L'article ci-dessous est publié en anglais. Chaque chiffre qu'il contient est le chiffre publié par l'éditeur lui-même, avec la date à laquelle nous l'avons lu.
Descript sits in AI video because that is where a buyer searching for video tools lands, but it is not competing in the same lane as the video generators beside it. Runway, Pika, Luma and Veo make footage from a prompt. Descript edits footage you already have, by editing the transcript. That difference decides which allowance question you ask, because an editor burns media minutes on everything it imports and AI credits on everything it generates, and the two are not the same meter.
The two meters are the purchase decision
The pricing page states the shape in its own words: all plans include two types of usage currency, media minutes and AI credits, and the help center names what each one buys. Media minutes cover uploading and recording, with a still image counted as one second. AI credits cover the generative features: Underlord, Studio Sound, Remove Filler Words, Green Screen, Eye Contact, AI speech, avatars and generated video. The media allowance is monthly, as is the AI credit allowance on the paid plans, and neither of those recurring allowances rolls over. The free plan is different on one point: its AI credits are a one-time grant rather than a monthly allowance.
That changes the buying question from "can it do the job" to "which meter will I exhaust first". A podcast that imports four hours of source audio spends media minutes on all of it, then spends AI credits on anything generative you apply to it. A YouTube editor working with one long recording spends media minutes quickly and AI credits slowly. The plan you need depends on which of those two lives you actually lead, and the pricing page keeps the two visible enough that you can look at them separately rather than treating the plan as one blob of included usage.
Monthly $24 reads together with annual $16
Hobbyist, the paid entry plan, prints both figures on the same card: $24 a month on the monthly toggle and $16 a month on the annual one, with the page's own line saying you save up to 35% over the monthly plan. Reading them together is the honest way to read the card, because the entry price is only half the decision: the annual figure makes a longer commitment look attractive, and the monthly figure is what you pay if you want the freedom to leave after a month.
What the monthly $24 actually buys is a faster meter than the free plan's: 600 media minutes a month against the free plan's 60, and 400 AI credits a month against the free plan's one-time 100. Export moves from 720p to 1080p and becomes watermark-free. You get one seat, and the card says so. It is a low entry for an editor with real limits, and it is also the plan where the two meters are most likely to be exhausted by real work, because the media allowance is about ten hours of source material a month, which a podcast that records weekly will brush against quickly.
The free tier is built for evaluation
The free plan asks for no credit card and grants 60 media minutes a month and a one-time 100 AI credits. That is enough to bring in a real episode, edit it by transcript, and see whether the workflow is the one you actually want, which is the honest job of a free tier.
Free is constrained for final delivery, and the reasons are export-specific: local video export is capped at 720p and watermark-free export is not included on the free plan, the one-time AI credits do not refresh, and no top-up is listed for the free plan. The export limits are what keep the plan an evaluation rather than a publishing surface. As an evaluation it is a real one, because the transcript-first workflow is fully present on the free plan, which is the thing worth testing before you pay.
What the AI credit meter commits you to
This is the part of Descript a buyer can least predict, and the vendor says so in writing. The help center states there is no fixed price per action, that the same Underlord prompt can cost different amounts on different runs, and that even chatting with Underlord, without invoking a specific tool, spends a few credits per message. Underlord itself is labeled beta, and the document is candid that it can overpromise or follow you into workflows it is not equipped to complete.
Two things make the meter more tractable. First, some costs have published caps: Studio Sound and Remove Filler Words are each capped at 30 credits per file, so a heavy-handed cleanup cannot drain the balance silently. Second, the top-up prices are printed, at 350 AI credits for $35.00, 1,000 for $80.00 and 4,000 for $200.00, and the docs say top-ups are used only after the monthly allowance is exhausted and expire 12 months after purchase. Top-ups are available on Creator and Business only, which is worth noting for anyone planning to run a Hobbyist plan past its credits without upgrading.
None of that makes the meter predictable, but it makes it legible: you can see the ceiling where it exists, the price of going past it where there is one, and the absence of a fixed per-action rate wherever Descript declines to print one.
Export is the portability story
The export surface is documented rather than implied. You can export a composition locally as video, audio, GIF, text or subtitles, export the timeline non-destructively to keep editing in most major audio and video tools, and publish a standalone web page or push directly to YouTube and podcast hosts. The comparison table prints the resolution per plan, a free 720p export rising to 1080p on Hobbyist and 4k on Creator and Business, with watermark-free local exports on all paid plans and a web-link duration of one hour on Free and Hobbyist and three hours on Creator and Business.
The practical edges are also in the table. Media uploads cap at 1 GB on Free, 10 GB on Hobbyist, 20 GB on Creator and 50 GB on Business, and cloud storage at 5 GB, 100 GB, 1 TB and 2 TB across the same four plans. Timeline export is not one format but a set that the table lists by name: Samplitude and Reaper, Final Cut Pro, Pro Tools and Logic, Adobe Audition and Adobe Premiere. For anyone whose workflow ends in another editor, that list is the portability answer, and it is printed rather than promised.
Who the editor actually fits
The "editor rather than generator" framing is real, and it is worth staying disciplined about what it does and does not claim. It does not mean Descript is a better product than the generators in this category. It means it is a different product for a different job: if your raw material is already recorded, you want the tool that turns that recording into an editable document; if your job is to make a shot from a prompt, Descript is not that tool, and the comparison with Runway or Pika is a category mismatch rather than a choice between two ways to do the same thing.
That is the honest test of the thesis. The transcript-first workflow genuinely belongs to editors working from existing audio and video, which is the podcast and interview world. It does not belong to the prompt-to-shot creator, and a review that passed that off as universal would be overgeneralizing. Descript's own positioning does not claim the generator lane, which is why the profile and the review keep the boundary drawn where the vendor draws it, and why the big question stays the meters rather than the media type.
What the score rests on
The five components and their weights are published on our methodology page, so here is what this review put under each of them.
Fit for the stated job, 30 percent, 8.5 out of 10. Descript is the transcript-first editor in a category of generators, and it is a clear fit for editors working from existing recordings. Docked because the same boundary makes it a non-fit for prompt-to-shot work, which is a large part of this category's buying audience, and because the AI co-editor that could broaden its reach is still labeled beta.
Cost to get started, 25 percent, 7.5 out of 10. A real free plan with no credit card and enough media minutes to test the workflow, an entry price printed both monthly and annual on the same card, and top-up prices published. Docked because the credit meter has no fixed per-action rate, because top-ups exist only on Creator and Business, and because a real working month spends media minutes and AI credits from two separate non-rolling buckets that the buyer has to budget for independently.
What the vendor commits to, 20 percent, 8.0 out of 10. The help center documents the two currencies, their reset rules, the non-deterministic credit cost, the credit caps on some tools, the top-up prices, and the export surface, each in a dated page. Docked because "no fixed price per action" is a real limit on the promise, and because the legacy and sunset plans are documented separately rather than in the same table, which matters if you are pricing an existing account.
Portability, 15 percent, 7.5 out of 10. Local export across five formats, non-destructive timeline export to named external editors, web publishing and host integrations, with resolution and duration spelled out per plan. Docked because the free plan's 720p with watermark is a hard ceiling for the no-payment evaluation, and because cloud storage is yours only within the plan's limit.
Documentation quality, 10 percent, 8.0 out of 10. The help center is thorough, dated, and covers billing, usage, export and the AI tools in separate pages, with the pricing page's FAQ answering the credit-card question directly. Docked because the pricing table and the help center do not agree on every number, which the final section names.
Weighted at those weights, the sum is 8.5 times 0.30 (2.55), plus 7.5 times 0.25 (1.875), plus 8.0 times 0.20 (1.60), plus 7.5 times 0.15 (1.125), plus 8.0 times 0.10 (0.80), which comes to 7.95, and the review rounds that to an 8.0.
What we could not settle
There is no score here for running the product. We did not install Descript, we did not edit a project in it, and we did not measure transcription accuracy, rendering speed, output quality or ease of use, because none of those can be claimed without authorized use of the product, and this review does not run one. Everything above is what Descript publishes and what its docs commit to, each figure dated and listed in the source ledger.
Two intra-vendor conflicts surfaced and neither was resolved by this reading. The pricing page's comparison table and the help center's transcription overview state different numbers for supported transcription languages, and this review has no way to know which the product currently accepts; we report the discrepancy rather than pick one.
The second is a live pricing conflict on two Free fields, and it comes from the same Descript-owned surface. The canonical pricing page's comparison table reads AI credits as 100 (one-time) on the free plan and prints watermark-free video export as Not included on Free. Both are the figures this review uses for the plan model, and the paid plans carry monthly AI credits and included watermark-free export. But a current tool page, the FLAC editor pricing fragment, embeds a free-plan row that contradicts those two figures: 100 AI credits / month, and Export 720p, watermark-free. So this review treats the canonical pricing table as current for the plan facts, and flags the tool-page fragment as an unresolved stale-surface inconsistency rather than silently picking one reading.
The remaining unknowns are the consumer behavior questions that the docs do not answer: how many credits a realistic month of Underlord use consumes, which no fixed per-action rate exists to predict, and how the two meters actually behave on a plan a real team runs for a year. Those are exactly the questions an authorized product test would answer, and none of them are claimed here.
The Descript tool profile carries the same entry price and free tier with the same dates, and AI video is the hub that places an editor beside the generators.