Brief the producer, approve the script, get a finished creator-style ad.
Bring your own keys. They're stored encrypted on this server and never shown again; leave a key field empty to keep the saved one.
When set, any chat turn with an attached photo is described by this model first; the main model reads that description instead of the raw photo, so its tool-calling still works. Leave blank to send photos to the main model directly instead.
From console.higgsfield.ai. The API has its own prepaid balance, separate from website credits.
Credits in the header: Higgsfield's API reports what's been spent, not the balance. After each top-up, enter the balance shown in console.higgsfield.ai and today's date; the header then shows what's left.
Prices for the "≈ $" on board buttons. Leave at 0 to use a guess for the chosen models (from this team's past usage: Flare ≈ $0.084 an image, Kling 3.0 ≈ $0.28 a second).
Higgsfield's own voice: the take model (e.g. Kling) speaks the line inside each take. Quick, but it can mispronounce names and the lip-sync is looser.A voice track first: a speech service records each line, you listen and approve it, then a lip-sync model animates the frame to that exact recording.
Searches the open web for reference images -- never scrapes Pinterest directly (no public API for that). Without a provider here, Brainstorm Mode still works, just without automated image search; uploading or pasting a link always works either way.