Submagic Review 2026: Pricing, AI Credits, Magic Clips & API
Submagic is a browser-based short-form production system built around animated captions, transcript cuts, automatic cleanup, b-roll, zooms, hooks, audio, brand styling, publishing, and long-to-short repurposing. Its current product extends beyond the ordinary editor with AI Actors Studio, REST and MCP automation, transcript review before API rendering, six-network publishing, and post-publication analytics. Those capabilities are real, but the buying model is unusually easy to misread: a plan's project count is only one of several independent meters, and current official pages contain material contradictions. This review checks the live purchase cards, June 2026 help articles, current API reference, feature pages, terms, privacy notice, and data-processing appendix separately, then turns them into a practical plan and workflow decision.
Is Submagic worth it in 2026?
Submagic is worth testing for creators and teams that repeatedly turn speech-led footage into captioned Shorts, Reels, TikToks, LinkedIn videos, and social ads. Pro is the practical self-serve plan: $39 month-to-month or $276 prepaid annually for 40 videos up to five minutes, 2K/30 FPS, six AI credits, cleanup, translation, Brand Kit, Storyblocks, and publishing. Business raises the ceiling to 100 thirty-minute videos, 4K/60 FPS, 15 AI credits, 100 API minutes, custom vocabulary, and priority service. The catch is that video count, duration, AI assets, API minutes, Magic Clips, seats, and templates are separate constraints. Test one complete production cycle before annual purchase, especially because official pages currently conflict on caption accuracy, source duration, AI-credit entitlements, voice translation, and Business template counts. Submagic is excellent at a repeatable social treatment; it is not a general professional NLE or an open-ended editing agent.
Submagic Pricing, Video Limits, and AI Credits
Current USD prices are per member. Month-to-month Starter, Pro, and Business cost $19, $39, and $69 according to Submagic's June 16, 2026 plan guide. The live annual cards display $12, $23, and $41 per member per month billed yearly, which means $144, $276, and $492 upfront. Each tier is governed separately by monthly video count, maximum video duration, output quality, AI credits, API minutes, and feature gates; Magic Clips is another per-member subscription.
| Plan | Monthly price | Included credits | What it means |
|---|---|---|---|
| Free | $0 | 3 videos / month | Up to 90 seconds and 200MB per video, Starter templates, free stock media, and a Submagic watermark. |
| Starter — monthly | $19/member/mo | 15 videos + 3 AI credits | Two-minute maximum, 1080p/30 FPS, no watermark, captions, free b-roll and audio, text trimming, and 10 API minutes. |
| Starter — annual | $144/member/year ($12/mo equivalent) | 15 videos + 3 AI credits / month | Starter features with annual prepayment. The published saving is $84 versus 12 monthly payments. |
| Pro — monthly | $39/member/mo | 40 videos + 6 AI credits | Five-minute maximum, up to 2K/30 FPS, Storyblocks, hooks, clean audio, silence and bad-take removal, translation, brand kit, publishing, and 10 API minutes. |
| Pro — annual | $276/member/year ($23/mo equivalent) | 40 videos + 6 AI credits / month | Pro features with annual prepayment. The published saving is $192 versus 12 monthly payments. |
| Business + API — monthly | $69/member/mo | 100 videos + 15 AI credits | Thirty-minute maximum, 4K/60 FPS, custom vocabulary and brand assets, priority rendering and support, and 100 API minutes. |
| Business + API — annual | $492/member/year ($41/mo equivalent) | 100 videos + 15 AI credits / month | Business features with annual prepayment. The published saving is $336 versus 12 monthly payments. |
| Magic Clips — monthly add-on | +$19/member/mo | 10 long sources / month | Long-to-short candidate generation with transcripts, virality scores, captions, hooks, b-roll, transitions, and effects. |
| Magic Clips — annual add-on | +$144/member/year ($12/mo equivalent) | 10 long sources / month | The same Magic Clips allowance, charged annually. It remains separate from the core editor subscription. |
| API pack — 500 minutes | $75/mo | 500 started minutes | $0.15 per API minute before rounding; every partial source minute consumes a full minute. |
| API pack — 5,000 minutes | $500/mo | 5,000 started minutes | $0.10 per API minute before rounding. Public packs also include 1,000, 2,000, and 10,000 minutes. |
AI credits reset monthly and are separate from ordinary video and API allowances. The live pricing cards give Starter 3, Pro 6, and Business 15; the March 2026 AI-credit help article lists Pro and Business but omits Starter. That omission conflicts with the purchase card. One AI image is documented as 1 credit and one AI b-roll or Avatar Studio result as 3, while the newer Actors landing page says a generation typically uses 3–4 credits depending on script length. Business also has a live template conflict: its headline card says 10 custom templates, while the lower comparison table shows 5; Pro's table shows 3. Capture the checkout entitlement and verify the signed-in meter. API minutes round every partial source minute upward. Magic Clips uses its own credits, including one Magic Clips credit per API project, rather than ordinary API-minute credits.
Choose Submagic when a known social-video recipe is the job: transcribe, tighten speech, apply a recognizable caption and b-roll treatment, review the result, then export or publish at volume. Start Free for caption and interface fit; use monthly Pro for a real five-minute project, six-credit budget, and publishing test; move to Business only when thirty-minute projects, 4K/60 FPS, 100 API minutes, custom vocabulary, priority rendering, or higher volume are demonstrated requirements. Add Magic Clips only after measuring usable candidates from a real episode. Choose Valmera when the edit is a broader outcome described in conversation and revised through rendered iterations; choose a conventional NLE when professional interchange, deterministic tracks, color, audio mixing, compositing, or long-form finishing is mandatory.
Method: this is an editorial analysis of Submagic's official product, help, and pricing material, checked August 24, 2026. It is not presented as a hands-on lab test. Valmera makes a competing video editor; there are no affiliate links, and the competitor's advantages are included because the right answer is sometimes Submagic.
Submagic at a Glance
What Submagic Actually Is
Submagic takes an uploaded recording through the standard social-video finishing pass: transcription, caption design, transcript trimming, silence or bad-take cleanup, hook text, stock or generated b-roll, music, audio cleaning, motion, brand styling, and export. The timeline and transcript remain editable. A user can delete a full sentence, word, or phrase and restore it later, adjust captions, reposition media, change layouts and ratios, and tune individual b-roll, zoom, and transition segments.
AI Auto Edit provides the fastest path. Its three published templates are Kelly for a minimal design-led treatment, Karl for a modern effective treatment, and Ella for a bold dynamic treatment. The selected system can coordinate caption style, b-roll, transitions, zooms or camera movement, pacing, and structure. That is a substantial first pass for a creator short, but it is still a template-directed automation pass that should be reviewed for meaning, timing, asset relevance, and brand fit.
The manual editor covers several recurring talking-head problems. Silence removal has slower, slow, and faster presets and previews the number of removals and seconds cut. Pro adds bad-take removal and clean audio. Caption translation and, where supported, voice translation address localization. Eye-contact correction, music, colors, hook titles, layouts, zooms, transitions, b-roll, and aspect-ratio conversion can complete a vertical or square delivery without a desktop editor.
Magic Clips is a distinct long-to-short workflow and a paid add-on. It accepts an upload, a YouTube URL, or a public Google Drive source, processes for a documented 10–25 minutes, and can return more than 20 candidate clips. Each candidate has a transcript and virality score and can be refined through the transcript and timeline before adding the familiar captions, hook, b-roll, transitions, and effects. Submagic's public help does not state a reliable hard source-duration ceiling for this add-on, so buyers should test their normal podcast or webinar length rather than infer it from broader marketing copy.
The API supports repeatable caption-video production through project creation parameters such as language, video URL, template, magic zooms, magic b-roll, a custom dictionary, and a webhook. A word-level transcript can be returned, and Submagic lists connections through Zapier, Make, n8n, and Google Drive. It advertises caption support in more than 100 languages. API capacity is its own duration meter and rounds every partial minute upward, separate from the monthly number of editor projects and the small AI-generation allowance.
Input guidance is unusually specific. Submagic recommends MP4 or MOV files, H.264 or HEVC video, whole-number frame rates instead of fractional rates such as 29.97 or 59.94, SDR, and Rec.709, with input up to 4K. Paid uploads can be up to 3,000MB. Starter outputs up to 1080p, Pro up to 2K, and Business up to 4K; the pricing matrix specifies 30 FPS on Starter and Pro and 60 FPS on Business. Ratios include 9:16, 16:9, 1:1, and 4:5, with automatic crop and manual repositioning.
Distribution is part of the Pro-and-higher value proposition. Submagic documents direct publish or scheduling to YouTube, TikTok, Facebook, Instagram, LinkedIn, and X. Subtitle users can export SRT and VTT or render captions into the video. A separate green- or blue-screen caption layer supports workflows that need to composite animated subtitles elsewhere, although it is not the same as an editable motion-graphics project file.
The commercial model is per member, but the public pricing page contains an important workspace contradiction. Its comparison grid says Starter supports one user per workspace, Pro three, and Business unlimited; the prices are simultaneously labeled per member. The Business headline card says 100 videos and 10 custom templates, while the lower comparison grid says unlimited for both. The official help article also says 100 uploads, so this review uses 100 as the prudent planning limit and advises teams to capture the checkout entitlement before committing.
A useful Submagic review has to separate four products that share one interface. The core editor turns a short source into a styled social video. Magic Clips finds candidate moments inside long sources. AI Actors Studio generates a presenter clip from text. The API and MCP surfaces automate creation, review, export, publishing, and analytics. They reuse templates and publishing infrastructure, but they do not share one unlimited allowance. Evaluate and cost each path independently.
The live pricing surface defaults to annual billing and displays monthly equivalents, not installments. Starter's $12 means $144 prepaid, Pro's $23 means $276, and Business's $41 means $492. The June 2026 help article still states month-to-month prices of $19, $39, and $69. Annual savings are therefore $84, $192, and $336 against twelve monthly payments, but fees are generally non-refundable and every seat multiplies the commitment. Verify currency, tax, renewal, and final cash charge at checkout.
Plan economics depend on accepted outputs, not only project slots. At the maximum count, annual Starter is $0.80 per included video, Pro about $0.58, and Business $0.41 before seats, Magic Clips, extra API packs, and human correction. Those figures rise if a creator uses fewer projects, hits the duration ceiling, exhausts AI assets, or needs more API minutes. Measure the last three months of raw projects by count and duration before choosing a tier.
The core editor remains constrained by both count and duration. Starter includes 15 projects up to two minutes; Pro 40 up to five; Business 100 up to thirty. A 31-minute source is not made eligible by having 99 projects left. Conversely, a series of one-minute clips can exhaust the project count while leaving most theoretical duration unused. Capacity planning needs a histogram of source lengths, not one average.
Submagic's newest Auto Edit landing page says videos up to two hours are accepted, while the April 2026 limit article and purchase cards cap Starter, Pro, and Business uploads and exports at two, five, and thirty minutes. These claims may describe different entry paths, source acceptance versus core-plan entitlement, or a stale generic FAQ. Do not use the two-hour statement to plan a paid deliverable. The checkout tier and signed-in upload gate are the operational test.
Caption accuracy also has a first-party contradiction. The Auto Edit page displays 99% near the top, describes 98% in its FAQ, and elsewhere uses both 99% and 98% across current feature pages. No public evaluation dataset, language breakdown, acoustic condition, dialect distribution, or error-counting method accompanies those figures. Treat them as marketing claims, not a transferable benchmark. Run a measured transcript test on your speakers and vocabulary.
The same page advertises 123 caption languages, while general product and API materials more often say 100+ and some feature headers say 48+. Those numbers may correspond to transcription, translation, voice, or a reduced landing-page set. The precise buying question is not how many languages appear in a badge. It is whether the target language supports transcription, translation, captions, voice translation if needed, dictionary terms, fonts, and acceptable accuracy in the exact workflow.
Smart Trim is the current cleanup name on the Auto Edit page. It describes silence, filler-word, and awkward-pause removal with Natural, Fast, and Extra Fast modes. Existing help also documents slower, slow, and faster silence presets. Either naming may appear in the current interface. More importantly, aggressiveness changes pacing and risk. Compare modes on the same file and listen for clipped consonants, breath loss, room-tone jumps, and unintentional meaning changes.
Bad-take removal has an editorial false-positive risk distinct from transcription accuracy. A later sentence is not automatically the best performance, intentional repetition may be rhetorical, and a stumble may be part of the personality. Review every deletion around jokes, emotional beats, sponsor claims, names, and transitions. Transcript editing makes restoration efficient, but it does not transfer final editorial responsibility to the model.
B-roll spans three sources. Starter includes free stock; Pro and higher add Storyblocks and the comparison table labels movie clips; users can upload their own media; AI generation consumes credits. The b-roll feature page says the system reads the transcript, places contextual clips, and offers a frequency slider. Relevance still needs review. A visually plausible insert can be factually wrong, geographically wrong, unlicensed for an edge case, or simply generic enough to weaken the brand.
The ordinary AI-credit pool is much smaller than the video count. At the documented help rate, Starter's three credits buy one AI b-roll, Pro's six buy two, and Business's fifteen buy five if nothing else is generated. An AI image costs one. The Actors page says an avatar generation typically uses three to four, while the help article says three. This means a Business user can process 100 ordinary videos yet generate only a handful of premium AI assets without another arrangement.
AI Actors Studio is not unlimited text-to-video. The current page advertises more than 120 presenters, voice selection, preview before generation, and a 720-character script ceiling—roughly 100 words. It explicitly positions the feature for short-form content under two minutes and recommends multiple clips for longer work. The page says licensed presenters may be used commercially, but users remain responsible for the script, claims, brand approvals, platform disclosure, and any uploaded assets.
Free preview is useful because the actor page says credits are consumed only when Generate is pressed. A serious test should still include the target presenter, voice, language, accent, difficult name, brand term, pacing, hand or body movement, captions, and final aspect ratio. Preview quality is not the same as the generated result, and a three- or four-credit generation can consume a meaningful portion of a monthly pool.
Magic Clips is a subscription inside a subscription. Monthly it adds $19 per member; annual it adds $144, displayed as $12 per month, for ten long sources each month. The help article says an upload, YouTube URL, or public Google Drive source can take 10–25 minutes to process and return more than 20 candidates. The number of candidates is not the number of publishable clips. Track usable yield after context, hook, crop, caption, and brand review.
The virality score is more detailed in the API response than on the marketing page. Current documentation exposes total, shareability, hook strength, story quality, and emotional impact. That is useful for prioritization, not a guarantee of distribution. No public methodology, training sample, confidence interval, platform-specific calibration, or causal performance validation is provided. Compare high- and low-scoring candidates against actual view and retention data on your own channels.
Magic Clips through the API has its own clear controls: create from a YouTube URL or upload a file; request minimum and maximum clip lengths from 15 to 300 seconds; choose a built-in template or a team theme; enable face tracking; and receive a webhook. The documentation says API Magic Clips consumes one regular Magic Clips credit per project rather than API-minute credits and requires the add-on. That separation should be explicit in any automation budget.
The API makes the core workflow more reviewable than a one-shot automation suggests. A project normally renders after transcription, but autoRender false pauses it. Once transcription completes, a developer can fetch the word array, correct text, retime words, add or remove entries, replace the array through Update Project, and then trigger export. Captions rebuild from the edited timing. This is the right path for names, regulated claims, accessibility, or any workflow where a bad transcript cannot silently reach publication.
API review has state semantics a production system must respect. With autoRender false, status completed means transcription is ready—not that a downloadable video exists. downloadUrl remains empty until export. Word edits are rejected before transcriptionStatus is COMPLETED. Webhooks are recommended; polling consumes request budget. Magic Clips uses a separate pipeline and does not support this exact transcript-before-render sequence.
Submagic's rate-limit documentation itself contains two levels of specificity. The API introduction broadly says lightweight operations receive 1,000 requests per hour, standard and uploads 500, and exports 50. The dedicated endpoint table narrows project creation and upload to 30 per hour and project retrieval to 100 per hour. Implement to the endpoint-specific figures, read the X-RateLimit headers, honor 429 Retry-After, use exponential backoff, and request custom limits when production volume requires them.
API minutes are a commercial limit separate from HTTP requests. Starter and Pro include ten monthly minutes and Business 100. Every partial source minute rounds upward, so 30 seconds costs one minute and 61 seconds costs two. This matters for batches of clips just over a boundary. Trim dead air before submission when safe, record charged duration from the API, and price the $75–$1,000 monthly packs by real rounded minutes rather than raw media duration.
The public packs do not all improve unit price. Five hundred, 1,000, and 2,000 minutes cost $0.15 each; 5,000 and 10,000 cost $0.10. A buyer expecting a discount at 1,000 or 2,000 gets more capacity but the same rate. The pack is monthly, so unused-capacity and reset behavior should be confirmed before provisioning a large tier. Custom plans can negotiate API limits separately.
Submagic's MCP server makes the API addressable from AI assistants. The official guide documents a Bearer-authenticated HTTP endpoint for Claude Code, Claude Desktop through mcp-remote, Cursor, Windsurf, and other MCP clients. It can create projects, generate Magic Clips, inspect status, export, publish, list posts, query analytics, add music, and list templates. MCP shares REST credits and rate limits; it is another control surface, not another allowance.
The MCP documentation also states a concrete limitation: claude.ai Custom Connectors are currently unsupported because they require OAuth rather than the Bearer-token pattern. Claude Desktop needs Node.js 20.18 or newer for the documented mcp-remote bridge. Production teams must keep the API key out of shared prompts and source control, scope access operationally, rotate it when exposed, and distinguish assistant authorization from human publishing approval.
Publishing now reaches beyond the three networks named on the Pro pricing card. The May 2026 help article documents YouTube, TikTok, Facebook, Instagram, LinkedIn, and X, with immediate or scheduled posting from exported projects. The API can publish to multiple connected profiles and later retrieve publication status, analytics, aggregated totals, daily time series, and profile-level performance. Test account permissions, aspect ratio, caption text, platform limits, time zone, failure retries, and duplicate protection before removing manual approval.
Direct publishing changes the risk boundary. A caption typo in an editor is a draft defect; the same typo sent automatically to six networks is a distribution incident. Use autoRender false, transcript approval, final preview, channel allowlists, schedule confirmation, and webhook logging. Verify whether failed posts are retried, whether partial multi-platform success is idempotent, and how a corrected export relates to already published versions.
Output quality follows the paid plan: Starter up to 1080p/30 FPS, Pro up to 2K/30 FPS, Business up to 4K/60 FPS. The pricing comparison renders Business as 1080p and 4K, while help explicitly says up to 4K. Resolution alone does not establish bitrate, color fidelity, HDR behavior, audio codec, loudness, chroma sampling, or source-frame preservation. Submagic recommends SDR Rec.709 and whole-number frame rates, so 29.97/59.94 and HDR camera workflows need a representative export test.
Input support has a similar marketing-versus-help split. Auto Edit says standard formats and any aspect ratio; current help recommends MP4 or MOV with H.264 or HEVC, up to 4K, whole-number frame rates, and SDR Rec.709, with a 3,000 MB paid file ceiling. Treat the technical help as the safer production specification. Transcode a copy rather than altering camera originals, and compare audio sync and color after round-trip.
The social aspect-ratio set includes 9:16, 4:5, 1:1, 16:9, and original on current feature pages. Automatic crop and manual repositioning can make a talking head fit, but subject tracking is not a substitute for composition review. Check every speaker, product, screen capture, lower third, caption safe area, and b-roll insert in every required ratio. One layout rarely works unchanged across TikTok, LinkedIn, and YouTube.
Translation has another meaningful documentation conflict. The current translator landing page describes subtitle translation that preserves the original audio and explicitly says it is not robotic voice cloning or lip sync. The help article says users can choose captions or voice where voice is supported. Until the signed-in workflow proves otherwise for the target language and plan, buy Submagic for translated captions and treat translated voice as conditional. Review names, numbers, claims, timing, line breaks, and cultural context with a fluent speaker.
Submagic is web-first rather than mobile-first. The official mobile article says there is no native iOS or Android app. A phone browser can upload a file, but the recommended next step is to open the project on a computer and edit there. That can help a field creator move footage into the queue, but it is not an end-to-end phone editor. Test large uploads, backgrounding, resumability, and sign-in on the actual device.
Team workspaces let members share projects, subscription, templates, and other workspace resources. Pricing remains per member, so adding collaborators increases annual cash cost even when the video allowance is also described per member. The public pages do not fully specify roles, approval permissions, guest access, audit history, folder-level boundaries, or whether every allowance pools cleanly. Create a test workspace with an owner, editor, and reviewer before migrating client work.
Custom-plan buyers receive the only public promise of advanced security and SSO, alongside custom members, duration, videos, Magic Clips, API, unlimited templates, and dedicated customer success. The self-serve Business card does not promise SSO. Procurement should request the authentication method, SCIM availability, audit logs, incident process, uptime commitment, security reports, pen-test summary, data locations, deletion SLA, and support escalation rather than assuming the words advanced security answer them.
The current privacy notice is more concrete than many creator tools. It names AWS, Vercel, Supabase, Stripe, PostHog, Google, TikTok, Customer.io, Crisp, FirstPromoter, Canny, and Churnkey; says servers are located in the United States; identifies EU and US processing; and describes Standard Contractual Clauses. It also says ordinary customer personal data is retained no longer than two years after the relationship and accounting data no longer than ten years.
The terms create a second retention view: account data up to three years after closure, billing data ten years, connection logs twelve months, and a data-return request after termination followed by destruction across environments and servers if no request is made within sixty days. Help says account deletions requested through support are batched at the end of each month, while the privacy page also offers in-account rights for EU, EEA, and UK users. These may govern different data categories, but a customer should ask for the exact media, project, backup, transcript, and analytics schedule.
Customer content rights require active configuration. The terms say customers retain their submitted-content rights but grant a worldwide nonexclusive license to use, modify, reproduce, and distribute content on and through the platform. They provide an email opt-out from promotional content use and allow Submagic to use the customer name as a business reference unless the customer objects in writing. A client-services team should send both opt-outs, keep confirmation, and incorporate them into onboarding.
AI output is not promised to be unique. The terms say other customers may receive identical or similar results and place responsibility for AI-generated results on the customer. That matters for avatars, generated b-roll, captions, hook titles, descriptions, and scripts. Review factual claims, likeness, trademarks, music, stock, originality, disclosure, and platform policy. A commercially licensed avatar does not validate the words placed in its mouth.
The public terms include a data-processing appendix that says personal data is processed only on customer instructions and for performing the services, with appropriate measures, breach notification, processor agreements, and return or destruction after termination. That is useful procurement language. The reviewed documents still do not contain a simple, product-wide promise that all uploaded video, transcripts, prompts, corrections, and outputs are never used for model training, so sensitive buyers should request that exact statement in writing.
Support is best effort on self-serve terms, available Monday through Friday from 9:00 to 18:00 CET. Scheduled maintenance is supposed to receive five business days' notice and not exceed one hour, while Business advertises priority support and rendering. No public response-time or render-time SLA accompanies those labels. A deadline-sensitive agency should test peak-hour render behavior and negotiate service levels under a custom agreement.
The strongest evaluation reel includes accented speech, a brand term, numbers, an intentional pause, a false start, music under voice, eye movement, a product close-up, vertical and landscape material, and a source at the actual frame rate and color profile. Produce one Auto Edit, one manual correction, one AI b-roll, one actor clip, one Magic Clips source, one API transcript review, one scheduled post, and one export at the tier maximum. Record time, credits, API minutes, false cuts, caption corrections, rights replacements, render failures, and accepted outputs.
A defensible purchase has four gates. First, does transcription and cleanup preserve meaning and voice? Second, do captions, b-roll, actor generation, crop, and export meet the brand and delivery specification? Third, can the chosen plan and add-ons support real monthly count, duration, AI, API, seat, and template demand? Fourth, do rights, privacy, security, deletion, support, and publishing controls fit the production risk? A fast demo that passes only the first gate is not yet an operational system.
Where Submagic Wins
The product maps cleanly to a social-video checklist
Captions, hooks, silence cleanup, b-roll, music, motion, ratios, brand styling, and publishing are the same operations creator teams repeat. Submagic keeps them in one workflow and exposes transcript and timeline controls for corrections after the automated pass.
Captions are a creative layer, not an afterthought
Animated templates, word-level correction, emphasis, custom styling, translation, SRT and VTT downloads, hardcoded subtitles, and a keyed caption-layer export cover both rapid publishing and workflows that finish in another editor.
AI Auto Edit creates a coherent first cut
The Kelly, Karl, and Ella systems do more than transcribe: they coordinate caption treatment, b-roll, transitions, camera movement, zooms, pacing, and structure. That can remove a meaningful amount of repetitive assembly before a human reviews the result.
Talking-head cleanup is well targeted
Text-based trimming, silence presets, bad-take removal, clean audio, eye-contact correction, hook titles, and b-roll attack the defects that usually slow down a podcast clip, expert tip, founder video, or direct-response ad.
The output ladder is explicit
Starter, Pro, and Business clearly move from 1080p/30 FPS to 2K/30 FPS and 4K/60 FPS, while project ceilings rise from two to five to thirty minutes. Buyers can quickly eliminate a tier that cannot meet a delivery specification.
Magic Clips has a reviewable candidate workflow
It proposes many moments rather than silently choosing one, displays transcripts and virality scores, and lets the creator refine each candidate before applying the short-form finish. That preserves useful editorial checkpoints in long-to-short repurposing.
The API reaches common automation systems
Webhooks, word-level transcripts, templates, language, b-roll, zoom, dictionary controls, and documented Zapier, Make, n8n, and Google Drive paths can turn a repeatable caption workflow into an operational pipeline.
Annual pricing materially reduces unit cost
At the published ceilings, annual Starter, Pro, and Business work out to about $0.80, $0.58, and $0.41 per included video before add-ons and AI usage. That can be attractive for a predictable calendar that actually consumes the allowance.
The free tier validates the visual language
Three no-card, watermarked projects let a creator test transcription, caption templates, text edits, b-roll rhythm, zooms, and interface fit on real footage before subscribing. That is enough to reject a poor workflow match early.
Pro is a coherent creator package
Forty five-minute videos, 2K output, Storyblocks, cleanup, translation, Brand Kit, publishing, and six AI credits cover a realistic recurring short-form workflow without forcing every buyer into the 100-video Business tier.
Business publishes a meaningful delivery step-up
Thirty-minute projects, 4K/60 FPS, 100 API minutes, custom vocabulary, priority rendering and support, and higher volume address real agency and media-team constraints rather than merely changing a badge.
AI assets are metered explicitly
The help center states one credit for an AI image and three for an AI b-roll or ordinary Actors Studio result. The pool is small, but a buyer can at least calculate the ceiling rather than discovering an invisible generator quota after purchase.
Actor previews do not spend credits
The current Actors page lets users check pacing and tone before Generate. With only a handful of monthly actor generations on self-serve plans, free preview is an important protection against avoidable credit waste.
Magic Clips exposes candidates and score components
Creators can inspect 20 or more suggestions, transcript, and virality score instead of accepting one hidden selection. API responses break that score into shareability, hook, story, and emotion, creating a useful human-review order.
The API supports a human approval gate
autoRender false allows transcription to complete without rendering, then permits word correction, retiming, addition, and removal before final export. That is materially safer than a pipeline that publishes the first ASR pass.
MCP makes automation accessible to agent users
Claude Code, Claude Desktop, Cursor, Windsurf, and other MCP clients can drive project creation, Magic Clips, status, export, publishing, analytics, music, and templates with the same API key and controls.
Publishing is paired with analytics
The platform can schedule or publish across six documented networks, while the API exposes status, per-platform analytics, aggregates, daily series, and profile performance. That closes more of the content operations loop than export alone.
The technical input guidance prevents silent quality loss
MP4 or MOV, H.264 or HEVC, whole-number frame rates, SDR Rec.709, and plan-specific output ceilings give production teams a concrete preflight target, even though generic landing pages sound broader.
The privacy page names the provider stack
A current subprocessor table identifies infrastructure, database, payments, analytics, marketing, support, affiliate, feedback, and retention vendors plus locations and transfer safeguards. That is more actionable than a generic vendors clause.
Custom procurement has a visible path
Custom videos, duration, members, Magic Clips, API, templates, success management, advanced security, and SSO are explicitly available through sales. High-volume teams know which requirements require negotiation rather than self-serve guesswork.
The editor and automation share the same asset system
Templates, brand styles, uploaded b-roll, captions, and publishing are usable across manual, API, Magic Clips, and actor workflows. When governance is configured well, that can reduce handoffs and visual drift.
Limits and Trade-offs
Four meters make the real capacity harder to see
The plan's video count, maximum duration, AI credits, and API minutes are independent. Magic Clips adds a fifth allowance. A team can have ordinary videos remaining while generated b-roll or API capacity is exhausted, so the headline project count is not total automation capacity.
The AI-credit pool is small
Starter's 3 monthly credits buy one AI b-roll or Avatar Studio output, or three AI images. Pro's 6 buy two such outputs and Business's 15 buy five if no other AI assets are generated. Stock b-roll remains available, but buyers expecting abundant generation should model the separate meter.
Long-to-short clipping costs extra per member
Magic Clips adds $19 monthly or $144 annually for ten long sources per month. An agency comparing dedicated clippers should add it to the base seat price and test whether source limits, candidate quality, processing time, and final clip count suit its workload.
Duration ceilings still define the core editor
Starter projects max out at two minutes and Pro at five. Business raises the ceiling to thirty minutes, but the system remains designed around compact social deliverables. A general long-form edit can exceed the product's intended shape even when the raw file uploads successfully.
Started-minute API rounding penalizes short overages
Every partial minute is billed as a whole API minute: a 61-second source uses two. A batch of clips barely over a minute can consume almost twice their raw duration. Normalize and trim inputs before submission when the process allows it.
Annual billing is a real upfront commitment
The $12, $23, and $41 figures are monthly equivalents, not monthly charges. Checkout costs $144, $276, or $492 per member for the year, before Magic Clips. The terms say fees are generally non-refundable, so the discounted rate trades flexibility for prepayment.
The refund policy is restrictive
Submagic's help center says it does not offer refunds, and its terms say paid fees are non-refundable. Consumer customers may retain a statutory fourteen-day withdrawal right unless performance is completed or the right is properly waived, but business buyers should assume no ordinary satisfaction refund.
Retention statements are not perfectly aligned
The privacy policy says ordinary customer personal data is kept no longer than two years after the relationship, while the terms mention account data for up to three years after closure. The terms also describe a post-termination data-return process and destruction after sixty days without a request. Preserve local masters and ask for contract-specific retention where it matters.
Content rights deserve a procurement read
Customers retain rights to submitted content, but the terms grant Submagic a worldwide nonexclusive license to use, modify, reproduce, and distribute it on or through the platform. They also permit use of the customer name as a business reference unless requested otherwise and describe an email opt-out from promotional content use. Confidential clients should exercise those controls and confirm the desired DPA terms.
There is no explicit public model-training promise
The current terms and DPA frame processing around providing the service and customer instructions, but the public documents reviewed do not contain a simple statement that uploaded video is never used for model training. That is not evidence of training; it is an unanswered procurement question for sensitive media.
Template consistency can become visual sameness
Recognizable captions, b-roll rhythms, hooks, and motion help a brand publish quickly. They can also produce a familiar creator-tool look. Distinctive storytelling still requires deliberate source selection, pacing, sound, graphic design, and revision beyond an Auto Edit preset.
This is not a professional interchange workflow
Submagic offers useful browser controls, 4K on Business, subtitle files, and a keyed caption layer. Its public product specification does not promise editable Premiere or Resolve timelines, advanced color management, multitrack audio routing, mezzanine codecs, or project interchange. Finish elsewhere when those deliverables are required.
The live template counts contradict each other
The Business headline card says ten custom templates, while the lower comparison table shows five; Pro's table shows three. The prior unlimited cell now belongs to Custom. Capture the checkout promise and verify the signed-in workspace before migration.
Caption accuracy is claimed as both 98% and 99%
The current Auto Edit page uses both figures and publishes no evaluation corpus, language breakdown, acoustic conditions, or error methodology. Neither percentage should be applied to a buyer's footage without a measured trial.
Language counts change by page and feature
Current pages use 48+, 100+, and 123. These may reflect translation, transcription, voice, or page drift, but the mapping is not consistently explained. Verify the exact feature-language combination rather than relying on the largest badge.
A two-hour marketing claim conflicts with plan limits
Auto Edit says videos up to two hours are accepted; April 2026 help and purchase cards cap the core plans at two, five, and thirty minutes. Production planning should follow the purchased entitlement and a signed-in test.
Starter AI credits are omitted from the credit help article
Pricing grants Starter three monthly AI credits, while the March 2026 credit article lists allocations only for Pro and Business. The live card is the purchase offer, but the mismatch weakens the documentation chain.
Actors Studio cost is documented two ways
The help article says one result uses three credits; the newer actor page says a generation typically uses three to four depending on script length. A small monthly pool makes the difference material. Read the in-app confirmation before generating.
Actor scripts are capped at 720 characters
The current page describes roughly 100 words and explicitly targets content under two minutes. Long training, sales, or course narration must be split into clips and recombined, increasing continuity and credit-management work.
Actor generation is not uniqueness protection
Commercial presenter licensing does not make the script, visuals, voice choice, or resulting message exclusive. The terms warn AI outputs may be identical or similar across customers and place use responsibility on the customer.
Magic Clips is one more seat-multiplied subscription
The $19 monthly or $144 annual add-on is priced per member and grants ten source projects each month. A multi-editor team can pay materially more than the simple add-on badge suggests.
Twenty candidates are not twenty usable clips
Help says Magic Clips may produce 20 or more suggestions, but context, hook, crop, captions, brand, rights, repetition, and platform fit determine publishable yield. Cost the accepted outputs, not the candidate count.
Virality scores have no public validation method
The API exposes helpful score dimensions, but Submagic does not publish training data, platform calibration, causal results, or confidence. The score prioritizes review; it cannot guarantee reach, retention, or revenue.
Magic Clips uses a separate credit pool
API Magic Clips consumes regular Magic Clips credits rather than ordinary API minutes, one per project. A pipeline can have API capacity remaining and still be unable to create another long-to-short job.
API rate-limit docs disagree at different levels
The introduction says upload operations receive 500 requests per hour, while the dedicated endpoint table says project POST and upload are limited to 30. Implement to the more specific table and inspect response headers.
MCP does not create free capacity
MCP calls use the same API key, credits, and rate limits as REST. Adding Claude or Cursor improves the control surface but does not expand project, minute, Magic Clips, or AI allowances.
Claude web connectors are not supported
The official MCP guide says claude.ai Custom Connectors require OAuth, while Submagic currently uses a Bearer token. Claude Code and Desktop are documented alternatives, but the web workflow is not equivalent.
Claude Desktop needs a local bridge
The documented configuration uses mcp-remote and requires Node.js 20.18 or later. This adds a local dependency, API-key handling, and support surface that a nontechnical team must maintain.
Default API rendering can publish errors faster
Projects render automatically unless autoRender is set false. A pipeline that trusts first-pass transcription can scale names, claims, timing, or caption errors. Human review needs to be designed explicitly.
Completed is an ambiguous API state
With autoRender false, completed means transcription is ready and downloadUrl is still empty. Systems that assume completed always means a downloadable render can fail or prematurely advance the workflow.
API minutes and HTTP requests are independent
A job must fit both commercial duration credits and endpoint rate limits. Buying a larger minute pack does not necessarily raise POST throughput, and a successful request can still exhaust rounded minutes.
The 1,000- and 2,000-minute packs do not lower unit price
They remain $0.15 per minute, the same as 500. The price drops to $0.10 only at 5,000 and 10,000. Larger commitment before that point buys capacity, not efficiency.
Publishing multiplies a bad approval
One incorrect export can be sent or scheduled to six networks. Public docs do not fully explain idempotency, partial multi-platform failure, retry behavior, or rollback. Keep a final human gate and event log.
The pricing card understates published network breadth
Pro names TikTok, YouTube, and Instagram, while current help documents Facebook, LinkedIn, and X as well. This is favorable product breadth but another example of entitlement information being distributed across sources.
Voice translation is not consistently promised
The translator page says the workflow preserves original audio and focuses on subtitles; help offers captions or voice where supported. Treat translated voice as conditional until the target language, plan, and signed-in feature are verified.
No native iOS or Android app
Official help says a phone browser can upload but instructs the user to edit later on a computer. That is useful ingestion, not a complete mobile editing workflow.
The recommended media profile excludes common camera modes
Whole-number frame rates and SDR Rec.709 are recommended, so 29.97, 59.94, HDR, log, unusual audio layouts, and some camera originals may require transcoding or careful export validation.
4K does not define professional master quality
Business advertises 4K/60 FPS, but the public plan does not specify bitrate, chroma sampling, mezzanine codecs, HDR, color metadata, audio configuration, or project interchange. Resolution is one delivery attribute.
Workspace roles and approvals are underspecified
Help says members share projects, subscription, templates, and everything else, but the public matrix does not fully define owner, editor, reviewer, guest, approval, audit, or client-isolation controls.
Self-serve Business does not include a public SSO promise
Advanced security and SSO appear on Custom, not the ordinary Business card. An organization requiring identity controls must enter a sales and procurement process rather than assume the highest checkout tier is sufficient.
Deletion rules vary by data category and document
Privacy says customer data up to two years after the relationship, terms say account data up to three, help batches account deletion at month-end, and the contract describes a sixty-day return window. Ask for a media-specific schedule.
Promotional-use opt-out is not automatic
Customers retain content rights, but the terms grant a platform license and require an email to refuse promotional content use. Customer-name reference use also continues unless rejected in writing. Agencies should send both requests.
No simple public no-training statement
The DPA language limits personal-data processing to service performance and customer instructions, but the reviewed documents do not plainly state that every uploaded video, transcript, prompt, correction, and output is excluded from model training.
Self-serve support has no public response SLA
Business promises priority support and rendering, while the terms describe best-efforts weekday CET support. Response and render commitments are not quantified on the purchasing page.
Annual add-ons compound the refund risk
Core plan and Magic Clips can both be prepaid per member, while monthly and annual fees are generally non-refundable. Test every required workflow before multiplying the commitment across a team.
Who Should Choose Submagic?
- Creators producing recurring talking-head Shorts, Reels, TikToks, and ads.
- Teams that value animated caption treatments and fast brand consistency.
- Marketers who want cleanup, b-roll, hooks, translation, and publishing in one short-form tool.
- Podcast and webinar teams willing to add Magic Clips for long-to-short candidate generation.
- Agencies with predictable monthly volumes, known duration ceilings, and API-ready templates.
- Business users that need 4K/60 FPS social exports and can verify the live seat and template entitlement.
- Solo creators who can fit most deliverables inside Pro's five-minute ceiling and forty-project allowance.
- Agencies willing to model separate per-seat, Magic Clips, AI-asset, and API-minute costs.
- Developers building a reviewed caption pipeline with autoRender false, webhooks, word-level corrections, and explicit export.
- Claude Code, Claude Desktop, Cursor, or Windsurf users who want to operate a video API through MCP.
- Social teams publishing or scheduling to YouTube, TikTok, Facebook, Instagram, LinkedIn, and X from one system.
- Teams that want publication status and performance analytics available through an API.
- Brands producing short synthetic-presenter clips within a 720-character script and small credit budget.
- Podcast teams that can justify Magic Clips by measuring accepted candidate yield from ten monthly sources.
- Business users whose actual requirements are 4K/60 FPS, thirty-minute sources, custom vocabulary, and 100 API minutes.
- Custom-plan organizations prepared to negotiate SSO, security, usage, support, retention, and higher rate limits.
- Teams that preserve camera originals and masters outside Submagic and can follow its recommended Rec.709 media profile.
- Operators who prefer a repeatable social production system over a blank professional timeline.
Submagic vs Valmera
Submagic is optimized for a known short-form recipe. A creator uploads or extracts a clip, applies a caption-and-b-roll treatment, corrects it, then exports or publishes. Valmera is optimized for outcome-level direction across a broader editing brief. The agent interprets uploaded footage, rewrites a versioned edit decision list, renders the requested operations, and continues revising through conversation.
Choose Submagic when the repeated deliverable is a branded talking-head short and animated captions, hooks, cleanup, stock media, ratios, and social scheduling should live together. Its template system and predictable video ceilings make that production line easy to understand, provided the separate AI, API, and Magic Clips meters fit the calendar.
Choose Valmera when each project may ask for a different structure, footage selection, pacing, sound treatment, freeze frame, montage, supporting media, motion, or long-form sequence. It is the more relevant test when the brief resembles directing an editor rather than selecting a social preset. A desktop NLE remains appropriate when a human needs deterministic track-level finishing or interchange.
The two systems' units are not directly comparable. Submagic charges seats, video quotas, AI assets, API minutes, and an optional clip-extraction allowance. Valmera meters agent work from its own credit balance. Benchmark one complete accepted video: include upload, first pass, transcript corrections, asset replacement, revision rounds, subtitle and master exports, human cleanup, and the value of direct publishing.
Against CapCut, Submagic is narrower and easier to operationalize for a speech-led, caption-heavy production line. CapCut has a much larger mobile, template, effects, generative-model, commerce, and social ecosystem. Submagic has clearer per-project tiers, a focused API, word-level pre-render review, MCP, and publishing analytics. Choose Submagic for a controlled repeatable pipeline; choose CapCut for a broad creative ecosystem and native mobile editing.
Against OpusClip, the decision begins with the source. OpusClip is deeply centered on long-to-short discovery, scoring, reframing, and distribution. Submagic makes that a paid Magic Clips layer around a more general short-form editor with transcript cleanup, captions, AI Actors, b-roll, and an API. Compare accepted clips per source, not candidate count or score labels, and include the Submagic add-on in the price.
Against Captions, both products cover automated social finishing, subtitles, cleanup, eye contact, translation, and synthetic presenters. Captions has a model-heavy creative ecosystem, mobile-centered history, cloning features, teams, and an API with its own economics. Submagic publishes a strong browser/API/MCP workflow and familiar caption-led editing. Test the exact platform, model, rights, credit, and output needs rather than treating the products as interchangeable caption apps.
Against Descript, Submagic is more explicitly designed for viral-style social output, animated captions, b-roll rhythms, and cross-network publishing. Descript offers a broader recording, document-editing, collaboration, audio, screen capture, and synthetic-media environment. A podcast team may use Descript for the primary production and Submagic for social derivatives—or select one after measuring total handoff and subscription cost.
Against a professional editor, Submagic can automate the repetitive assembly and distribution surface but cannot own the narrative, performance, factual, legal, or brand decision. The terms do not promise unique AI output or performance results. The best operation lets automation generate and format options, then makes a named human responsible for the accepted cut and publication.
Against Valmera, Submagic exposes a feature-and-template system for a known short-form recipe, plus an API/MCP surface for programmatic control. Valmera's differentiation is a broader agentic edit: describe a multi-operation outcome in conversation, let the agent rewrite a versioned edit decision list, inspect a rendered preview, and revise. Submagic currently has stronger self-serve social publishing and caption-template operations; Valmera is the more natural test when the brief itself should be the interface.
For the full feature-by-feature comparison, read Valmera vs Submagic.
Primary Sources Checked
- Submagic — Pricing and Plan Comparison
- Submagic — AI Video Editor
- Submagic Help — Current Price
- Submagic Help — Free Trial
- Submagic Help — Video Size and Length Limits
- Submagic Help — AI Credit Costs
- Submagic Help — AI Auto Edit
- Submagic Help — Magic Clips
- Submagic Help — API Credit Rounding
- Submagic Help — Recommended Video Formats
- Submagic Help — Export Resolution by Plan
- Submagic Help — Caption and Voice Translation
- Submagic Help — Complete Editing Guide
- Submagic Help — B-Roll Controls
- Submagic Help — API Workflow
- Submagic — API Product and Language Claims
- Submagic — API Documentation
- Submagic Help — Social Publishing and Scheduling
- Submagic Help — Cancellation and Pause
- Submagic Help — Refund Policy
- Submagic Help — Delete a Video
- Submagic Help — Delete Account Data
- Submagic Help — Team Workspaces
- Submagic Help — Transcript-Based Trimming
- Submagic Help — Silence Removal
- Submagic Help — Aspect Ratios
- Submagic Help — Video Layouts
- Submagic Help — Zoom Effects
- Submagic Help — Transitions
- Submagic Help — Caption-Layer Export
- Submagic — SRT Video-to-Text Export
- Submagic — SRT, VTT, and Hardcoded Subtitles
- Submagic — Terms of Use
- Submagic — Privacy Policy
- Submagic — Current AI Auto Edit feature, Smart Trim, accuracy, languages, and duration claims
- Submagic — Current Magic Clips product page
- Submagic — AI Actors Studio presenters, script, preview, credits, and commercial use
- Submagic — AI eye-contact correction
- Submagic — Subtitle translation scope and current language claims
- Submagic — Contextual, stock, uploaded, and generated b-roll
- Submagic — Social publishing product page
- Submagic Help — No native mobile app and browser-upload workflow
- Submagic API — Introduction, capabilities, and broad limits
- Submagic API — Endpoint rate limits and response headers
- Submagic API — MCP server for Claude, Cursor, Windsurf, and other clients
- Submagic API — Transcript review and correction before rendering
- Submagic API — Create Magic Clips from YouTube
- Submagic API — Upload a file for Magic Clips
- Submagic API — Update a project
- Submagic API — Export a reviewed project
- Submagic API — Publish or schedule a project
- Submagic API — List published projects and platform analytics
- Submagic API — Aggregated publishing statistics
- Submagic API — Profile-level social performance
- Submagic — Custom integrations and automation paths
Frequently Asked Questions
Want the AI to Perform the Edit?
Upload your footage to Valmera, describe the outcome, and refine the rendered edit through conversation.
Edit with Valmera →