your ElevenLabs subscription tier, usage, and available features.
  • Stability AI (Stable Audio) — required for the Variator and for SFX on the Stability provider. Add your key in Options → AI (get a key). A missing key shows an orange banner naming the provider… clicking it opens Options → AI.

Generation options

Add to Taglist

Sends generated files to the active Taglist for quick previewing. If off, the output folder opens in Finder / Explorer when generation completes.

Use Custom Path

Specifies a custom output folder for generated files. If blank, files go to the global Transfer Path.

Share to Community Library (SFX only)

Uploads a copy of the generated file to our server for possible inclusion in a future free basehead community library. Adds a few seconds of upload time per file.

Settings

SFX

Provider

ElevenLabs or Stability (Stable Audio). On Stability the Duration becomes whole seconds (1–30 s) with a Model and Seed… the ElevenLabs-only controls (sub-second Duration, Prompt Influence) hide.

Duration

0.1 to 22 seconds.

Seed value

Set a seed for consistent results, or leave at 0 for random.

Prompt Influence

How closely the result follows your prompt text.

Variations

Number of variations to generate per prompt.

Speech

Voice Selection

Pick from voices available in your ElevenLabs account.

Voice Stability

Higher values = more consistent results.

Voice Similarity

How closely the output matches the source voice character.

Voice Expression

Adjusts emotional range and expressiveness.

Speaker Boost

Toggle to enhance audio output.

Model Selection

Choose between ElevenLabs’ speech models.

Variator Stability AI

Pick a sound in Results or the Taglist and make AI variations of it through Stability Stable Audio. The source audio is the anchor… you just steer how the takes differ.

How it works

  • Source — the Source line shows the selected sound… Use Selected Sound re-captures whatever is selected. (The Source never auto-switches to a file the Variator just made.)
  • Base prompt — the AI is always told what the sound is, from a base prompt built automatically from the source. Base prompt from: Description | File Name picks which (Description = its description + keywords, catalog noise stripped). The base shows as a dim line.
  • Input Text is optional — it only adds how the takes should differ (e.g. “more gritty and hissy, in your face”). What gets sent is the base prompt plus your text.
  • Similarity to source — 0.2 (Close) to 0.8 (Loose), default 0.5. The Auto toggle (on by default) moves the slider as you type — closer with no extra text, looser the more you add (empty ≈ 0.25, ~150+ characters ≈ 0.70). Turn Auto off to set it by hand.

Takes & output

  • Number of Variations — 1 to 10 (default 4). A batch makes one file holding all the takes, with a cue marker at the start of each (a “sausage”), named <source>_varx<N>.wav. A single take is <source>_var01.wav with no markers. Takes are separated by 0.5 s of silence… each is the source length plus up to 1 s of tail.
  • Model — stable-audio-2.5 (fast), stable-audio-2, or stable-audio-3 (default — clearest, queued so results land a little later). The batch must fit the model’s input limit (190 s on 2 / 2.5, 380 s on 3) or Generate declines with a length message.
  • Save beside the source file (default on… off = the Generator output folder) and Match source sample rate / bit depth / channels (default on… the API returns 44.1 k stereo).
  • The sausage carries the source’s description, keywords, category, subcategory and library in its metadata, plus a provenance note, and lands in the Taglist with its real rate / bits / channels.

Cost & credits. One call per batch, whatever the take count — the cost line reads like 1 × 26 credits. The credits label at the top shows your Stability balance and refreshes when you enter the mode and after each run.

It runs in the background. A batch takes a while… the panel shows elapsed time and tells you to carry on working. A toast — “Variator finished — N take(s) ready” — pops when it lands, wherever you are in the app.