Credits and billing

• Image, video, speech, edit, upscale, and assembly jobs
• Product, marketing, brand, commerce, explainer, and thumbnail workflow outputs
• Character training

generate_audio costs 2–8 FoxCut credits for up to 2,000 emitted UTF-8 bytes, including cues and speaker markers. A short read costs 2 credits; 1,000 ASCII bytes cost 4 and 2,000 cost 8. Both supported Fish models use the same customer-credit formula. Free in s2.1-pro-free describes provider inference, not free FoxCut credits. Estimate the exact script before submitting.

foxcut tool call estimate_generation_cost --json '{"kind":"tts","model":"foxcut-speak","params":{"script":"Third late-night text this month. Give me a reason to stay.","speechModel":"s2.1-pro-free"}}'
# Expected: 2 FoxCut credits. Select a voice with list_voices before submitting.
foxcut tool call generate_audio --json '{"transcript":"Third late-night text this month. Give me a reason to stay.","voice_id":"VOICE_ID_FROM_LIST_VOICES","speech_model":"s2.1-pro-free"}'

generate_audio.transcript maps to estimator params.script, direction to params.speechDirection, speed to params.speechSpeed, speakers to params.speakerVoices, and speech_model to params.speechModel. Pass the same values to both calls. Explicit models are honored; omission uses the configured default with no automatic paid fallback.

• Account, model, preset, voice, workflow, asset, character, and job reads
• Cost estimates
• OAuth connection and token refresh
• Creating a draft character or moodboard record

Agents should estimate first, state the total expected charge, and seek confirmation before a material spend or multi-output batch. Credits are reserved when each job is accepted; failed-provider refund behavior depends on the generation worker and ledger policy.