sula.ai

Gallery

About

A finished, self-hosted AI content platform. Customers sign up, buy credits on a plan or as a one-off top-up, and generate video and images from 23 models. It has never launched, so there is no revenue and no users. What you are buying is the build, not a book of business.Why it is worth more than the model wrappers you have seenCalling a video API is the easy part and everyone has done it. The months go into everything around it, and that part is finished:Credits ledger with pooled balances, charge and refund, and per-block pricing where the first block bills at the model rate and later blocks lower.A boot-time margin guard. Every one of the 23 models carries its real upstream cost next to what it charges. If an edit would ever sell credits below cost, the server refuses to start. You cannot lose money on a typo.A pricing invariant, also enforced at boot: pay-as-you-go credits are always dearer per credit than the dearest plan, so top-ups can never cannibalise subscriptions.Stripe proven, not merely wired. £5 checkout paid, signature-verified webhook, exactly 200 credits granted, and the same event replayed granted nothing. Stripe retries in production and a webhook that double-credits is real money lost.Multi-tenant, with account isolation proven from outside by an automated test: one account cannot read, change or spend another's rows.A written security review covering injection, path traversal, SSRF, uploads, sessions, escaping and headers, with the two known weaknesses stated rather than hidden.39 automated checks that pass from a clean install on an empty database. 18 of them cover the payment path.Also includedAuto-clipper (long video to vertical captioned clips, running on local ffmpeg and free-tier Whisper at zero provider cost), timeline editor, face swap, upscaling, caption removal, voiceover, YouTube publishing, ten CapCut-style caption presets, a scheduling calendar with reminders, an admin dashboard, a public API with key management, an MCP server for AI agents, and the full marketing site with terms, privacy and acceptable-use pages.TechPython 3.9+ with no web framework: the server and the entire paid path run on the standard library, so there is no framework to migrate and no dependency tree to audit. SQLite. Plain HTML, CSS and JavaScript, no build step, no bundler, no npm install to break in six months. Heavier features import their packages lazily, so requirements.txt is tiered and you install only what you want. Runs with python3 server.py.What it costs to operateOnly fal.ai, per generation, billed when a customer actually generates. Everything else is free tier: Cloudflare Workers AI for scripts and captions, local ffmpeg, edge-tts for voice. The auto-clipper costs nothing per run. No fixed platform fees, so an idle month costs nothing.What you needYour own fal.ai account for the models and your own Stripe account to take money. SETUP.md lists every variable and what it costs, and there is a guided setup script for each.What is included in the saleFull source as a clean single-commit repository (405 files), complete documentation (setup, security review, model licensing, runbook, deployment), the model catalogue with real upstream costs, brand and logo files, the live marketing site, and the 39-check test suite. Not included: my API keys and provider accounts, which you replace with your own, my social channels, and no domain.Why I am sellingI built it, finished it and enjoyed the engineering, but I have no interest in doing the customer acquisition it now needs. I would rather sell it working than let it sit unlaunched.Known weaknesses, since you will find them anywayTwo, both in SECURITY.md. The Content Security Policy allows inline script because the UI uses inline handlers, so CSP is not a second line of defence behind the output escaping. And password hashing is PBKDF2 at 200,000 iterations against OWASP's current 600,000; the count is not stored per hash, so raising it needs a rehash-on-login migration rather than a one-line edit.