Clarity AI: Revenue, team and how it was built
Clarity AI (AI image and video upscaler with a web app and an API, built on the founder's open-source model, Not stated for the business; the founder's GitHub profile says he is from Germany and lives in Thailand, and the terms name the laws of Singapore): US$5K MRR, July 2024. Evidence grade D: reported, not verified. Team: 1 founder, by the site's own account ("built by its founder, Philip"); no employees reported. Launched May 2024; still running.Added 27 October 2026
- Outcome
- US$5K MRR, July 2024
- Evidence
- D reported, not verified
- Launch
- May 2024
- Status
- Still running
- Team
- 1 founder, by the site's own account ("built by its founder, Philip"); no employees reported
- Outside funding
- None reported
- Group
- Photos and portraits
- Main channel
- Open source
What is not verified
The headline US$5,000 of monthly recurring revenue rests on one founder post on X, which this page could read only as eBiz Facts quoted it on 1 July 2024, and the post itself gives no date beyond "two months after launching". No payment processor, filing or second interview with figures was found, and no later revenue figure exists in the sources. A reader could look for a payment-verified profile, for example on TrustMRR (the site takes payments through Stripe), or for a newer interview with the founder. The headline number is grade D: one claim or one estimate, with no second source. Aiolum shows it because the story is useful, and marks it so that it is not mistaken for a verified figure.
Clarity AI is an AI upscaler: a program that takes a low-resolution image and produces a larger one, using an image model that draws in plausible new detail instead of only stretching the pixels. Its open-source version is built on Stable Diffusion, an open image-generation model, and the user decides how much new detail the AI may add. The founder first released the code as open source, code that anyone may read, run and change, here under the AGPL-3.0 license, on GitHub on 15 March 2024; the repository had 5.2 thousand stars in October 2026, and the public model on Replicate, a service that runs AI models for a fee per use, showed 30.7 million runs.
The site's FAQ says that Clarity AI "was built by its founder, Philip", who posts on X as @philz1337x; no source gives his surname. In a post that eBiz Facts quoted on 1 July 2024, he wrote that he had been a mechanical engineer in Switzerland earning US$100,000 a year, quit, learned to code and to build AI models, and worked with the founder Pieter Levels to build and publish models on Replicate. He wrote that two months after launching, the app was making US$5,000 of MRR, monthly recurring revenue, enough to finance his life.
That US$5,000 figure rests on one founder post and no payment processor has confirmed it, so this page is marked "Reported, not verified" with evidence grade D. In October 2026 the site sells plans from US$19 to US$299 a month and a pay-as-you-go API, and lists Modal, fal and Replicate as its GPU providers. The page also shows earlier photo-editing companies that failed.
How much revenue does Clarity AI make?
Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.
| Fact | Value | Grade | How we know | Sources |
|---|---|---|---|---|
| Monthly recurring revenue, by 1 July 2024 | US$5,000 | D | The founder's post on X as quoted by eBiz Facts on 1 July 2024: "Two months after launching, it's making $5,000 MRR, and I can finally finance my life with my own projects." X was not opened, and no second source gives a revenue figure. | [1] |
| Run rate from that figure | US$60,000 a year | calc | Calculated: US$5,000 times 12. | [1] |
| Earlier job | mechanical engineer in Switzerland, US$100,000 a year | D | The same post: "Before that I worked as a mechanical engineer in Switzerland and made $100k per year, but it was so boring." | [1] |
| Work before the app | built and published models on Replicate with Pieter Levels | D | The same post: "Then I worked with @levelsio to help him build and push models to Replicate". eBiz Facts adds that Levels runs PhotoAI and InteriorAI. | [1] |
| Open-source release | 15 March 2024, AGPL-3.0 license | C | The repository README lists "03/15/2024: Code release" and later updates, the first with a date in month/day/year order. | [7] |
| GitHub repository, 27 October 2026 | 5.2 thousand stars, 532 forks | B | GitHub's counters for philz1337x/clarity-upscaler, described as "free and open-source Magnific Alternative". | [7] |
| Public model on Replicate, 27 October 2026 | 30.7 million runs; about US$0.019 per run | B | Replicate's page for philz1337x/clarity-upscaler, on Nvidia A100 (40GB) hardware, with typical runs of 17 seconds. The fee goes to Replicate; no source says whether the founder receives any of it. | [9] |
| Launch of the paid app | about May 2024 | calc | Calculated: two months before the founder's post, which eBiz Facts quoted on 1 July 2024. The first archived copy of clarityai.co with content is from 19 May 2024; earlier captures in December 2023 and March 2024 are redirects. | [1, 10] |
| Team | 1 person | C | The site's FAQ: "Clarity AI was built by its founder, Philip." Support answers say "mail me" and "message me on X at @philz1337x". | [2] |
| Founder's identity | first name Philip; @philz1337x on X and GitHub | C | No downloaded source gives a surname or a registered company. The GitHub profile says "AI developer from 🇩🇪 Germany living in 🇹🇭 Thailand building open source AI apps". The terms name the laws of Singapore. | [2, 8, 4] |
| Usage claimed by the site | "over 30 million upscales" | C | Text on the pricing page in October 2026. The figure is close to the 30.7 million runs of the public model on Replicate; no source says whether they are the same count. | [3, 9] |
| Model and hosting providers | Modal, fal.ai and Replicate for GPU inference; OpenAI for image descriptions and prompt enhancement | C | The site's subprocessor list, published 24 September 2026, also names Cloudflare, Google Cloud and Firebase, Stripe, MailerSend and Rewardful. | [6] |
| Rules for users | 18 or older; no sexual content involving minors, no non-consensual intimate imagery, no impersonation | C | Terms of service, updated 20 July 2026. The privacy policy says uploads are not used to train AI models. | [4, 5] |
| Time from launch to the headline figure | 2 months | calc | Calculated from the founder's own words: about May 2024 to the post quoted on 1 July 2024. | [1] |
How to read this table: a grade D value is a founder statement that only one post contains, quoted by eBiz Facts. A grade B value is a counter that GitHub or Replicate shows. A grade C value comes from the company's own pages and code, or from the founder's README and profile. A "calc" value was computed on this page from the rows above it.
| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| API | 0 | pay as you go, from US$0.005 per megapixel | |
| Starter | 13 | 19 | 1,200 credits a month, commercial use, API access |
| Pro | 33 | 49 | 4,000 credits a month, adds Flux upscaling and ComfyUI |
| Business | 199 | 299 | 30,000 credits a month, custom options and support |
The pricing page loads by script; the prices were read from the page's own code on 27 October 2026. Yearly billing is US$156, US$396 or US$2,388 a year, which the page shows as US$13, US$33 and US$199 a month and calls a saving of about 30%. The API has no subscription and is charged per output megapixel: US$0.005 for the Clarity upscaler, US$0.016 for Crystal and US$0.03 for Clarity Pro, and US$0.10 per megapixel-second of video for Crystal Video. Unused, recent payments can be refunded by the user within 14 days, the FAQ says.
When did Clarity AI launch, and how did it grow?
From a mechanical engineering job in Switzerland to an open-source upscaler and a paid app, as the founder tells it. Each entry carries its grade.
- Before March 2024DThe founder leaves a job as a mechanical engineer in Switzerland, learns to code, to build websites and to use Stable Diffusion, and sees his first company fail, he wrote. eBiz Facts notes that his GitHub account dates from 2022. [1]
- Before March 2024DHe helps Pieter Levels build and publish AI models on Replicate, according to the same post. [1]
- 15 March 2024CCode release of the open-source Clarity upscaler on GitHub. [7]
- March to June 2024CREADME updates: anime upscaling, LoRA support, a ComfyUI node, resolution up to 13,000 by 13,000 pixels, pattern upscaling. [7]
- 19 May 2024CFirst archived copy of clarityai.co with content, titled as an upscaler with "API and ComfyUI Support". [10]
- By 1 July 2024DUS$5,000 MRR two months after launch, according to the founder's post quoted by eBiz Facts. [1]
- 5 March 2025CFlux upscaling is offered only through the paid app and API, not as open source. [7]
- 20 July 2026CNew terms of service and privacy policy. [4, 5]
- 24 September 2026CThe site publishes its list of subprocessors. [6]
- 27 October 2026BPlans from US$19 to US$299 a month and a pay-as-you-go API; 30.7 million runs of the public model on Replicate. [3, 9]
Why did Clarity AI work?
Four causes, in the order they mattered. Each one is a reusable lesson.
Free code as the advertisement
The founder released the upscaler as open source and called it a free alternative to Magnific, a paid upscaler. The README tells readers that the easy, paid options are his app and API, and the Replicate model page links to the app. Developers who found the code also found the hosted version.
Paid convenience on top of free code
Running the open-source model needs a GPU and setup work. The app sells the same result without either, and later modes such as Crystal, Clarity Pro and Flux upscaling are closed source and available only through the paid app and API.
Skills learned on someone else's products
The founder wrote that he built and published models on Replicate for Pieter Levels before building his own app, and that working with Levels taught him to focus on the right things. He launched with the model work already done.
Credits and per-megapixel prices
Subscriptions give monthly credits and the API charges per output megapixel, so larger images cost more. Each job runs on rented GPUs at Modal, fal or Replicate, so the cost of serving a customer rises with use and the founder owns no hardware.
How it was built
These details come from the founder's README and profile, GitHub and Replicate (grades B and C), the site's own code and pages (grade C), and the founder's post (grade D). [7, 9, 6, 2, 3]
- The open-source upscaler is built on the AUTOMATIC1111 Stable Diffusion web UI and packaged with Cog, Replicate's tool for model containers; it also runs in ComfyUI and A1111.
- The public model runs on Replicate on Nvidia A100 GPUs. The paid service sends jobs to Modal, fal.ai and Replicate, and uses OpenAI only for image descriptions and prompt enhancement.
- Files are stored on Cloudflare R2; the backend, job processing and logins run on Google Cloud and Firebase.
- Stripe takes payments, MailerSend sends account emails, and Rewardful tracks the affiliate program.
- The web app is a single-page JavaScript application; the prices and texts of the site are part of its code.
Tools named in the sources
Each one opens its page in the Tools tab, with every story that uses it.
Who tried this before Clarity AI, and why did they fail?
Four companies in the Loot Drop database sold photo editing or creative image tools and failed. The values below are from that database and are not independently verified.
Picnik
The promiseSimple photo editing in the browser, without installing software.
Why it diedGoogle bought it in 2010, folded its technology into other products and closed it in 2012, while Adobe and mobile editing apps took the market.
Snipshot
The promiseEdit photos in the browser without the Adobe fuss.
Why it diedLarger free tools such as Picnik offered more, and it never found a clear way to make money.
Splashup
The promiseOnline photo editing with real-time collaboration.
Why it diedIt was built on Flash as the web moved to HTML5, and it had no clear business model.
FiftyThree
The promiseA digital sketchbook for the iPad with its own stylus.
Why it diedApple's own Pencil and stronger drawing apps took away what made it different.
The failed startups come from Loot Drop, a free database of dead startups, retold here in our own words. Loot Drop describes its entries as AI-assisted summaries of public sources, so these cards carry no evidence grade.
What changed
Picnik, Snipshot and Splashup put simple photo editing in the browser and lost to larger free tools or to a change of web technology. FiftyThree built its own hardware and lost when Apple shipped the same idea.
An upscaler built on an image model adds detail that was never in the file, as the site describes it. The founder used free, open models, published his version as open source, and rented GPUs for each job, so one person could offer it. The same openness is the risk: anyone can run the free code, and model platforms list many competing upscalers next to his.
Can one person copy Clarity AI now?
Yes, if you can build the model work yourself.
The method is clear and was repeated by others: publish a useful model as free code, show the results on X, and sell a hosted app and an API to everyone who does not want to run it. It needs real skill with image models and GPUs, and the founder learned that by building models for another founder first.
Where it is still open: The same pattern for other image or video tasks where open models are good but hard to run: background removal for one industry, restoration of old photos, or upscaling for print shops.
- 1Pick one task that open image models do well but that is hard to set up, such as upscaling.
- 2Release a free, open-source version with a clear README that points to your paid app.
- 3Sell the hosted version with credits and an API priced per unit of output.
- 4Keep the newest and best modes closed and available only in the paid product.
What could go wrong for Clarity AI?
The risks are the mirror image of the causes above.
What is not verified
The headline US$5,000 of monthly recurring revenue rests on one founder post on X, which this page could read only as eBiz Facts quoted it on 1 July 2024, and the post itself gives no date beyond "two months after launching". No payment processor, filing or second interview with figures was found, and no later revenue figure exists in the sources. A reader could look for a payment-verified profile, for example on TrustMRR (the site takes payments through Stripe), or for a newer interview with the founder.
Founder known by first name and handle only
The site names the founder only as "Philip", his email address uses "philipp", and he is @philz1337x on X and GitHub, with a newer X handle @philipp1337x in the site's code. No source gives his surname or the legal company that runs the service, although the terms name the laws of Singapore.
Free code competes with the paid app
The open-source model is free under AGPL-3.0 and has 30.7 million runs on Replicate, where anyone can pay Replicate to use it without paying Clarity AI. The founder keeps newer modes closed, but Replicate and fal list many other upscalers, and the site's own benchmark pages compare about 50 of them.
Dependence on GPU and model providers
Every job runs on rented GPUs at Modal, fal or Replicate. A price change at these providers changes the cost of each image, and the open models underneath, such as Stable Diffusion and Flux, are controlled by other companies' licenses.
Images of real people
Upscalers that add detail can change faces, and the README says the Flux mode supports LoRAs "with a style or a face". The terms forbid impersonating real people without consent and non-consensual intimate imagery, and require users to be 18 or older; no source says how the service enforces these rules.
Sources
The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.
- D eBiz Facts, 1 July 2024, by Fardeen Khan: His AI App: $5000/Month 2 Months After LaunchRead in full in the site's RSS feed (page 2), downloaded by the pipeline on 27 October 2026; the robots.txt group for all agents allows /feed/ and posts. It quotes the founder's post on X, which was not opened. Source of the headline figure and the founder's background.
- C Clarity AI home page and web app code, read 27 October 2026The page loads by script, so its texts were read from the downloaded JavaScript files of the site: the FAQ, the support answers and the founder link. The robots.txt allows every path.
- C Clarity AI pricing page, read from the page code on 27 October 2026Read the plan objects, the API prices and the usage claim in the downloaded script of the pricing page.
- C Clarity AI terms of service, last updated 20 July 2026Read in full from the page code. Source for the age rule, the content rules and the governing law.
- C Clarity AI privacy policy, last updated 20 July 2026Read in full from the page code. Source for the rule against training on uploads.
- C Clarity AI subprocessors, published 24 September 2026Read in full from the page code. Source for the GPU, model, hosting, payment, email and affiliate providers.
- B GitHub: philz1337x/clarity-upscaler, read 27 October 2026Read the repository page and README: stars, forks, license, release date, update log and the links to the paid app. The robots.txt group for ClaudeBot does not block this path.
- C GitHub profile: philz1337x, read 27 October 2026Read the profile name and bio. The profile shows no full name.
- B Replicate: philz1337x/clarity-upscaler, read 27 October 2026Read in full: run count, cost per run, hardware and the link to the app. The robots.txt allows every path.
- C Clarity AI home page, Wayback Machine copy of 19 May 2024Read the title of the first archived copy with content; the capture list shows only redirects before that date.
- C Loot Drop database export, 1,749 failed startupsLocal export. AI-written summaries, not independently verified. Source for the failed predecessors.