imbrgr
imbrgr.website — images, served hot. Community image and short-video hosting.
Built by ChloReform Studios (Irie Pharm · Victor Birkle). Visual identity: BRANDING.md.
Disclaimer: imbrgr is not affiliated with, endorsed by, or connected to Imgur. This project does not use Imgur’s name, logos, trademarks, or visual identity.
How it works
Plain-language mechanics for operators and contributors. End users see friendly labels in the app, not this section.
Cook count
- Cook count on a post is the number of unique viewers (
Post.viewCount), not page reloads.
- Each viewer is keyed by
voterKey: signed-in users use a stable user-based key; guests use the imbrgr_vid cookie (a:… prefix).
- The post owner (and anonymous owners via matching
voterKey on media) do not increment cook count when they view their own post.
- First visit from a key creates a
PostView row and increments cook count; later visits from the same key update hitCount but do not increment cook count again.
- View recording is rate-limited per post and voter (see
recordPostView in src/lib/post-views.ts): excess requests within the window are ignored without error.
Cheese pull and kitchen heat (spice)
spiceScore is kitchen-themed engagement heat on a post (separate from the unique cook counter).- On each counted view event, spice may increase by
spiceDeltaForView: +1 for a guest’s first unique cook, +1 on repeat visits from the same key, +0 on a signed-in user’s first unique cook (repeat signed-in visits still add +1).
- The UI maps spice to cheese pull / kitchen heat labels and meters (
src/lib/cheese-spice.ts); levels are derived from score bands, not separate counters.
- Spice updates run in the same pipeline as views and refresh
hotScore on the post.
Hot ranking
-
The /hot feed sorts public, non–admin-hidden posts by hotScore (desc), then createdAt.
-
hotScore is recomputed when views/spice change:
engagement = viewCount × 3 + spiceScore × 2 + 1
ageHours = max(0.25, hours since createdAt)
hotScore = engagement / (ageHours + 2)^1.35
(see src/lib/hot-score.ts).
-
Gallery “Most viral” uses a separate Reddit-style hotScore on votes (src/lib/ranking.ts), not the Hot page formula.
Visibility
| Level | Meaning | Typical surfacing |
|---|
| PUBLIC | Listed for everyone | Home/Hot feeds (if not hidden), search, tags, public profiles |
| UNLISTED | Link-only | Direct /p / /i URLs; not promoted in public feeds/search |
| PRIVATE | Owner-only | Hidden from others; file route returns 404 to non-owners |
Admin-hidden (hiddenByAdmin) | Moderation hold | Treated like hidden from the public; owners may still see their post depending on route |
Uploads and studio creations default to public on the gallery (Home / Hot). Signed-in users can choose unlisted or private before or after upload; their default post visibility in settings applies when they do not pick one (migration 20261007200000_default_post_visibility_public sets existing UNLISTED defaults to PUBLIC; users who chose PRIVATE are unchanged). Guests are always public (server-enforced); private and unlisted controls are hidden until they sign in.
Thumbnails
- WebP sm/md thumbs are generated with sharp (
src/lib/thumbnails.ts) and stored via putObject (local disk or Vercel Blob when STORAGE_DRIVER=blob).
- Feeds call
warmFeedThumbnails after the response (after() from next/server) to generate missing thumbs for a capped number of cards—best-effort, deduped per media id.
- Privacy: thumbs are served only through
/api/media/file/* with the same access checks as originals; going private or admin-hidden purges thumb blobs and clears thumb columns; deletes remove storageKey and both thumb keys.
- Cache: private/hidden media uses
Cache-Control: private, no-store; other gated gallery media avoids long immutable public cache (see src/lib/serve-media-file.ts).
Studio library, locked original, and folders
- Click-to-save: In the studio, click the preview image for Save, Save to folder…, Download, or Revert to original (keyboard-friendly menu; Esc closes).
- Click menu (always): Save, Save to folder…, and Download work on any version (original, edits, generated, uploaded) regardless of the keep-original toggle.
- Keep original (toggle): Controls automatic copies in the user’s My images library cache only (default on). On: auto-save keeps the locked first version plus new edits (
LibrarySave rows with label __studio_auto__). Off: auto-save tracks only the latest version in that family (no duplicate original in the auto cache). Manual saves from the image menu always work for whichever version is on screen. Preference: User.studioKeepOriginal or guest localStorage (imbrgr_studio_keep_original). Sync runs after AI generate, studio import, and versioned edits (src/lib/library-auto-save.ts).
- Locked original (editing): The first file in each studio chain stays an immutable row (
Media.rootMediaId); manual and AI edits always create new versions; Revert to original uses the version strip. This is separate from the auto-library toggle above.
- Versions strip: Thumbnails for all versions in the family; the original shows a lock badge.
- Library saves:
POST /api/library/save records the current media version in LibrarySave (signed-in userId or guest voterKey). Visibility on save respects the studio visibility control (AI gens default public unless the user chose otherwise).
- Folders:
LibraryFolder per owner; create/rename/delete via /api/library/folders. Saves can target a folder by id or new folder name from the studio menu.
- My images:
/library and the My images tab on your own profile list folders and saves (GET /api/library). Guest library rows are claimed on sign-in with posts/media ().
Remix, collections, guests, reports
- Remix creates a new post linked to
remixedFromPostId from a public source post; studio can open with ?remixFrom=.
- Collections group posts with their own visibility; collection pages filter member posts by access rules.
- Guests upload with
voterKey and optional delete tokens; on sign-in, POST /api/auth/claim-guest reassigns guest posts/media to the new user when the cookie matches.
- Reports (
POST /api/reports) accept targets POST, MEDIA, COLLECTION, USER with resolved canonical ids; admins triage in the console.
AI pipeline (studio)
- Prompt enhance (optional checkbox): cache lookup → Ollama → Groq → Gemini for complex prompts (
src/lib/ai/prompt-enhance.ts); skipped for very short prompts.
- Image generation: Pollinations Flux (
gen.pollinations.ai); optional POLLINATIONS_API_KEY; mock path when AI_MOCK=true (E2E).
- Blocked prompts:
src/lib/prompt-safety.ts rejects disallowed patterns before generate/edit.
- Daily allowance: per-user count in
AiGenerationUsage (limit AI_DAILY_LIMIT or per-user override); guests use AiAnonymousUsage keyed by IP hash with ANON_AI_DAILY_LIMIT / admin site setting.
- Assist: natural-language image edits (Gemini image route) and quick chips; preview before keep.
- Hourly caps also apply via
UPLOAD_RATE_LIMIT_PER_HOUR and AI_RATE_LIMIT_PER_HOUR.
Admin console
- Super-admins (
SUPERADMIN role + SUPERADMIN_USERNAMES) get /admin: dashboard stats, user/post/media moderation, reports, site settings (e.g. guest AI limit), thumbnail backfill batches, audit log.
Environment variables (names only)
See the table below for descriptions. Never commit values. Common names: DATABASE_URL, DATABASE_URL_UNPOOLED, AUTH_SECRET, STORAGE_DRIVER, BLOB_READ_WRITE_TOKEN, NEXT_PUBLIC_SITE_URL, USER_STORAGE_QUOTA_BYTES, ANON_STORAGE_QUOTA_BYTES, AI_DAILY_LIMIT, ANON_AI_DAILY_LIMIT, POLLINATIONS_API_KEY, OLLAMA_HOST, OLLAMA_MODEL, OLLAMA_API_KEY, GROQ_API_KEY, GEMINI_API_KEY, GEMINI_MODEL, GEMINI_IMAGE_MODEL, UPLOAD_RATE_LIMIT_PER_HOUR, AI_RATE_LIMIT_PER_HOUR, SUPERADMIN_USERNAMES, AI_MOCK, AI_ENHANCE_TIMEOUT_MS.
Features
- Multi-file upload (drag-and-drop, file picker, clipboard paste, URL fetch) — JPG, PNG, GIF, WebP, MP4, WebM with size limits
- Posts with title, description, tags; short
/p/{id} links, direct /i/{id} pages, embed HTML
- Gallery feeds: viral (hot score), newest, top — infinite scroll
- Tag index and per-tag feeds
- Post page: votes, favorites, views, threaded comments, share/embed panel
- Accounts (self-serve sign-up), editable settings (username, display name, avatar/banner with crop, bio, links, public/private favorites, storage meter, account delete)
- Public profiles (
/u/{username}): posts, favorites (if public), comments, stats
- Storage quotas per user (default 1 GB,
USER_STORAGE_QUOTA_BYTES) and smaller anonymous cap; server-enforced on every upload
- In-browser image editor on upload, studio Refine, settings avatar/banner, and your own images: crop, rotate, looks (filters), Adjust sliders (brightness, contrast, saturation, exposure, warmth) with live preview and per-slider reset, Touch up (auto-enhance, vignette, sharpen, soft blur, spot fix, smooth brush), Draw (ember palette + custom color, brush size, eraser), undo/redo, save as new version or replace
- Image studio (
/studio): import (upload / paste / URL), convert (PNG/JPEG/WebP/AVIF + resize), manual editor, AI Flux generation, Gemini natural-language edits, share/embed links
- Media lab (
/studio?tab=media): local browser image/video/GIF/audio import, reorder + trim controls, side-by-side/stacked/grid/overlay/picture-in-picture/collage layouts, WebM export fallback, and pacNcap-inspired eco filter preview
- Mature AI generation policy: mature/unsafe AI outputs require sign-in and are automatically kept off the public gallery (unlisted/private only)
- Prompt enhance: Ollama Cloud → Groq; Gemini only for complex prompts or fallback; cached repeats in Postgres (
AiPromptCache)
- Anonymous uploads are always public on the gallery; sign in to use unlisted or private. Owners can delete posts and report content
- Search across titles and tags
- Ember burger brand system, light/dark themes, responsive layout
Stack
| Layer | Technology |
|---|
| App | Next.js 16 App Router, React 19, TypeScript, Tailwind CSS 4 |
| Auth | Auth.js (next-auth) credentials + JWT sessions |
| Database | Prisma 7 + PostgreSQL (Neon in production) |
| Neon config | @neon/config + root neon.ts (owner runs neon link / neon deploy on their machine) |
| Storage | Local disk (data/uploads) or Vercel Blob (STORAGE_DRIVER=blob) |
Environment variables
Copy .env.example to .env, or after linking Neon run neon env pull (never commit .env).
| Variable | Required | Description |
|---|
DATABASE_URL | Yes | Pooled Postgres URL — app runtime (@prisma/adapter-pg) |
DATABASE_URL_UNPOOLED | Yes for migrations | Direct URL — Prisma migrate / db execute |
NEON_BRANCH | No | Branch name from Neon CLI (informational) |
AUTH_SECRET | Yes | Session signing (openssl rand -base64 32) |
STORAGE_DRIVER | No | local (default) or blob |
BLOB_READ_WRITE_TOKEN | If blob | Vercel Blob token |
NEXT_PUBLIC_SITE_URL | No | Canonical URL for share links |
USER_STORAGE_QUOTA_BYTES | No | Per-account storage cap (default 1 GiB) |
ANON_STORAGE_QUOTA_BYTES | No | Guest upload cap (default 50 MiB) |
AI_DAILY_LIMIT | No | AI generations per user per UTC day (default 20) |
OLLAMA_HOST, OLLAMA_MODEL, OLLAMA_API_KEY | No | First tier for prompt enhance |
GROQ_API_KEY | No | Second tier (Llama 3.3) |
GEMINI_API_KEY, GEMINI_MODEL | No | Gemini for complex prompt enhance + image edits |
GEMINI_IMAGE_MODEL | No | Gemini model for natural-language image edits (studio) |
POLLINATIONS_API_KEY | No | Optional key for gen.pollinations.ai Flux renders |
ANON_AI_DAILY_LIMIT | No | Guest AI generations per UTC day (default 5) |
UPLOAD_RATE_LIMIT_PER_HOUR | No | Per-IP/user upload throttle (default 60) |
|
E2E (Playwright)
Global setup seeds users via tsx scripts/e2e-seed.ts (avoids loading the Prisma ESM client from Playwright’s CJS global-setup).
docker compose up -d # or local Postgres
npm run db:migrate:deploy
export DATABASE_URL=postgresql://imbrgr:imbrgr@localhost:5432/imbrgr?schema=public
npm run build
STORAGE_DRIVER=local AI_MOCK=true npm run test:e2e
npm run test:e2e:artifacts # optional screenshots → /opt/cursor/artifacts
See docs/testing.md for CI, environment gotchas and the guest-cookie rule.
AR Camera Assets
The AR Camera (/tools/ar) uses @mediapipe/tasks-vision for client-side face landmarking. Run:
npm run ar:assets
This copies the MediaPipe WASM runtime into public/mediapipe/wasm/ and downloads face_landmarker.task into public/mediapipe/models/. These binaries are git-ignored.
Production builds must run ar:assets before next build (e.g. npm run ar:assets && npm run build) so that /mediapipe/ static assets are included in the deployment artifact. Otherwise, the AR tool will 404 on local asset routes.
Offline & CI Caveat: Downloading face_landmarker.task requires outbound HTTPS network access to storage.googleapis.com at build/sync time. In offline or air-gapped CI pipelines, the model file must be pre-cached or supplied via pipeline cache to prevent build failures.
For deployment options, CDN hosting alternatives, and Content Security Policy (CSP) requirements, see docs/ar-deployment.md.
Private images & Vercel Blob
On Vercel set STORAGE_DRIVER=blob and BLOB_READ_WRITE_TOKEN (required for uploads and for /api/media/file/* streaming). Without them, production defaults to local disk and files 404.
Image bytes are served at /api/media/file/<storageKey> (e.g. /api/media/file/uuid.jpg?mime=image%2Fjpeg). JSON metadata stays at /api/media/<shortId>.
Production uses a public Blob store (imbrgr13-media). Private files are never listed in feeds/search and the file route returns 404 to non-owners, with Cache-Control: private, no-store. Direct Blob CDN URLs are not exposed in the UI; however, anyone who obtained a raw Blob URL could still fetch the object. For stricter isolation, use a private Blob store or proxy-only delivery and rotate storage keys when visibility changes from public to private.
Do not commit .env, .neon, or connection strings.
Local database (Docker)
docker compose up -d
cp .env.example .env
# DATABASE_URL and DATABASE_URL_UNPOOLED can be the same local URL
npm run db:migrate:deploy
npm run db:seed
Neon (production)
On a machine with the Neon CLI linked to project purple-sea-37945850 (branch production):
neon env pull # writes DATABASE_URL + DATABASE_URL_UNPOOLED (+ NEON_BRANCH) to .env
Apply migrations against the unpooled URL (Prisma config reads DATABASE_URL_UNPOOLED automatically):
export $(grep -v '^#' .env | xargs) # or use dotenv
npm run db:migrate:deploy
For a new migration during development:
npm run db:migrate
Deploy app env to Vercel with at least DATABASE_URL (pooled) and DATABASE_URL_UNPOOLED (for build-time migrate if you run it in CI).
Local setup
npm install
cp .env.example .env
docker compose up -d
npm run db:migrate:deploy
npm run db:seed
npm run brand:assets
npm run dev
Open http://localhost:3000.
Demo login: [email protected] / password123 (after seed).
Scripts
| Command | Description |
|---|
npm run dev | Development server |
npm run build | Production build |
npm run start | Run production server |
npm run lint | ESLint |
npm run typecheck | tsc --noEmit |
npm run test | Vitest (ranking, validation, storage quota, AI usage & prompt failover) |
npm run db:migrate | Create/apply migrations in dev (migrate dev) |
npm run db:migrate:deploy | Apply migrations (migrate deploy) — use on Neon/production |
npm run db:seed | Seed demo data |
npm run brand:assets | Export PNG/ICO brand assets |
Deploy to Vercel
- Import the GitHub repo.
- Set
DATABASE_URL (pooled), DATABASE_URL_UNPOOLED, AUTH_SECRET, and blob vars as needed.
- Run
npm run db:migrate:deploy against production (CI step or once manually).
npm run build (runs prisma generate via postinstall).
License
Proprietary — ChloReform Studios / Irie Pharm unless otherwise noted.
