Framer AI 2

用自然语言生成并发布生产级整站的 AI 建站平台

Vibe Coding

https://www.framer.com/

Visit Website

In-depth Report

Framer is a no-code website building platform built by a team in Amsterdam, the Netherlands. It has established a foothold in the designer circle with its "design and publish" canvas experience. The "Framer AI 2" included in tokenfly this time actually corresponds to the Framer 3.0 generation AI capability upgrade released in June 2026 - the core change is to plug the AI Agent directly into the canvas, so that natural language descriptions can generate the entire site structure, pages and components, and publish them to production-level hosting with one click. Its strengths are speed and design quality, but its pain points are vendor lock-in and a set of overlay billing that catches people off guard.
Framer was founded by Koen Bok and Jorn van Dijk as a prototyping tool Motif Tools around 2013 and later transformed into a website design and publishing platform. It has always been headquartered in Amsterdam. The company completed a $100 million Series D financing in August 2025, with a valuation of over $2 billion and a total of approximately $160 million in financing. The team is located in Amsterdam, San Francisco, New York and Barcelona, with a scale of about one to two hundred people. It was known for its interactive prototyping in the early days, and now positions itself as a "designer-led, AI-accelerated" website production and operation platform. Its customer list includes Perplexity, Miro, Scale AI, etc. CB Insights ranks it as the leader in the AI website builder track, along with Webflow, Wix, and HubSpot.
The release of Framer 3.0 threw together Agent, Branching, new community and design. It ranked first on Product Hunt on June 17, 2026, receiving 509 votes and a reputation of 4.84 points. The core of the so-called "Framer AI 2" experience is to allow AI to no longer stand outside the door delivering pictures, but actually work inside the room: the user selects several layers on the canvas as context, describes the requirements in one sentence, and the Agent directly generates or rewrites pages, components, styles, CMS content, SEO metadata and even the release process; every change is a native Framer project, which can be checked, fine-tuned, branched, and online. Specifically, the AI suite contains several aspects: Wireframer is responsible for growing a full-page structure with hierarchies and responsive breakpoints from natural language, eliminating the fear of a blank canvas; Workshop is responsible for turning an English requirement into a reusable interactive component; AI Translate translates the entire site into multiple languages with one click; AI Plugins allows access to external models such as OpenAI, Anthropic, Gemini, etc.; Generate Images supplements materials. More importantly, External Agents can directly connect the Claude Code, Codex, Cursor, and Gemini CLI already used by the team to Framer, drive site updates from the terminal or chat, publish PR from GitHub, and change copywriting from Slack. Branching installs a fuse for Agent changes. The team can first open a branch to test the layout and test the variants, confirm that they are correct, and then merge them online to avoid directly affecting the production environment. The underlying platform capabilities are also complete: the visual canvas is close to Figma's layer and attribute panel; the built-in CMS manages structured content; the SEO tool generates meta, sitemap and structured data; the built-in analysis is cookie-free and GDPR compliant, and EU companies are exempt from consent banners; it is hosted on a global CDN, automatic SSL, and the official nominal Core Web Vitals is good (LCP 1.1 seconds, INP 95 milliseconds). Perplexity reportedly built the entire site with zero developers, and about 85% of Superhuman’s site work was completed by the design and marketing teams themselves.
Framer will simplify the seven pricing levels to five levels in October 2025. By mid-2026, the prices will be roughly: Free $0 (including framer subdomain and "Made in Framer" logo, but open to all AI tools, 1000 pages and 10 CMS collections); Basic monthly payment is $15, equivalent to annual payment of $10; Pro monthly payment is $45, annual payment is equivalent to $30 USD, plus advanced analytics, A/B testing, password protection, custom code injection, and 10 CMS collections; Scale is only $100 per year and scales by usage; Enterprise is customizable. What really drives up the actual bill are the add-ons: additional editor seats are $20 at the Basic level and $40 per person per month at the Pro/Scale level; each additional translation language is $20 per month; A/B testing is added based on the amount of events. A common real-world scenario is that for Pro plus 3 editors plus 2 languages, the monthly fee jumps from the list price of $30 to about $190. The Free version is enough to test the idea. If you really want to bind a custom domain name and go online, the minimum is Basic; but Basic only has 1 CMS collection, and most real sites will be pushed to Pro.
The front sound focuses on speed and texture. Many developers say that it only takes 2 to 3 hours from conception to launch of the landing page, and the output is not like a template but like customization; animation and micro-interaction are recognized as the best in its class; Figma users can be migrated at almost zero cost; AI tools are fully open in the free version, which is very popular. Some people in the community use Workshop to make paid components and Wireframer to quickly build shareable site prototypes, saying that the iteration feels "superior". The negative voices are equally clear. "Vendor lock-in is driving me crazy" is a frequently used sentence on Reddit; on Trustpilot, Framer has only 1.7 points, with negative reviews focusing on opaque billing, hidden seat fees, and cancellation friction rather than design quality; G2 has about 4.3 points, and Capterra has about 4.4 points, which are more positive. A two-star Trustpilot review is typical: AI frequently produces results that deviate from expectations on complex requests, but the author also fairly admits that the speed of no-code website building is indeed industry-leading.
From an industry perspective, Framer is regarded as a design representative among AI website generators. It, Wix ADI, and Figma AI can automatically create layouts, but Framer retains canvas-first in-depth editing and is likened to "a design tool that will grow into a complete product platform." The media generally recognizes that it completely smoothes the development handover between design and release. Entrepreneurial teams can sell strategies, do design, and deliver real websites in one tool. Shortcomings have also been repeatedly named: CMS is not as deep as Webflow and WordPress, and is not suitable for content-intensive or e-commerce-focused sites; there is no native e-commerce, and it can only rely on third-party embedding.
The controversy mainly revolves around three things. The first is vendor lock-in: the site cannot export clean HTML/CSS, the design is locked in a proprietary runtime, the larger the scale, the higher the migration cost, and the less leverage users have when the platform raises prices. The second is pricing transparency: there is a big gap between the listed price and the actual bill. Seats and languages are superimposed on a volume basis. The monthly fee for complex sites can easily exceed one hundred or even two hundred dollars, while self-hosted static site hosting of similar quality may only cost a few to twenty dollars per month. The third is the uncertainty and cost of AI itself: Agent output is unstable under complex prompts, and one prompt may burn about 1,000 credits (about 10 US dollars), which puts obvious pressure on budget-conscious individual creators; the community is also worried that the Agent will "drift" during automatic revision and destroy the responsive layout, so system pages and branching mechanisms need to be designed to avoid this.
This set of tools is most suitable for three types of people: entrepreneurs who need to launch a beautiful landing page or brand official website within a day; freelance designers who regard design quality as their business card and want the most creative control; and small studios and agencies that quickly deliver high-looking customer sites. It is not suitable for e-commerce sellers (no native e-commerce), content-intensive site operators (CMS is not deep enough), and budget-sensitive large teams that require the collaboration of multiple editors (seat fees can quickly get out of hand). If you want to own the code yourself and be able to move it out at any time, Webflow, WordPress or a lower-level solution is more stable.
Framer 3.0 advances AI from "drawing pictures in one sentence" to "working in the room". The generation and release of the entire production-level site is truly closed-loop, which is a real acceleration for design-driven teams. However, the two hidden costs of locking and superimposed billing determine that it is more suitable for teams that treat the site as an ongoing operating asset rather than a one-time deliverable.

User Reviews

AHillQ2

—

The Agent's step into the canvas is really amazing. It finally doesn't generate a piece of HTML and then paste it into the wall.

آریارضایی

—

Migrated from Wordpress, a simple click becomes an entire task in Framer, so forget about complex content sites.

Kelly_Gutierrez_8806

—

It’s a good thing that the editor’s seats have been reduced from 40 to 20 dollars, but AI credits have another cost, which essentially changes from buying seats to paying for electricity.

JonathanThompson

—

I was tricked by hidden charges once, and I was silently charged $30 for inviting a collaborator to edit. There was no prompt at all during the invitation process. Unsubscribing is also difficult, the customer service page is still old, and the chatbot is completely useless. I have nothing to say about the quality of the design, but the billing transparency and unsubscription friction are really discouraging. It is recommended that the team should figure out the billing logic clearly before collaborating with multiple people.

KSanchezX

—

The free version is 500 points per day. It’s enough to test the water by making a landing page. Let’s talk about it first.

Bruce_Johnson369

—

I do think the lockdown argument is a bit over the top. If something goes wrong and the website goes down at 2 a.m., a closed source platform at least has an SLA and a support team to answer the phone; no one knows the details of the stack that I vibed out. To a growing company, the maturity of the platform and the company behind it are worth more than theoretical portability.

Elizabeth_Nguyen_9915

—

I use Workshop more than Agent, mainly because Agent takes credits every time, which is a bit distressing.

lAUREN158

—

The token was burned so hard that one prompt killed a thousand points.

CMartinezX

—

The first draft generated by AI is amazing, but it does go astray under complex requirements, and you often have to finish it manually. It’s more realistic to think of it as an intern rather than a deliverer.

Kimberly126

—

It is said that the concern about Agent drift is not unreasonable. Automatically changing the layout may really blow up the responsiveness. Fortunately, Branching is equivalent to a fuse, letting it toss in the branch first.

Catherine.Howard168

—

After Framer changed AI billing to credits, it cost about 300 points to generate an entire landing page. Opus was 80% more expensive than Sonnet. The free version resets every day and does not accumulate. If you really want to build a website on a large scale, you still have to buy Pro or add additional purchases. The drop-down action of selecting a model actually turns "which model to use" into a cost item that you check yourself, which is quite honest.

海浪422

—

The summary of 255 reviews on Product Hunt illustrates the problem: the advantages are concentrated on ease of use, no code, and fast speed. The disadvantages are also very honest - complex configuration, limited CMS capabilities, high pricing, and slow support response. It is good at putting beautiful marketing websites online quickly, but it may not be suitable for carrying complex business systems. Before buying, you must first think clearly whether you want a marketing website or a business system.

brownbird165

—

Before a single person started a studio, the amount of work he could do was very limited. Now Agent takes care of all the chores such as adapting the desktop and moving breakpoints, cleaning fonts, and writing alt text. It is like having an extra helper that does not require management. The real fun part is left to the designer himself, and the repetitive work is left to it, and the efficiency improvement is real.

Alice_LopezIII

—

It's great to have External Agents connected to Claude Code. You can change the CMS of the client site from the terminal, eliminating the need for context switches. The website has changed from an app that needs to be clicked into an endpoint that can be called by scripts.

dq49fcbz5s

—

Branching is great for our small team. We open branches first when revising the version, and then merge them after confirming that there are no problems. We are no longer afraid of moving to the online site. Especially suitable for non-profit projects that frequently try out different layouts.

Gloria980

—

Throw in a picture reference and let it generate a navigation bar. Agent does a decent job.

NashPowell777

—

In comparison, Webflow is like in ancient times. This version of Framer completely connects design to release.

MReed

—

As a designer, using Framer is really smooth. The interface is so similar to Figma that there is almost no barrier to getting started.

Shirley_Nelson_77

—

Branching is a god. It is very important to let AI branch first and not start production.

Framer AI 2

In-depth Report

User Reviews

Similar Products