Murf

Murf is three buying decisions under one brand. Murf Studio is a browser production environment that meters voice-generation time and project credits. Murf TTS API charges by input characters and has region/model-specific concurrency. Murf Dub charges a localization workflow by minutes, languages and service credits. A plan or sample from one does not prove fit for the others.

> Distinctive strength: Murf offers distinct Studio, API and dubbing products so creator production, embedded synthesis and localization can be sourced from one vendor. > > Where it stops being an advantage: Those products use different meters, models, regions and rights and cannot be priced as one plan.

Murf's published latency, voice-count and dubbing claims are inputs to a trial, not results from BenPicks. The decision here is narrower and more useful: which Murf product solves your job, what is its real accepted-output cost, and do its consent, model-lifecycle and data terms fit your risk?

Choose Murf Studio, API or Dub in 60 seconds

What exactly are you buying from Murf?

Start with the product boundary. Murf Studio is intended for script-to-voiceover and voiceover-video production. It organizes work into workspaces, folders, projects, blocks, sub-blocks and tracks. You combine narration with uploaded media, control voice delivery and export audio, video or script/subtitle files.

Murf API is infrastructure. It offers batch and streaming TTS, regional endpoints, HTTP/WebSocket delivery, model selection and format options. It is billed in characters and must be judged by concurrency, latency, errors, versioning and integration behavior.

Murf Dub is localization. It supports a transient API-only mode and persistent editable projects, with language/minute-based credits and separate speaker/rightsholder terms.

Choose the wrong comparison unit and the recommendation fails:

• Studio hour ≠ API character;
• API character ≠ finished accepted audio;
• Dub minute × language ≠ source-video minute;
• free Studio preview ≠ export test;
• Enterprise maximum concurrency ≠ default pay-as-you-go capacity.

How do Murf Studio plans work?

The current help center provides stable allowances even though the JavaScript pricing page did not expose stable dollar amounts in the retained public response:

Voice-generation time and project credits solve different constraints. A project credit limits how many active projects/work items the workspace can support; generation time limits synthesized output. Confirm how deletion, duplication, import and archived projects affect credits before migrating a library.

The free trial has no time expiry and ends when its 10 minutes of voice generation and 10 minutes of transcription are exhausted. It includes access to 300+ voices across 33 languages but no downloads. That is enough to compare editor behavior and previews, not enough to inspect exported audio/video or validate commercial delivery.

Because current Studio dollar prices were not reliably available from the canonical public surface, this draft omits an `Offer` rather than importing a stale snippet. Verify the intended workspace, cadence, currency, tax and allowance in checkout.

How much does the Murf API cost?

Pay-as-you-go is published at $0.03 per 1,000 input characters, with a $2 minimum purchase. Murf's product-specific terms count spaces, punctuation and most SSML tags. Only `<speak>` and `<voice>` tags are excluded from billing under the documented rule.

Simple base scenarios:

These figures exclude minimum/top-up effects, failed/retried requests, SSML overhead, multiple takes, storage/delivery, engineering, monitoring and human QA.

If a one-million-character corpus carries 5% SSML/punctuation overhead and 10% is regenerated, billed input becomes about 1.155 million characters, or $34.65 at the public rate. If two voice candidates are generated for 20% of the corpus, add another 200,000 characters: approximately $40.65 total base synthesis.

The free API allocation is 100,000 characters, one key, concurrency five and 1,000 requests/minute, with no time expiry documented. Pay-as-you-go supports up to three keys, concurrency 15 and 10,000 requests/minute. Those headline values need a regional correction.

What are the real API concurrency limits?

Streaming concurrency depends on region and context. Current docs say US-East can reach 15 contexts on pay-as-you-go, while global/regional routing outside US-East can be limited to two. WebSocket connections can reach up to ten times the streaming context limit, but requests sharing one context run sequentially and inactive sockets close after three minutes.

That creates a real architecture choice:

• route to US-East for capacity and accept geography latency/data-path implications;
• use the closest region with lower documented concurrency;
• negotiate Enterprise capacity;
• queue/batch requests;
• split environments by keys only where the contract permits.

Enterprise materials market reserved capacity up to 2,000 Gen2 and 10,000 Falcon 2 simultaneous calls. Those are negotiable maximums, not default plan entitlements. Require the order form to state model, region, burst/sustained limits, throttling response and remedy.

What changed with Falcon 2 and Gen2?

The current streaming documentation says legacy Gen2 streaming was scheduled for deprecation on 16 August 2026 and recommends Falcon 2, marketed around 100 ms latency for conversational use.

Do not treat a model-name change as transparent. Before migration, compare:

• target voices/locales and style support;
• pause/SSML behavior;
• output formats/sample rates;
• word timestamps;
• pronunciation and number handling;
• first-byte and total latency;
• audio consistency across chunks;
• error and retry semantics;
• regional availability;
• cost and concurrency;
• version pinning and future retirement notice.

Run the same frozen corpus through both models where legacy access still exists. A faster first byte is not a success if pronunciation, voice identity or long-turn stability regresses.

Which audio formats does Murf support?

Paid Studio exports document MP3, WAV, FLAC, ALAW and ULAW; video includes MP4 and MOV; text/subtitle exports include TXT, DOCX, SRT and VTT.

API documentation adds OGG, PCM and Base64, word timestamps and sample rates of 8, 24, 44.1 and 48 kHz. Streaming model/endpoint combinations may differ.

Validate the actual file:

• codec/container and sample rate;
• mono/stereo behavior;
• loudness and peak clipping;
• telephony compatibility for ALAW/ULAW/8 kHz;
• timestamp alignment after punctuation/SSML;
• chunk boundaries in streaming;
• import into the real NLE/LMS/telephony stack;
• whether Base64 changes retention or only delivery.

An export list is not proof of downstream fidelity.

Can Murf clone a voice instantly?

A current self-serve instant-clone entitlement was not established. Murf documents professional, sales-assisted custom cloning using roughly one to two hours of recordings, with a stated creation timeline around one to four weeks. Enterprise API describes cloning on request.

Terms require explicit written consent from every speaker for training and synthesis. They also permit Murf personnel, service providers and affiliates to listen to training/synthesized audio for clone quality and create non-defamatory internal QA utterances using the clone.

For a voice engagement, retain:

• signed speaker consent and identity;
• scripts/recordings submitted for training;
• allowed use cases, territories and term;
• internal QA/human-access disclosure;
• who can generate/share/export;
• revocation, termination and deletion process;
• model ownership and portability;
• replacement/retraining policy;
• rights after account termination.

Confidentiality language does not erase the explicitly permitted QA access. If human listening is unacceptable, negotiate the controlling terms before submitting training audio.

What does Murf Dub require?

Murf Dub has two operational modes:

• Transient: temporary API workflow; output links expire after 72 hours and editing is not available.
• Persistent: API/project workflow with stored editable projects, QA, collaboration, re-synthesis and re-download.

The API documentation lists source/destination language sets, up to one-hour video, 1080p, concurrency up to five on Free/PAYG and 15 on Enterprise, and watermark differences. Credits depend on minutes and language count.

The product-specific terms are the more important buyer evidence. They require valid explicit written speaker/rightsholder consent and include language concerning learning by Murf's AI model and moral-right waivers that may survive termination. Murf Dub QC permits personnel/providers/subcontractors to listen to and modify source/output.

Obtain legal review for any professional performer, union, deceased speaker, regulated material or revocable-consent workflow. Do not infer that “dubbing” merely translates text while leaving likeness rights untouched.

How rights change across Studio, API and Dub

Paid Studio output is described as commercially usable subject to the User Agreement, underlying rights and resale restrictions. Murf does not clear your script, uploaded footage, music, trademarks, voice or likeness.

Murf Dub terms prohibit representing output as Murf-approved/vetted, wholly human-generated or original human work. Automated significant decisions require adequate human review and legal compliance.

For publication, preserve:

• source and media licences;
• speaker/likeness consents;
• plan/order and terms version;
• clone/Dub-specific authorizations;
• final human review;
• synthetic-media disclosure decision;
• exported master and provenance.

Commercial eligibility is not uniqueness, non-infringement or indemnity.

How does Murf handle data?

General Terms treat submitted content as confidential, with access for Murf personnel, providers and affiliates as needed for service and defined exceptions. Privacy says personal data is generally retained while the account remains open and account information—including projects, media and uploaded text—is permanently deleted within 30 days after termination, subject to legal/security/fraud exceptions.

Enterprise API materials advertise stronger product-scoped controls:

• no-training covenant;
• Base64 inline audio with no object-storage persistence;
• ordinary pre-signed output URL available for 24 hours by default;
• regional data processing for Falcon 2;
• SOC 2 Type II and ISO 27001 evidence under NDA.

Do not generalize Enterprise API controls to Studio, cloning or persistent Dub. Obtain the actual report/certificate and map its scope to the product, region and legal entity in the order.

When three Murf products are the wrong answer

Choose another route when:

• the team cannot identify whether Studio, API or Dub is the controlling product;
• self-serve instant cloning is required but unavailable;
• professional clone QA access conflicts with confidentiality;
• regional API concurrency of two is insufficient and US-East routing is unacceptable;
• Falcon 2 migration does not preserve required voice/control behavior;
• Studio price/allowance is not confirmed in checkout;
• Dub's consent/model/moral-right terms are unacceptable;
• output requires a public contractual SLA not available;
• the workflow depends on long-term platform storage;
• zero retention is needed outside the scoped Enterprise API option;
• target voices/languages fail native or blind review.

Compare ElevenLabs for a broader self-serve cloning/API ladder, Descript Regenerate for correction of existing recorded speech, or browse the AI Voice category. Use Compare after fixing the product boundary and workload.

A Murf evaluation that produces a buying decision

Studio

1. Freeze 20 minutes of script in two languages with names, numbers, acronyms and pauses.
2. Select two voices and define acceptance thresholds before generating.
3. Record voice-generation time and project-credit behavior for first pass and 25% regeneration.
4. Add fixed media, export every needed format and inspect real file properties.
5. Have two blinded reviewers plus a native reviewer score pronunciation, artifacts and consistency.

API

1. Freeze 30,000 characters including SSML and compute billed length locally.
2. Test non-streaming, HTTP streaming and WebSocket in US-East plus the intended region.
3. Measure p50/p95/p99 first-byte/total latency, errors, 429s and retry amplification.
4. Compare Falcon 2 with the legacy baseline where possible.
5. Validate timestamps, formats, context behavior, key separation and output retention.

Dub

1. Use a consented two-minute clip into two languages.
2. Run transient and persistent modes; record credits, link expiry and stored artifacts.
3. Have native reviewers score meaning, names, timing, emotion and artifacts.
4. Test one declared edit and re-synthesis path.
5. Confirm consent, human-QA, deletion and disclosure requirements.

Approve each product independently. A Studio win does not approve the API, and an API latency win does not approve Dub rights.

The unresolved Murf purchase gate

Current self-serve Studio dollar prices, a controlling public SLA and instant-clone availability remain unresolved. Confirm them in checkout/order documentation rather than filling them from stale snippets.

Sources checked

• Murf Studio workspace ↗ — workspace types, hours, project credits and Enterprise collaboration; checked 27 August 2026.
• Free trial ↗ — voice inventory, 10-minute grants, no-card/no-download limits; checked 27 August 2026.
• Trial duration ↗ — no expiry before allowance exhaustion; checked 27 August 2026.
• Studio basics ↗ — editor/workflow structure; checked 27 August 2026.
• Voice export ↗ — Studio audio export formats; checked 27 August 2026.
• Commercial rights ↗ — paid output eligibility; checked 27 August 2026.
• Murf terms of service ↗ — confidentiality, cloning consent and human QA access; checked 27 August 2026.
• Product-specific terms ↗ — API character billing, Dub consent, output restrictions and service credits; checked 27 August 2026.
• Murf privacy policy ↗ — subprocessors, transfers, retention and account deletion; checked 27 August 2026.
• API plans ↗ — free/PAYG price, keys, concurrency and rate limits; checked 27 August 2026.
• API rate limits ↗ — model/region/context limits; checked 27 August 2026.
• Streaming ↗ — Falcon 2, Gen2 deprecation, endpoints and formats; checked 27 August 2026.
• API FAQ ↗ — character/request limits, formats, timestamps and language inventory; checked 27 August 2026.
• Enterprise API ↗ — reserved capacity, retention, training, support and certification claims; checked 27 August 2026.
• Dub API ↗ — modes, languages, limits, retention and failures; checked 27 August 2026.