In “Dear Higgsfield, I cloned you in 1 hour,” Albert Olgaard asks a coding agent to build an AI generation studio around Higgsfield's public template and API. The result is an interesting prototype: after a repair pass, he shows one video and one image being generated in the local app. It is not evidence that an entire production platform, or Higgsfield's underlying models, were recreated.
Albert reports the coding session took about three hours, despite the video's one-hour title, and puts its API-equivalent agent cost at roughly $100. Two generation paths are shown working. Stripe billing, every listed model, account isolation, security, production deployment and a paying business are not demonstrated end to end. The valuable takeaway is a narrower build-and-test workflow, not a verified one-to-one clone.
Watch Albert Build and Test the Studio
Credit and scope: The observations below come from Albert's video and the supplied transcript. We separately checked Higgsfield's template repository and API documentation on 21 October 2026. We did not run or audit Albert's generated application or receive its source code.
What Higgsfield Actually Released
The relevant release is the public higgsfield-ai/app-templates repository. It includes a Studio starter distributed as a shadcn registry, with UI patterns and typed generation plumbing that can shorten the path to a similar interface. It does not give someone Higgsfield's model weights, customer base, billing operation, moderation system or production service. Generation in this workflow still calls the Higgsfield API using account credentials and credits.
There is also a licensing distinction. The repository is publicly viewable, but we did not find a license file in its root when reviewing it. Public visibility alone does not grant general reuse rights. Check the current repository license or obtain permission before copying template code into a commercial product; build an original brand and visual identity rather than using Higgsfield's name or assets. The attached prompt below asks for close visual similarity, so treat that instruction as a reference to evaluate, not an automatic legal clearance.
What the Video Demonstrates, and What It Does Not
| In the video | What is established | What remains open |
|---|---|---|
| Build session | Albert enters a long prompt into Claude and later reports a three-hour session with about $100 in API-equivalent coding usage (02:51). | That is his reported session cost, not a repeatable price or the total cost of operating the app. |
| Local studio | A branded local app called VIO shows video, image, effects, community, pricing, billing and admin views (04:29). | Seeing a screen is not proof that every control, model, permission and checkout path works. |
| Real generation | The first video requests error. Albert adds API credit, asks Claude to inspect logs, retries, and later shows a generated video and a generated image (05:10; 06:41). | The video does not demonstrate all models, failure recovery, concurrent requests, durable storage or production traffic. |
| Pricing and payments | The UI displays credits and plans. Albert says Stripe would work after keys are connected; he then notices an image job appears to cost only one app credit (06:58). | Stripe checkout, webhook handling and profitable per-request pricing are not verified in the demo. |
The distinction is not pedantic. Higgsfield's own quickstart describes a queued request, an idempotency key and a later status check. A reliable wrapper must handle those states, duplicate submissions and credits, not merely display a “Generating” tile. Its billing guidance also says cost varies by model and parameters, failed or moderated jobs are refunded, and generated outputs may disappear after at least seven days unless you save them yourself.
A $5.4 Billion Valuation Is Not a Clone Business Model
Higgsfield announced a $400 million financing at a $5.4 billion valuation in August 2026. That is the company's reported financing valuation; it says nothing about the value of a new API wrapper. Albert's example of 400 users spending $50 each, with $25 left per user, is arithmetic for a hypothetical $10,000 monthly gross contribution. It is not observed demand or net profit. API charges, payment fees, failed runs, refunds, storage, support, acquisition, taxes and engineering time can all change the result.
The stronger idea is the one Albert reaches near the end: pick a reachable buyer and a specific visual job, such as showing a contractor a proposed kitchen renovation. Then test whether the buyer will pay for a reviewed deliverable before building a generic creative platform. Ask for permission to use input photos, mark generated previews as illustrative, and keep a human responsible for claims about real construction outcomes.
Full Attached Build Prompt
The following is the complete build brief supplied with this request, preserved in a copyable field. It is long because it specifies architecture, models, credit accounting, Stripe, storage, security and verification in phases. The video description's Google Docs URL was supplied only as a truncated link, so we cannot verify that this attachment is word-for-word the document Albert shared. The brief is source material, not a tested recipe; check every dependency, license, API path and cost before acting on it. Never put real API keys into a prompt or client-side code.
Download the complete prompt as plain text. It begins with editable brand and pricing placeholders, then covers: ground-truth sources; the Studio architecture; model catalog; Auth.js and MongoDB; credits and Stripe; UI routes; hardening; environment variables; nine build phases; and a definition of done.
Read the full attached prompt on this page
Higgsfield-style image, video and audio studio
Preserved as attached; review source rights and live API details before use.
Before You Use the Prompt
- Check reuse rights. Confirm the current terms for the public template, and replace third-party branding, copy and media with assets you own or may use.
- Verify the API surface. Start with one documented image model and one documented video model. Follow live API documentation for endpoint names, credentials, polling, retries and idempotency. Do not assume that a model shown in a UI is available to your account.
- Price from actual estimates. Use Higgsfield's request estimate for the exact model and parameters, then test debit/refund behavior and payment webhooks. Albert's one-credit image price is a warning, not a validated margin.
- Protect people and outputs. Keep API secrets server-side, isolate users and generations, review input-photo rights, and copy completed media to storage you control if it must persist beyond the provider's retention window.
- Prove the whole buyer journey. A useful acceptance test is: an authorized user signs in, funds a test account, generates an asset, sees the correct charge, downloads it after a refresh, and can get an accurate refund on a failed job. Do this in test mode before taking real payments.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | The one-to-one clone claim |
| 00:58 | Reseller pricing and the hypothetical 400-user model |
| 02:01 | The long build prompt and Claude setup |
| 02:51 | Three-hour session and reported API-equivalent cost |
| 03:15 | Environment variables and API credentials |
| 04:29 | The VIO interface and first video attempt |
| 05:10 | Failed requests, credits and the repair loop |
| 06:00 | Tabs, pricing UI and admin screens |
| 06:41 | A video output appears |
| 06:58 | Image test and the pricing warning |
| 07:29 | Why Albert recommends a distinct niche |
| 08:10 | Provider pricing discussion |
| 09:03 | Kitchen-remodeling example and shared code |
Sources and Verification Limits
Video: Albert Olgaard's walkthrough and its supplied transcript. Template: Higgsfield's public app-templates repository. API: quickstart and billing and retention. Valuation: Higgsfield's financing announcement. The prompt above is a user-provided attachment, not an authenticated copy of the truncated Google Doc. This article does not verify the generated code, revenue projections, all model access or the legal status of commercial template reuse.