Lesson narration
Split a lesson into manageable sections and save each script with its audio. Regenerate a corrected section without recreating the entire lesson.
Build an AI voice product
Turn a reviewed script into useful audio.
Build narration around lessons, product explanations and approved scenes. Keep the words, voice choice and generated file together in the customer’s project.
Demos start with prepared briefs and shot suggestions. You choose when to generate; preparing a plan does not spend AI credits.
Put the script before the Generate button.
A course creator needs accurate lesson narration. A business needs its name pronounced correctly. A video editor needs a line that fits the scene. Give users a script they can edit and approve, then submit only the intended spoken words to an appropriate audio endpoint.
Business voiceovers
Review names, numbers and claims before generation. Offer only voices and languages that the selected endpoint actually supports.
Audio for a video scene
Create a voiceover as a separate asset, then measure and align it with the selected clip in your assembly workflow. Generated audio is not automatically synchronized to video.
From script to a saved voiceover
Approve the words
Show the exact narration in an editable field. Keep instructions such as “warm delivery” separate from the text unless the endpoint explicitly expects inline direction.
Choose supported settings
Validate language, voice, length and any reference-audio requirements against the model’s inputs. Custom voices require the necessary rights and permissions.
Listen and accept
Play the result, check pronunciation and pacing, then save the accepted file. Store its script and model choice so a later correction has context.
Keep the workflow precise
An audio result might be a file URL, a structured output or several segments, depending on the model. Map the actual response to your player and storage flow. If the product needs subtitles, timings or a combined movie, treat those as explicit additional steps. Start by proving one short narration before adding long documents or batch jobs.
Choose a model that fits the job.
Search the fal catalog and connect compatible queue inference endpoints through a model ID and model-specific JSON inputs. Running a request requires schema validation, provider access and verified pricing. A catalog listing is not a guarantee that the model can run: models without a supported usage quantity stay unavailable until configured. The service does not promise that every catalog model has been tested.
Realtime sessions, streaming generation and training are separate integrations. A new model may need different input fields and result handling; choosing its name alone does not adapt your interface.
Technical references: fal queue and result documentation · Bubble server-side plugin actions.
One managed plan
Build it. Run it. Get support.
$49.99 / month
$30 in AI credits included.
Plan details & account →Subscription availability is shown on the account page.
- Free Bubble connector for the managed service
- Building skill for adapting an existing Bubble app
- Support from The Wise Bubbler
- Tracked generation jobs and a shared app credit allowance
Credits reset monthly with no rollover. Your app’s users share its allowance. Eligible requests use verified fal unit prices and supported usage quantities. Credits are reserved before generation and reconciled to actual provider costs afterward, which can take over an hour. Different models and settings use different amounts. There are no automatic overage charges.
Bubble hosting, your AI assistant and any other services your app uses are separate. Review the selected model’s terms for your intended use.
Before you build
Does every fal model support speech?
No. Choose an endpoint built for the audio task and inspect its inputs. A speech model, transcription endpoint and music model solve different jobs.
Is this a live voice agent?
This workflow submits and retrieves queued generation jobs. Realtime conversations, streaming audio sessions and telephony need separate integrations.
Can customers use a cloned voice?
Only offer a reference-voice workflow when the provider supports it and the customer has the required permission. Do not promise that every voice option is available through every model.