# Diffusion Studio Aiolum story: https://aiolum.com/story/diffusion-studio/ · added 2026-11-06 · updated 2026-11-06 **In short.** Diffusion Studio (Open-source video editor that coding agents can drive, with paid credits for AI models, Germany (Y Combinator listing); Delaware company, site footer says San Francisco): 3.5K GitHub stars on the open-source editor, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 people on the Y Combinator page; founder Konstantin Paulus. Launched June 2023; still running. Open-source video editor that coding agents can drive, with paid credits for AI models · Germany (Y Combinator listing); Delaware company, site footer says San Francisco · AI-native · team of 2 to 10 · still running · group: Video generation - Outcome: 3.5K GitHub stars on the open-source editor, November 2026 (evidence grade of the headline number: E; no revenue figure) - Launch: June 2023; months from launch to the outcome: 41 - Team: 2 people on the Y Combinator page; founder Konstantin Paulus - Outside funding: Y Combinator Fall 2024 batch; amount not published - Main factor: Distribution - Meters, 1 to 4: evidence strength 1, copyable pattern 2, capital needed 2 - Tags: AI-native, Small team, Funded, Running, B2C, Video editing Links: - Visit diffusion.studio: https://diffusion.studio/ **The product.** A video editor for macOS, Windows and the browser in which every project is a set of code files, so a coding agent such as Claude Code, Codex or Cursor can read the project, cut clips, add captions and motion graphics, and export the video. The editor itself is free and open source; the paid plans sell credits for AI actions such as transcription, footage analysis and generating video, images and voice. **Why it worked.** Coding agents became good at writing code in 2025 and 2026, but video editors keep their projects in closed binary files that an agent cannot read. Diffusion Studio stores each composition as JSX/TSX code, so the same agent that edits a website can edit a video, and a person can still change the result on a canvas and a timeline. Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts. ## Summary Diffusion Studio is a non-linear video editor, a program in which clips, text and effects are placed on a timeline and can be rearranged at any time. Its projects are written as JSX/TSX files, the component markup used by web developers, and the app exposes its tools through MCP (Model Context Protocol, a standard way for an AI agent to call the tools of another program) and through a command-line tool called dapi. A coding agent can therefore inspect footage, edit the code of a scene and render it, while a person edits the same project by hand. The company's GitHub organization was created on 6 June 2023, and Y Combinator lists Diffusion Studio in its Fall 2024 batch with a team of 2 and Konstantin Paulus as founder and CEO. An earlier open-source engine, a browser video compositing library built on the WebCodecs API (the browser interface for decoding and encoding video frames), came first; the current editor became an open alpha on 4 May 2026 and a Mac app with agent support on 10 August 2026. No revenue figure was found. The headline is the number of GitHub stars on the open-source editor, 3,474 on 6 November 2026, a measure of developer interest and not of sales. The pricing page shows paid plans from US$20 a month billed yearly, so the business does sell today. ## Pricing Diffusion Studio prices on its pricing page, read 6 November 2026 (grade C); monthly prices from the page's structured price data | Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit | |---|---|---|---| | Free | 0 | 0 | 50 credits one time; full editor, unlimited projects and exports | | Pro | 20 | 25 | 2,500 credits a month | | Plus | 35 | 50 | 5,000 credits a month, priority support | | Max | 100 | 150 | 15,000 credits a month, priority support | The editor, exports up to 4K without a watermark, the dapi CLI and agent editing are free on every plan. Paid plans only add credits, which are used when Diffusion Studio runs AI models: transcription, footage analysis and generation of video, images, audio and speech. Unused monthly credits do not roll over; credits can be topped up. ## How much revenue does Diffusion Studio make? The numbers, graded A free open-source video editor for agents and people, funded by Y Combinator, with paid AI-credit plans from US$20 a month and no published revenue | Fact | Value | Grade | Note | Sources | |---|---|---|---|---| | GitHub stars on the editor repository, 6 November 2026 | 3,474, with 320 forks | C | GitHub's public API for diffusionstudio/editor, read 6 November 2026. The repository was created on 7 July 2026 and is licensed under MPL 2.0. Stars show interest, not paying users. | [4] | | GitHub stars on the Lottie animation skill, 6 November 2026 | 5,694 | C | Repository diffusionstudio/lottie, created 4 June 2026, described as "Generate production-ready Lottie animations with Claude Code or Codex". | [5] | | GitHub organization created | 6 June 2023 | C | GitHub's public data for the diffusionstudio organization. The first own repository, ffmpeg-js, dates from 28 June 2023; Y Combinator gives 2023 as the year founded. | [5, 3] | | Months from the organization's creation to the star count | 41 | calc | Calculated: June 2023 to November 2026. | [4, 5] | | Earlier engine | "A fast, browser based video compositing engine powered by WebCodecs", 1,251 stars | C | Repository diffusionstudio/core, created 30 July 2024, last pushed 18 November 2025. | [5] | | Team | 2 people | C | Team size on the Y Combinator company page, read 6 November 2026. One founder is named there: Konstantin Paulus, Founder/CEO, formerly at Rohde & Schwarz in Munich. | [3] | | Funding | Y Combinator, Fall 2024 batch | C | Y Combinator company page. No downloaded source gives the amount or another investor. | [3] | | Company | Diffusion Studio Inc., Wilmington, Delaware | C | Terms and conditions, with registration number 7698816. The site footer says the product is built in San Francisco; Y Combinator lists the location as Germany. | [7, 1, 3] | | Prices | Free; US$20, US$35 and US$100 a month billed yearly (US$25, US$50 and US$150 monthly) | C | Pricing page. Paid plans sell AI credits only; the editor and agent workflows are free. | [2] | | Payment processor | Stripe | C | Terms and conditions: "All payments are processed by Stripe." | [7] | | Coding agents named as supported | Claude Code, Codex, Cursor, VS Code with Copilot, Antigravity, Gemini CLI, Devin (Windsurf) | C | Home-page FAQ. Other agents can connect through MCP. | [1] | | Third-party AI providers | OpenAI, Google (Gemini/Veo), ElevenLabs, AssemblyAI, Fal.ai | C | Terms and conditions, list of integrated providers, together with Supabase, Google Cloud, AWS, Fly.io, Sentry and Umami. | [7] | How to read this table: grade C is what the company writes on its own site, on its Y Combinator page or in GitHub's public data. A "calc" value was computed on this page. No value is grade A or B, and no row is a revenue figure. ## When did Diffusion Studio launch, and how did it grow? Timeline From a GitHub organization for browser video tools in June 2023 to a desktop editor that coding agents control, in about three and a half years. Each entry carries its grade. - **June 2023** (grade C [5]): The diffusionstudio GitHub organization is created on 6 June; the first own repository, an FFmpeg build for the browser, follows on 28 June. - **2024** (grade C [5, 3]): The open-source WebCodecs compositing engine "core" is published on 30 July, and the company joins Y Combinator's Fall 2024 batch. - **4 May 2026** (grade C [6]): A rebuilt editor around an infinite canvas opens as an alpha in the browser, with auto-captions and generation inside the editor. The previous editor stays available as a legacy version. - **10 August 2026** (grade C [6]): Diffusion Studio for Mac and the dapi CLI: coding agents such as Claude Code, Codex and Cursor can edit videos in the desktop app. - **27 August 2026** (grade C [6]): Projects move to local folders, with code as the database of a video project and Git to track changes. - **September to November 2026** (grade C [6, 4]): Windows app on 23 September and AI object masking on 29 September. On 6 November the editor repository has 3,474 GitHub stars. ## Why did Diffusion Studio work? ### Projects as code A video composition is a JSX/TSX file. Changes made in the code update the editor, and changes made on the canvas or timeline update the code, so an agent and a person can take turns on the same project without exporting anything. ### Free editor, paid models The editor, exports and agent workflows cost nothing. Money comes from credits that pay for model calls: transcription, footage analysis and generated media. The company's cost of each paid action is a model provider's bill, so price and cost move together. ### Open source as the sales channel The editor, the desktop app, the CLI and the agent skills are on GitHub under MPL 2.0. Developers find the repositories, install the app with Homebrew or a skills command, and the agent then uses credits when it needs a model. ## How it was built These details come from the company's GitHub README, terms and site. They are grade C. - The editor UI is written in Solid with Vite; the desktop app is an Electron shell around it. - A headless runtime package decodes media and captures frames without a browser window, so the same engine renders for the app, the CLI and agents. - Agents reach the app through its MCP server or the dapi CLI, with commands such as probe, grab, filmstrip, waveform, transcribe and listen to inspect media they cannot watch. - Model work is bought from third parties: OpenAI, Google Gemini and Veo, ElevenLabs, AssemblyAI and Fal.ai. - Accounts and data run on Supabase, Google Cloud, AWS and Fly.io; payments run on Stripe. Sources: [4, 7, 1] ## Who tried this before Diffusion Studio, and why did they fail? Four companies in our database of failed startups, three of them also from Y Combinator, promised easier video or media editing and failed. The values below are from that database and are not independently verified. The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade. ### ReelSurfer (2012 to 2015; funding US$1.25M; cause: Competition) - Promise: Clip and share segments of online videos without editing skills. - Why it died: YouTube, Facebook and mobile apps built clipping in, and it found no way to earn money from it. - Record: ### HighlightCam (2009 to 2012; funding US$500K; cause: Competition) - Promise: Automatic editing of raw footage into a short highlight video. - Why it died: Smartphone apps and do-it-yourself software did the same job before the market was ready to pay. - Record: ### Snipshot (2006 to 2009; funding US$20K; cause: Competition) - Promise: Edit photos in the browser without installing software. - Why it died: Larger rivals and free tools offered more, and simplicity alone was not enough. - Record: ### ClearMix (2020 to 2023; funding US$2M; cause: Competition) - Promise: High-quality marketing video production without a studio budget. - Why it died: Canva and Adobe offered self-service tools, and AI made its manual production process obsolete. - Record: ## What changed The four predecessors sold an easier editor to people, and larger platforms or free tools added the same feature. Diffusion Studio gives its editor away and sells the model calls that an agent makes inside it, so a free competitor does not remove its paid part. The other change is the user: coding agents that can read and write code files did not exist when these companies ran. An editor whose projects are code is useful to such an agent in a way that a closed project file is not. ## Can one person copy Diffusion Studio now? **Verdict: Partly. The open-source editor is years of engine work; the business model is copyable.** A solo founder would not rebuild a 4K video engine quickly. The structure can be copied in other fields: make a tool's files readable by coding agents, give the tool away, and charge credits for the model calls the agent makes through it. 1. Pick a creative or office tool whose files coding agents cannot read today. 2. Store projects as plain code or text, and expose the tool's actions through MCP and a CLI. 3. Publish the tool as open source so developers find it on GitHub. 4. Sell credits that pay for model calls, priced above the providers' cost. **What is still open.** Slide decks, diagrams, audio projects and 3D scenes edited by coding agents, with credits for the models they call. ## What could go wrong for Diffusion Studio? - **What is not verified.** No revenue figure was found. There is no TrustMRR profile under the name, no founder interview with figures among the downloaded sources, and no filing. The headline, 3,474 GitHub stars on 6 November 2026, measures interest from developers, not sales. A reader could look for a revenue statement by the founder, a payment-processor profile or a funding announcement with figures. - **Unchecked numbers and logos on the site.** The home page shows "100K+ Creators and counting" and a row titled "Folks onboarded from" with ElevenLabs, Warner Bros. Discovery, Ramp, X, Microsoft, Canva, fal, Lovable and Shopify. No downloaded source explains how the 100,000 were counted or what the named companies use; the founder's Y Combinator bio separately mentions a social-media audience of over 100,000. - **Dependence on model providers.** The paid credits pay for third-party models: the terms name OpenAI, Google (Gemini/Veo), ElevenLabs, AssemblyAI and Fal.ai. A price change or a feature from one of these providers changes Diffusion Studio's margin or its offer. The editor also depends on coding agents from other companies, such as Claude Code and Codex, to do the agent work. - **Generated voice and other people's likeness.** The paid plans include generated voice and speech. No page read offers voice cloning. The terms forbid content that impersonates a person, "including the creation of misleading deepfakes of real individuals", and make the user responsible for releases for the likeness of identifiable people. - **Competition.** On this site, Hera also sells an AI motion-graphics canvas and BlitzCut AI edits video by chat; Diffusion Studio differs in giving the editor away as open source and selling only model credits. ## Sources 1. Diffusion Studio home page, read 6 November 2026 — https://diffusion.studio/ (grade C). Read in full. Source for the product description, the supported agents, the free and open-source statements, the 100K+ counter and the logo row. 2. Diffusion Studio pricing page, read 6 November 2026 — https://diffusion.studio/pricing (grade C). Read in full, including the structured price data for the monthly prices. Source for the plans and the credit rules. 3. Y Combinator company page: Diffusion Studio — https://www.ycombinator.com/companies/diffusion-studio (grade C). Read the company data on the page. Source for the batch, the team size of 2, the year founded, the location and the founder. 4. GitHub: diffusionstudio/editor repository and its public API record, read 6 November 2026 — https://github.com/diffusionstudio/editor (grade C). Read the README and the API record (api.github.com/repos/diffusionstudio/editor). Source for the star count, the creation date, the license and the technical stack. 5. GitHub: diffusionstudio organization and its repository list, read 6 November 2026 — https://github.com/diffusionstudio (grade C). Read the API records (api.github.com/orgs/diffusionstudio and its repos list). Source for the organization date and the older repositories. 6. Diffusion Studio changelog, with the Open alpha entry of 4 May 2026 — https://diffusion.studio/changelog (grade C). Read the Markdown index and the Open alpha entry in full. Source for the release dates. 7. Diffusion Studio terms and conditions — https://diffusion.studio/legal/terms-and-conditions (grade C). Read in full (Markdown version). Source for the company, Stripe, the third-party providers and the deepfake rule. 8. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors. Contact: aiolum@livyden.com. Run Diffusion Studio? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Diffusion Studio).