Used this tool? Rate it
Used this tool? Rate it
Recraft is an AI image and design platform whose distinguishing bet is vector output. Most text-to-image tools hand you pixels: a PNG that looks right at the size it was generated and degrades the moment someone needs it on a billboard or inside a favicon. Recraft generates SVG — paths, strokes and fills — alongside conventional raster images, and that single architectural difference determines who this tool is for and where it disappoints.
The company presents itself on its homepage with the line "tastefully crafted AI image models," and organises the product as a studio plus a set of named tools: Recraft Studio, AI Image Generator, AI Image Vectorizer, AI Vector Generator, AI Photo Editor, Mockup Generator, Image Upscaler, Background Remover and AI Eraser. The capabilities the homepage foregrounds are revealing of the intended user: "Complex editable vector graphics," "Consistent styles," and "Design-ready typography." Those are designer concerns, not hobbyist ones.
The legal entity is Recraft Inc., named as such at the head of its terms of service, which carry a last-updated and effective date of 11 August 2026 — an unusually current document for this category. The About page states the company was founded in 2022 in the US, yet lists its headquarters as London, UK. TechCrunch's coverage places the company in San Francisco. Both are reported here rather than reconciled; the discrepancy is in the public record and this page does not resolve it.
Funding is well documented by named reporting. TechCrunch's Charles Rollet reported on 5 May 2025 that Recraft closed a $30 million Series B led by Accel, with Khosla Ventures and Madrona participating, following a $12 million Series A led by Khosla in 2024. The company's own press release for the same round adds Nat Friedman, Elad Gil and RTP Global as participants. Scale figures come from the company itself: TechCrunch relays that the startup "recently passed $5 million in ARR and 4 million users," while the About page claims "over 3 million users across 200 countries" and "more than 350 million images." Note that these two company-sourced user counts disagree with each other; neither is independently audited, and both are presented here as claims rather than facts.
The product's technical credibility rests on a specific, checkable event. TechCrunch reports that Recraft's V3 model, "codenamed red_panda" because early users kept generating images of the cute mammal, topped the Artificial Analysis benchmark, beating DALL-E and Midjourney. CEO Anna Veronika Dorogush — previously in machine learning at Yandex, Google and Microsoft — states the company builds its own models from scratch and competes with Midjourney, Adobe Firefly, Stable Diffusion and Black Forest Labs.
One thing to settle before going further, because it changes the arithmetic of trying this tool: on the free plan you do not own what you generate. That is not an inference; it is the company's own stated policy, quoted in full further down.
This is the reason to consider Recraft over a general image generator. The AI Vector Generator page claims output is "fully editable SVGs — paths, strokes, and fills stay clean and accessible," that you can "export to your own editor (Figma, Framer, Illustrator) and keep working from there," and describes results as "production-ready: clean geometry, editable layers, and consistent design quality."
That claim deserves precision rather than repetition. An SVG with clean, sparse, named paths is a design asset. An SVG that is technically vector but composed of thousands of auto-traced fragments is a file that opens in Illustrator and is nonetheless unworkable. The company publishes no quantitative measure of path economy, no complexity ceiling, and no compatibility caveats for specific editors — and, as the limitations section documents, several users report the opposite of the marketing claim for icon work. Treat "production-ready" as an assertion to test against your own use case in the first session, not as a specification.
The homepage states that "Creating a custom style is as easy as dropping in your images." This is the feature that separates a design platform from a novelty generator. A brand needs twenty illustrations that look like siblings, not twenty individually impressive images with nothing in common. Brand style consistency is what makes an AI generator usable for an actual campaign, and Recraft's positioning — closer to a design tool than to an art toy — rests on it.
Beyond generation, the platform bundles the operations a designer would otherwise scatter across several tools: vectorising an existing raster image, removing and generating backgrounds, inpainting and outpainting, erasing regions, crisp and creative upscaling, and mockup generation. Typography is called out separately as "Design-ready typography," which matters because text rendering is where most image models still fail visibly.
Recraft exposes the platform programmatically. The official documentation lists seventeen operations including generating raster or vector images, image-to-image, inpainting, outpainting, background replacement and generation, vectorising, background removal, crisp and creative upscale, region erase, remix, and style creation. Authentication is straightforward: "We use Bearer API tokens for authentication," with the key passed as Authorization: Bearer RECRAFT_API_TOKEN; generating a key requires that "your API units balance is above zero."
The pricing page adds that the API covers "image and vector generation, editing, background removal, inpainting, outpainting, batch jobs, and asynchronous processing," and that "All models across the V2, V3, V4 and V4.1 generations are available via the API." The documentation as retrieved states no rate limits and no per-operation unit prices, so this page does not assert any.
Icon sets and logo exploration. The single strongest fit on paper, because these are inherently vector deliverables. Read the limitations section before committing to this use — it is also where the sharpest user complaints cluster.
Marketing campaigns needing visual consistency. Custom styles let a series of assets share a visual language. For a brand producing weekly social graphics, consistency is worth more than any individual image's quality.
Print and large-format work. Vector output scales without resampling. If an asset might end up on packaging or a trade-show banner, starting in vector removes an entire class of problem.
Replacing stock illustration. Generating an illustration matching your brand style beats searching a stock library for something approximately right — provided you are on a paid plan, since free-tier output cannot be used commercially at all.
Product mockups. The Mockup Generator places artwork onto physical-product renders, serving sellers who need listing imagery without a photoshoot.
Programmatic pipelines. With seventeen API operations and batch plus asynchronous processing, teams generating visuals at volume — catalogue imagery, per-customer assets — can drive the platform from their own code.
Test the vector quality inside your first thirty credits. This is a hard-edged, practical reason and not a general caution: the published refund policy is discretionary and requires either no credits used since payment, or fewer than thirty credits used over the account's lifetime. Once past that line, your money is committed regardless of whether the output serves you. Run your hardest real test — the actual icon, the actual logo — immediately.
Never evaluate on the free tier if your purpose is commercial. Free output is owned by Recraft, published in the community gallery, and unlicensed for commercial use. You cannot generate a logo on free and later license it by upgrading; the asset was never yours.
Budget in credits, not images. Vectors cost two, mockups two, creative upscale twenty. A workflow that leans on creative upscale burns credits an order of magnitude faster than one that does not.
Decide about model training deliberately. Training on your inputs and outputs is enabled by default and opt-out. If you are feeding in client brand references under an NDA, change this setting before the first upload, not after.
Remember that unused subscription credits vanish. They "reset at the start of each billing cycle — unused credits don't roll over." Top-up credits, by contrast, never expire. For irregular workloads, a smaller subscription plus top-ups wastes less than a large subscription used half the time.
Do not delete your account while a refund is pending. The company states plainly it cannot issue refunds once an account is deleted, because it no longer holds the records.
Keep the source files. Export and archive the SVGs you rely on. Assets created during a paid subscription remain yours after cancellation, but your access to the platform's editing surface does not.
A good fit for: designers and brand teams who need vector deliverables and repeatable style; sellers needing product mockups and listing imagery; marketing teams producing consistent asset series; and developers building visual generation into a product via the API.
A poor fit for: anyone whose primary need is a single photorealistic image, where general-purpose generators compete on equal terms and often cheaper; casual users testing the water, because the free tier's ownership terms make it unsuitable as a commercial trial; anyone needing a documented accuracy or quality metric for vector output, since none is published; and buyers who require a straightforward refund path, given the thirty-credit lifetime threshold.
Requires deliberate thought from: studios handling client IP. Training on your inputs is on by default, the liability cap is the greater of $50 or six months of fees, and you indemnify Recraft against claims arising from assets you generate. Each term is common enough in this category; together they mean the commercial risk of a bad output sits with you, not the vendor. That is a decision to make explicitly.
Recraft runs primarily in the browser as Recraft Studio. The site's real paths include /pricing, /terms, /privacy, /about, /blog, /docs/api-reference/getting-started, /contact, /press-releases, /careers, /enterprise and /community.
An iOS app exists and was verified through Apple's official lookup interface. Published by Recraft, Inc. under the name "Recraft: AI Images and Videos," it is free to download, rated 4.8 out of 5 from 130 ratings, and carries a 17+ content rating. Version 3.2.11 was released on 20 August 2026 and requires iOS 17.6 or later. The app is localised into English only — worth noting given the platform's international user base. It first shipped on 19 June 2024.
A same-name caution applies here. App Store search also surfaces a separate application simply called "ReCraft," published by ReCraft ApS, a different legal entity with zero ratings. That product is unrelated to the tool described on this page. The anchor for identification throughout is recraft.ai and the publisher Recraft, Inc.
The API is documented at /docs/api-reference/getting-started and uses Bearer token authentication.
Recraft sells subscriptions metered in credits, with an option to buy additional credits on top.
How credits are consumed, per the company's own pricing-update post: a raster image costs 1 credit, a vector image 2 credits, a mockup 2 credits, and a creative upscale 20 credits. Credits "are reset daily for free users and monthly for subscribers."
Subscription versus top-up credits. The pricing page draws the distinction clearly: "Subscription credits are included in your plan and reset at the start of each billing cycle — unused credits don't roll over. Top-up credits are purchased separately, never expire, and are used automatically once your monthly credits run out." Top-ups are available on the Basic, Pro and Teams plans, and on Teams workspaces they are shared across all seats.
Billing. Annual billing is offered at a stated saving of 20% against monthly. Payment is by credit card, debit card or bank transfer, processed through Stripe.
On the actual tier prices: the plan names and amounts on the official pricing page are injected by client-side script and were not present in the retrievable page text. Rather than repeat figures from third-party summaries that could not be confirmed at source, this page does not state them. Check the official pricing page directly for current amounts.
Ownership is the most consequential pricing term, and it is stated unambiguously. For the free plan: "Images generated on the Free plan are owned by Recraft. They are publicly visible in the community gallery and are not licensed for commercial use. For example, they cannot typically be listed on stock platforms, which require proof of ownership." For paid plans: "Paid plans grant you full ownership and commercial rights for all images generated while subscribed... Images remain private and are not shown in the community gallery." After cancellation, "Images created during a paid subscription remain private and under your ownership even after the subscription ends."
Refunds are discretionary and narrowly gated. The stated policy: "We may offer a refund at our discretion where your most recent payment was made within the past 30 days and either you have not used any credits since that payment, or your account has used fewer than 30 credits in total since it was created." The terms of service put it more bluntly still: "If you cancel your subscription... you will not receive a refund or credit for any amounts that have already been billed or paid."
API billing is separate. Usage is metered in API Units, which are prepaid, purchased apart from subscription credits, do not expire, and are "non-cancellable and non-refundable except as required by applicable law."
Adobe Illustrator with Firefly is the incumbent for vector work, generating inside the tool where the paths will be edited, with enterprise-grade indemnification — the strongest alternative for organisations where legal comfort outweighs cost.
Midjourney produces arguably the most striking raster imagery in the category but offers no vector output, making it a different tool rather than a direct competitor for SVG work.
Kittl targets similar design ground with templates and a strong typography focus, and appeals to users who want more scaffolding and less prompting.
Canva covers the broader design workflow with AI generation attached; it is the pragmatic choice when generation is one task among many rather than the centre of the work.
Vectorizer.ai and comparable converters solve only the raster-to-vector conversion problem, which is worth knowing: if you already have artwork and need it in vector form, a dedicated converter may serve better than a generator.
Ideogram competes specifically on rendering legible text within images, an area where it has a strong reputation.
The competitive framing from the company itself is useful. Dorogush told TechCrunch that Recraft builds its own models from scratch and competes with Midjourney, Adobe Firefly, Stable Diffusion and Black Forest Labs, while positioning the product closer to design tools like Canva than to pure art generators. If your requirement is a single beautiful raster image, that is a crowded field and Recraft holds no structural advantage. If the requirement is an editable vector in a consistent brand style, the field narrows considerably.
Free-tier output is not yours, and this is the most important term on the page. The terms of service impose three restrictions on Free Tier Assets: "(i) no commercial use of Free Tier Assets is permitted; (ii) Free Tier Assets are owned by Recraft and may not be sold, licensed or transferred; and (iii) Free Tier Assets may not be used to train artificial intelligence models, systems or networks or similar technology." Breach results in "the immediate termination of this Agreement." Free output is also public in the community gallery, meaning your creative direction is visible to others while you explore.
Even paid ownership carries a restriction most users will not notice. The grant reads: "You own all Assets you create with the Services and Recraft hereby assigns to you all copyright rights it may have in the Assets subject to the following restriction: Assets may not be used to train artificial intelligence models, systems or networks or similar technology." Full commercial ownership, with one carve-out — you cannot use your own generated assets as training data.
Vector quality draws the sharpest user complaints, and they contradict the marketing directly. On Trustpilot, one reviewer wrote on 3 July 2026: "Tried to use it for SVG icons for toolbar. The results were absolutely unusable... All attempts to improve made icons even worse. In other words, SVG generation is far from usable in real work." Another, on 30 June 2026: "While attempting to create simple vectors (such as a star using reference images), the results repeatedly failed or produced completely different designs. Even straightforward instructions, like enlarging a vector or removing the background just redesigned it." Set these against the product page's claim of "production-ready" SVGs with "clean geometry, editable layers." Both statements are on the public record; the honest conclusion is that quality is workload-dependent and must be tested against your specific deliverable before you spend past the refund threshold.
Credits are consumed by failed generations too. Reviewers describe this repeatedly. One on 13 July 2026: "it intentionally eats credits on messed up results when it outputs exactly the same image not making any changes." Another on 12 August 2026: "The chat agent deliberately makes tiny progress on each prompt and forces you to repeat yourself over and over again... This eats credits and excludes you from any eligibility of refund." The imputation of intent is the reviewer's own and is not established; the mechanical fact underneath it — that iteration spends credits and spent credits foreclose refunds — is confirmed by the published policy.
The Trustpilot rating is very low, and the sample is very small. The profile shows a TrustScore of 1.7 from 28 reviews, with 89% at one star and 21 reviews in the past twelve months. The profile is unclaimed, and Trustpilot notes the company "hasn't invited their customers, so reviews may not be representative." Twenty-eight reviews is below any threshold at which a quality verdict can responsibly be drawn — treat this as a map of what goes wrong when it goes wrong, not as a measure of typical experience. The absence of vendor solicitation also means the sample carries no positive selection bias.
Ratings diverge sharply across channels. The iOS app sits at 4.8 from 130 ratings against Trustpilot's 1.7 from 28. Both samples are small, and they measure different populations: mobile app users who chose to install, versus desktop subscribers who paid and then sought refunds. Neither number should be reported alone.
The refund threshold is unusually tight. Thirty credits over an account's entire lifetime is fifteen vector images, or a single creative upscale plus ten rasters. It is entirely possible to exhaust the refund window during a genuine evaluation and discover only afterwards that the tool does not suit the work.
Your inputs train their models by default. The terms state Recraft "may use your Inputs you provide and the Assets produced by the Services to train, fine-tune, and otherwise improve Recraft's artificial intelligence models and systems. This training use is enabled by default, but you may opt out at any time by adjusting your preferences in your account profile settings."
Liability is capped at $50 or six months of fees. In the terms' own capitals, liability "WILL AT ALL TIMES BE LIMITED IN THE AGGREGATE TO THE GREATER OF FIFTY US DOLLARS (U.S. $50), OR THE AMOUNT PAID BY YOU TO RECRAFT IN THE SIX (6) MONTHS IMMEDIATELY PRIOR TO THE DATE OF THE EVENT GIVING RISE TO SUCH ASSERTED CLAIM."
You indemnify them. Users agree to "defend, indemnify, and hold harmless Recraft, its affiliates, and each of its and their respective officers, directors, employees, contractors, agents, suppliers, licensors and representatives" — including, expressly, against claims arising from assets you generate.
No published quality metric for vector output. The vector page asserts production readiness but offers no path-complexity ceiling, no geometry benchmark, and no compatibility caveats for particular editors. For the feature that is the product's core differentiator, that absence is a genuine gap.
Company-sourced user numbers conflict. TechCrunch relays 4 million users; the About page states over 3 million. Neither is audited. Where a vendor's own figures disagree, both are reported here rather than averaged.
Age requirement. The terms state in capitals that you must be "AT LEAST 18 YEARS OF AGE," and the iOS app carries a 17+ rating — notably stricter than the 4+ ratings typical of productivity software, and worth factoring into any deployment decision involving younger users.
Content restrictions. The terms prohibit generating non-consensual intimate imagery, "including AI-generated intimate images or deepfakes depicting real individuals without their consent," and separately bar using the service for automated decisions producing legal or similarly significant effects on individuals without human oversight.
There is a free tier, but read its terms before treating it as a trial. Images generated on it are owned by Recraft, appear publicly in the community gallery, and are not licensed for commercial use. Paid plans grant full ownership and commercial rights. Current tier prices are on the official pricing page; annual billing is stated to save 20%.
On a paid plan, yes: "Paid plans grant you full ownership and commercial rights for all images generated while subscribed," usable for marketing, branding and product packaging. On the free plan, no — commercial use of free-tier assets is expressly not permitted. One restriction applies to everyone: assets may not be used to train AI models.
It produces genuine SVG files, and the company states paths, strokes and fills "stay clean and accessible" and can be reopened in Figma, Framer or Illustrator. However, no quantitative measure of path quality is published, and several users report that SVG output for icon work was unusable in practice. Test your hardest real case within your first thirty credits, while a refund request is still possible.
Per the official pricing update: a raster image is 1 credit, a vector image 2, a mockup 2, and a creative upscale 20. Subscription credits reset each billing cycle and do not roll over; separately purchased top-up credits never expire.
Only within narrow limits and at the company's discretion: the payment must be within the past 30 days, and either no credits used since that payment or fewer than 30 credits used across the account's entire lifetime. Refunds cannot be issued after an account is deleted. API Unit purchases are non-refundable except where law requires otherwise.
By default, yes. The terms state training use "is enabled by default, but you may opt out at any time by adjusting your preferences in your account profile settings." The privacy policy applies a stricter standard to third parties: Recraft will not permit third-party AI providers or contractors to train on your AI input or output "unless you have affirmatively authorized" it. The looser default applies to Recraft's own proprietary models.
They remain yours. "Images created during a paid subscription remain private and under your ownership even after the subscription ends. They will not be made public. You keep full ownership and commercial rights to those images even after cancellation." Export what you need before losing platform access.
Yes. The documentation lists seventeen operations covering raster and vector generation, image-to-image, inpainting, outpainting, background replacement and removal, vectorising, upscaling and style creation, with models from the V2, V3, V4 and V4.1 generations available. Authentication uses Bearer tokens. API usage is billed in prepaid API Units, separate from subscription credits.
The legal entity is Recraft Inc., named in terms last updated 11 August 2026. The About page states founding in 2022 in the US with headquarters in London, UK; TechCrunch places the company in San Francisco. The CEO is Anna Veronika Dorogush, previously in machine learning at Yandex, Google and Microsoft. The company raised a $30 million Series B led by Accel in May 2025, after a $12 million Series A led by Khosla.
The profile shows 1.7 from just 28 reviews, and the complaints cluster tightly around refund refusals under the thirty-credit rule, credits consumed by failed generations, and disappointing SVG results. Two caveats matter in both directions. Twenty-eight reviews is far too small a sample for a quality verdict. But the profile is unclaimed and the company has never solicited reviews there, so the sample is at least free of vendor-driven inflation. Read it as a catalogue of specific, checkable failure modes — each of which you can test cheaply and early — rather than as a score.