Open source AI video platform with 5 tools in one: Clip Generator, AI Shorts (UGC videos with AI actors), YouTube Studio, VoiceOver (AI-narrated shorts) and Game Profiles (per-game viral scoring).
Two people on camera? OpenShorts stacks them instead of shrinking the wide shot, puts the captions on the seam where they cover nobody, and switches back to a face-tracked crop when the cut goes to one person. The AI picks the layout per video; nothing to configure.
Two ways to run it, same software either way:
| Self-hosted (this repo) | Hosted on openshorts.app | |
|---|---|---|
| Price | Free forever, MIT | Free plan, paid from $12/mo |
| Speed | 5 to 8 min per 8-min video on CPU | About 50s on our NVIDIA GPU |
| API keys | Bring your own Gemini, ElevenLabs, fal.ai | Gemini included, nothing to set up |
| Watermark / limits | None, ever | Watermark and 20 min/mo on the free plan, neither on paid |
| Setup | Docker, 8GB+ RAM, model downloads | Sign in and paste a link |
| MCP / API for agents | Same /mcp endpoint, but only while your machine is on |
Always-on endpoint at mcp.openshorts.app, API keys in one click |
| Your data | Your server | Ours |
Self-hosting is genuinely free and always will be. It costs you a machine, your own API keys and the time to keep it running. The hosted plans exist to cover that hardware and those keys, not to unlock features.
demo-openshorts.mp4
Click the image above to watch the full walkthrough.
Turn your long-form videos — podcasts, webinars, livestreams, vlogs, interviews — into viral-ready 9:16 shorts for TikTok, Instagram Reels, and YouTube Shorts.
Generate marketing videos with AI actors for any product or business. No camera, no studio, no influencer budget. Just describe your product or paste a URL.
- Two cost modes: Low Cost (
$0.65/video) and Premium ($2/video) - Works for any business: SaaS, restaurants, e-commerce, coaching, local businesses
- AI-generated actors with lip-sync, voiceover, b-roll, and TikTok-style subtitles
- Choose from a shared avatar gallery or upload your own photo
- Publish directly to TikTok, Instagram, and YouTube
Complete free AI YouTube toolkit: thumbnails, titles, descriptions, and direct publishing.
- AI thumbnail generator with face overlay
- 10 viral title suggestions with refinement chat
- Auto-generated descriptions with chapter timestamps
- One-click publish to YouTube
All generated videos and avatars are saved to a public gallery with SEO pages for each video.
- Public gallery page with hover-to-play (
/gallery) - Individual SEO video pages with og:video meta tags (
/video/{id}) - JSON-LD structured data for search engines
- Avatar gallery with prompt history
Turn any clip into a narrated short: AI caption suggestions → TTS voiceover → ducked mix → burned captions, all in one guided workflow.
- Caption suggestions with live preview: the AI proposes caption sets (Gaming, Dramatic, GenZ, Creepypasta…), each one previewed in a real Remotion player beside the options — words highlight as the clip plays, exactly like the final burn
- Two TTS providers: local Qwen3-TTS VoiceDesign (free, runs on your GPU, 12-dimension voice personas) or ElevenLabs — selectable per job
- Voice & Style Presets manager: 10 seeded voice personas and 10 caption style prompts, fully editable. Voice presets are tagged with caption styles (a "Gaming Hype" voice carries the
gamingtag); picking a style reorders matching voices to the top of the dropdown and auto-selects the first one - Emoji-aware captions: the LLM sprinkles emojis into captions, timed to hold 1–2 s on screen; display text keeps emojis while the spoken audio stays clean. Emoji words burn through Remotion's AnimatedEmoji set (Google/Android animation), and every later re-style of the clip (subtitle editor, hook, recut) re-burns through the same path so the animation survives chaining
- Non-English TTS done right: Italian (and any language) gets slang expansion (
nn→ non,1v1→ uno contro uno), number-to-words conversion, and article-elision splitting so the voice says "l' ho fatto" correctly — while burned captions keep the natural written form - Auto ducking: background audio is sidechain-compressed under the narration and the mix is extended if the voiceover runs long
- Session persistence: switch tabs or refresh mid-generation and the workflow keeps its state; projects reopen from History with captions restored
- Steam auto-import: type the game name, pick the match from live Steam search — the profile pulls the official description, genres and tags automatically (or fill everything by hand)
- AI analysis on demand: one click runs an LLM analysis over the Steam data to produce game type, gameplay characteristics and the key moments that matter for virality (clutches, fails, jumpscares…), suggested weights are copied onto the profile with one click
- Custom prompt / notes: your own instructions get prompt priority in every selection pass — "prefer clutch saves" or "find the funny fails" steer clip selection explicitly
- Where it applies: cheap-signal seeding, transcript scoring, the deep full-VOD scan and the semantic vision pass all receive the profile context, so a horror game is judged on scares and a fighter on combos
- Profiles are stored, editable, importable/exportable
Open on any generated clip for a full caption studio rather than a font picker:
- 10 style presets (TikTok, Reels, Shorts Pop, Gold Glow, Neon, Cyber, Karaoke, Beast, Boxed…) plus save your own preset/default
- Live Remotion preview with karaoke word-highlighting that plays in sync with the video — what you see is exactly what gets burned
- Fine styling sliders: vertical position, size, words per block, block duration, spacing, letter spacing, colors, outline, opacity, per-word highlight color, animations (pop, glow, karaoke, none)
- Word-level editing: click any word in the text to fix a mishear or drop in an emoji at the playhead — edits keep the original word timings (LCS-matched), so nothing drifts
- Timing strip with playhead: drag words and trim handles directly on a zoomable timeline (3s/5s/10s/20s/Full), pan by dragging empty track, and follow-playhead mode keeps the active word centered as the red playhead moves (edit text blurred above for privacy)
- Burn anywhere: renders through the same Remotion/ASS pipeline used by auto-captions, so browser preview and final clip match; emoji renders are handled with a proper color-emoji font
- All generated projects grouped per source video — clip jobs and VoiceOver jobs together
- One click reopens a project: clips, edit state, and (for VoiceOver) the caption suggestions are restored as if you never left
- Per-job provider selector: run the analysis with Gemini or OpenAI (any OpenAI-compatible endpoint) straight from the dashboard — keys and models come from Settings and ride on the job, no server env needed
- Independent deep-analysis selector: the deep full-VOD scan can use a different provider/model than the main job (e.g. cheap flash for scoring, a stronger model for the deep watch), with a "same as job" default
- Viral Moment Detection: the LLM analyzes transcripts and scene boundaries to detect 3-15 high-potential moments
- Runs fully local if you want: point
LLM_BASE_URLat Ollama, LM Studio, vLLM or any OpenAI-compatible server and the moment picker runs on your own model, no Google key needed (see Run without a Google key) - Multimodal candidate detection (all toggleable per job): the transcript is always analyzed, plus optional cheap signals that catch moments nobody is narrating — PySceneDetect scene cuts, audio spikes/screams, visual motion — each seeding extra candidate windows so a silent jumpscare followed by a scream still becomes a clip
- Vision pass: each shortlisted window gets 6 frames (3 uniform + 3 peak-biased around screams/loudness) analyzed against the Game Profile, adjusting scores on visual evidence
- Deep full-VOD scan: for long videos, the AI watches the whole VOD (native low-res video proxy for Gemini, 12-16 sampled frames for OpenAI-compatible models) and its best moments bypass candidate scoring entirely — they are prepended straight into the final selection with their own titles, and the remaining slots are filled from the scored candidates. Game-profile aware when a profile is selected. Because deep already saw the whole footage, enabling deep auto-disables the per-window vision pass (explicit header override can force both)
- VOD title & description for Twitch: after clip selection, a dedicated LLM pass writes a Twitch-ready VOD title plus a description with timestamped highlights (~10 moments ranked by viral potential, absolute seconds converted to
MM:SS), surfaced in the dashboard with copy buttons — independent of whether deep ran - Robust LLM output handling: tolerant JSON parsing with repair, per-clip validation with duration swap-repair, snap-to-word-boundary cuts, and IoU-based overlap dedup so two windows finding the same moment never render twice
- Parallel transcript enhancement: an optional LLM polish pass fixes Whisper mishears (homophones, game jargon) in concurrent chunks while preserving word timings; emojis are added conservatively and timed 1-2 s so their animation plays
- Smart 9:16 Cropping: AI reframing per scene — TRACK mode (MediaPipe + YOLOv8 face tracking), GENERAL mode (blurred background), SPLIT mode (two speakers stacked, captions on the seam) and SCREENCAST mode (screen over presenter); the layout is picked per video by Gemini or forced from the dashboard
- Auto Subtitles: faster-whisper with word-level timestamps, styled and burned into clips
- AI Voice Dubbing: ElevenLabs integration for 30+ languages with voice cloning
- Hook Text Overlays: AI-generated attention-grabbing text overlays
- AI Video Effects: Gemini-generated FFmpeg filters for professional effects
flowchart TD
A[Source video] --> B[faster-whisper transcription\nword-level timestamps]
B --> C{Silent video?}
C -- yes --> V[Gemini watches the full video\nvisual clip picking]
C -- no --> E[Optional LLM polish pass\nfix mishears + emojis]
E --> D{Deep full-VOD scan?\nprovider/model selectable independently}
D -- on --> F[Gemini: 12-16 frames or native\nlow-res proxy over whole VOD\ntop moments as priority candidates\n+ game profile context]
D -- off --> G
F --> G[Cheap signals\nscene cuts · audio spikes · visual motion]
G --> H[Transcript windows\n+ cheap-seeded + deep candidates]
H --> I[Pass 1 — score every window\nbatched LLM calls]
I --> J{Vision pass on?\nskipped automatically when deep is on\n+ Game Profile}
J -- yes --> K[6 frames per window\nsemantic score adjustment]
J -- no --> L
K --> L[Shortlist best windows]
L --> M[Pass 2 — detail extraction\nsurrounding transcript + word timestamps\n+ scene-cut hints per window]
M --> N[Validate · snap to words\ndedup overlapping picks]
N --> O[Render clips\nreframe · captions · hooks]
N --> P[VOD title + description\nTwitch-ready with MM:SS timestamps]
V --> O
- Analyze: Scrape website URL + web research, or generate from manual description
- Script: AI writes viral scripts (hook - problem - solution - CTA format)
- Actor: Generate AI actors with Flux 2 Pro or select from shared gallery
- Voice: ElevenLabs TTS voiceover (English/Spanish, male/female)
- Video: Talking head generation (Hailuo 2.3 Fast img2video + VEED Lipsync)
- B-roll: AI-generated visuals with Ken Burns effect
- Composite: FFmpeg final assembly with subtitles and hook overlays
- Publish: Direct posting to TikTok, Instagram Reels, YouTube Shorts via Upload-Post
- AI-powered title generation with 10 viral options
- Interactive refinement chat for titles
- AI thumbnail generation with custom face + background
- Auto descriptions with chapter timestamps from Whisper transcript
- Direct YouTube publishing via Upload-Post
- One-click posting to TikTok, Instagram Reels, and YouTube Shorts simultaneously
- Schedule uploads for any date and time — plan your content calendar and let OpenShorts publish automatically
- Multi-platform distribution — publish to all your social networks at once from a single interface
- Upload-Post integration with async uploads
- S3 cloud backup (private bucket for clips, public bucket for gallery/avatars)
- SEO gallery pages served by FastAPI with JSON-LD structured data
- Shared avatar gallery across all users
- Async job queue with configurable concurrency
- Content creators — Turn long videos into shorts automatically, publish to all platforms at once
- Marketing agencies — Generate UGC videos for clients at scale, no actors or studios needed
- SaaS founders — Create product demos and marketing shorts from just a URL
- E-commerce brands — Product videos with AI actors for TikTok Shop, Instagram, YouTube
- Local businesses — Restaurants, gyms, real estate, coaching — affordable video marketing
- Developers — Self-host, customize the pipeline, integrate via API
Videos generated with OpenShorts AI Shorts — no camera, no studio, no actors:
 |
 |
 |
| Biohacking for Investors · LOW COST | Secret Weapon for Devs · LOW COST | El Secreto de los Agentes de IA · PREMIUM |
Browse all videos at openshorts.app/gallery
| Feature | OpenShorts | Opus Clip | CapCut | Vizard | Klap | Descript |
|---|---|---|---|---|---|---|
| Price | Free self-hosted from $12/mo hosted |
$15-29/mo | $8/mo | $15-20/mo | $23-63/mo | $24-65/mo |
| Self-hosted | Yes | No | No | No | No | No |
| Open source | Yes | No | No | No | No | No |
| Watermark | Never self-hosted free plan only when hosted |
Free tier | Some | Free tier | Free tier | Free tier |
| Upload limits | None self-hosted by plan when hosted |
10-30GB | Credit-based | 60min-10hr | 10-100 vids/mo | 60min-40hr |
| AI clip detection | Yes | Yes | Yes | Yes | Yes | Yes |
| Smart 9:16 reframing | Yes | Yes | Yes | Yes | Yes | No |
| Auto subtitles | Yes | Yes | Yes | Yes | Yes | Yes |
| Voice dubbing (30+ langs) | Yes | No | Pro only | No | Pro only | Business only |
| AI UGC actors | Yes | No | No | No | No | No |
| AI video effects | Yes | No | Yes | No | No | No |
| Hook text overlays | Yes | No | No | No | No | No |
| YouTube Studio (titles, thumbnails) | Yes | No | No | No | No | No |
| Social auto-publishing | Yes | Pro only | TikTok only | Paid only | Paid only | No |
| Schedule uploads | Yes | Pro only | No | Paid only | Paid only | No |
| Data privacy | Your server | Their cloud | Their cloud | Their cloud | Their cloud | Their cloud |
| Works with a local LLM (Ollama) | Yes | No | No | No | No | No |
Self-hosting OpenShorts is free. You provide the machine and you only pay for the AI APIs you use, and most have generous free tiers:
| Service | Free Tier | Paid Cost | Used For |
|---|---|---|---|
| Google Gemini | Free trial with generous limits | < $0.01 per 10-min video | Viral moment detection, script generation, web research |
| Local LLM (Ollama, LM Studio, vLLM...) | Free, your hardware | $0 | Viral moment detection instead of Gemini (LLM_BASE_URL) |
| fal.ai | Pay-per-use | ~$0.50-1.50 per AI Short | Actor generation, talking head video, lip-sync |
| ElevenLabs | Free tier available | Pay-per-use | Voiceover, voice dubbing |
| Upload-Post | 10 free uploads/month to all networks (no credit card) | Pay-per-use | Auto-publishing to TikTok, Instagram, YouTube |
| AWS S3 | Optional | ~$0.023/GB | Cloud backup for clips and gallery |
Bottom line: You can clip videos for practically free with Gemini, and publish 10 videos/month to all social networks at zero cost with Upload-Post.
Don't want to run any of that? openshorts.app is the same software on our hardware: our NVIDIA GPU clips an 8-minute video in about 50 seconds instead of the 5 to 8 minutes it takes on a typical CPU, the Gemini key is included, and auto-publishing is already wired up. Free plan is 20 minutes a month with a watermark and no credit card; paid plans start at $12/mo for 100 minutes without watermark.
- Docker & Docker Compose
- Google Gemini API Key (Free — get it here) — required for all AI features
- fal.ai API Key (Pay-per-use) — required for AI Shorts (actor generation, video, lip-sync)
- ElevenLabs API Key (Free tier) — required for voiceover/dubbing
- Upload-Post API Key (free tier) — required for direct social posting
The VoiceOver workflow can narrate clips with a local Qwen3-TTS model instead of ElevenLabs.
It's installed automatically in Docker GPU builds (--build-arg GPU=1, the default in
docker-compose.yml). The model (~4 GB) downloads lazily on first use into a persistent
hf-cache Docker volume — deliberately not baked into the image, which would bloat builds.
For a local (non-Docker) dev setup:
pip install qwen-tts # plus SoX on your PATH (apt install sox / brew install sox)The first voiceover generation downloads the Qwen3-TTS VoiceDesign model (~4 GB) into the HuggingFace cache. Without the package, the VoiceOver page simply offers ElevenLabs only.
Voices are designed, not cloned: each voice preset is a 12-dimension persona (gender, pitch, speed, volume, timbre…) sent as a design instruction to the model. Create and edit personas in the Voice/Style Presets page — every voice is tagged with the caption styles it fits, and the VoiceOver page brings matching voices to the top when you pick a style. For non-English narration the pipeline preprocesses the speak-text per language (slang expansion, numbers to words, article elisions) so pronunciation is correct, while the burned captions keep the original written form.
git clone https://github.com/Marsic1/openshorts.git
cd openshortscp .env.example .env
# Edit .env with your AWS keys for S3 backupdocker compose up --buildNavigate to http://localhost:5175
- Go to Settings and enter your API keys (Gemini, fal.ai, ElevenLabs, Upload-Post)
- Clip Generator: Upload a long-form video to generate viral shorts
- AI Shorts: Describe your product or paste a URL to generate UGC marketing videos
- YouTube Studio: Generate thumbnails, titles, and descriptions for YouTube
- UGC Gallery: Browse all generated videos and avatars
The default image is CPU-only. With an NVIDIA card (any card with NVENC, e.g. RTX 4060) an 8-minute video clips in about a minute instead of 5 to 8. Nothing is passed through in the VM sense — the container just gets access to the host GPU.
Host: install the NVIDIA driver (nvidia-smi must work) and the NVIDIA Container Toolkit:
sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker
docker run --rm --gpus all nvidia/cuda:12.4.0-base-ubuntu22.04 nvidia-smi # sanity checkOn Windows use Docker Desktop with the WSL2 backend and the Windows NVIDIA driver; no driver inside WSL.
Compose: create docker-compose.override.yml next to docker-compose.yml (picked up automatically). GPU: "1" adds cuBLAS/cuDNN and onnxruntime-gpu to the image (~2 GB); video is required for NVENC.
services:
backend:
build:
context: .
args:
GPU: "1"
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu, video].env:
WHISPER_MODEL=large-v3-turbo
WHISPER_DEVICE=cuda
WHISPER_COMPUTE=float16
FFMPEG_ENCODER=auto # probes h264_nvenc at startup, falls back to x264
TRANSCRIBE_BACKEND=parakeet # optional: ~2x faster than whisper, 25 European languages, auto-falls back to whisper
ASR_GPU_CONCURRENCY=1
Verify:
docker compose up --build -d
docker exec openshorts-backend nvidia-smi -L
docker exec openshorts-backend ffmpeg -hide_banner -f lavfi -i testsrc=size=256x256:rate=1 -frames:v 1 -c:v h264_nvenc -f null -The backend log on the first job reports the chosen encoder and transcription device. A CUDA error in whisper (e.g. VRAM exhausted) retries once on CPU automatically. 8 GB of VRAM is enough for large-v3-turbo fp16 plus the detection models.
The only cloud call in the clip pipeline is the moment picker: it sends the transcript (never the video) to Gemini. Point it at any OpenAI-compatible server instead and the whole pipeline stays on your box:
# .env
LLM_BASE_URL=http://host.docker.internal:11434/v1 # Ollama on the host
LLM_MODEL=qwen2.5:14b # any chat model that follows instructions
# LLM_API_KEY=... # only if your server checks one (vLLM --api-key, OpenRouter)Works with Ollama, LM Studio, vLLM, llama.cpp server, LocalAI and OpenRouter. The dashboard stops asking for a Gemini key when this is set. Two things to know:
- Context length. A scoring call carries three transcript windows
(~2-3k tokens) and the detail call up to ten (~5k on a long podcast).
Ollama defaults to a 4096-token context and truncates silently, so run it
with
OLLAMA_CONTEXT_LENGTH=16384(or setnum_ctxin a Modelfile); raiseLLM_SCORE_BATCHabove 3 only if your context allows it. 7-8B models return valid JSON reliably, 3B ones do not. - What still needs Gemini. Anything that has to look at frames: the
automatic layout picker (
AUTO_LAYOUT), the on-screen content detector and silent videos (no speech to clip by). Without a Gemini key those fall back to the plain face-tracking crop, and a silent video fails with a message that says so. Add a key alongsideLLM_BASE_URLand you get both.
- Ingest — Local video upload (or self-hosted URL ingest via yt-dlp)
- Transcribe — faster-whisper with word-level timestamps, optional LLM polish pass
- Candidates — transcript windows + cheap signals (scene cuts, audio events, visual motion) + optional deep full-VOD scan
- Score & detail — two LLM passes (provider selectable: Gemini or OpenAI): batch scoring, then detailed extraction on the shortlist with validation, word-snapping and overlap dedup
- VOD metadata — Twitch-ready title + description with MM:SS highlight timestamps
- Extract — FFmpeg precise clip cutting
- Reframe — AI vertical cropping with subject tracking
- Effects — Subtitles, hooks, AI video effects
- Publish — S3 backup + Upload-Post social distribution
- Analyze — Website scraping + Gemini web research (or manual description)
- Script — Gemini generates viral scripts with segments
- Actor — Flux 2 Pro portrait generation (or gallery/upload)
- Voice — ElevenLabs TTS voiceover
- Video — Hailuo 2.3 Fast img2video + VEED Lipsync (Low Cost) or Kling Avatar v2 (Premium)
- B-roll — Flux 2 Pro image generation + Ken Burns effect
- Composite — FFmpeg assembly with ASS subtitles and hook overlays
- Gallery — Upload to public S3 with metadata for SEO pages
- Publish — Upload-Post to TikTok, Instagram, YouTube
- Captions — the selected AI (Gemini or OpenAI) proposes a caption set in the chosen style; each suggestion is previewed live in Remotion
- Edit — per-word editing in the preview, style sliders (size, density, spacing, colors) with real-time re-render
- Speak — per-caption TTS (Qwen3-TTS VoiceDesign locally or ElevenLabs); Italian and other languages get slang/number/elision preprocessing on the speak side only
- Time — word-level captions are matched 1:1 to the spoken audio; emojis get guaranteed 1-2 s windows so their animation plays
- Mix — narration ducked over the original audio (sidechain compression, timeline extension if the voice runs long)
- Burn — captions burned via Remotion (animated emoji included), result saved and openable from History; later re-styles from the subtitle editor re-burn through the same path
flowchart LR
C[Clip] --> CS[Caption suggestions\nstyled, with emojis]
CS --> PV[Live Remotion preview\n+ word editor]
PV --> TTS[TTS per caption\nQwen local / ElevenLabs]
TTS --> W[Word timing sync\nemoji 1-2s windows]
W --> MX[Ducked mix\nvoiceover over original]
MX --> BR[Burn captions\nfinal narrated short]
You don't need the dashboard. The whole pipeline is callable by AI agents and scripts.
OpenShorts ships a built-in MCP server, so Claude, ChatGPT, Cursor or any MCP client can clip and publish videos for you:
claude.ai and ChatGPT: paste https://mcp.openshorts.app/mcp as a custom connector (Settings → Connectors) and approve the access on openshorts.app. The server does OAuth 2.1 with dynamic client registration, so there is no key to copy; the connection shows up under Account → API keys, where revoking it disconnects the app.
# Claude Code / Cursor / n8n (hosted): create an API key in your account page
claude mcp add --transport http openshorts https://mcp.openshorts.app/mcp \
--header "Authorization: Bearer osk_..."
# Self-hosted (no key needed, BYOK rules apply):
claude mcp add --transport http openshorts http://localhost:8000/mcpTools: process_video (URL or upload_id; captions: false when the source already has subtitles, auto_hook: false to skip the hook line, burned by default like the dashboard), create_upload (hand the agent a local file: PUT the bytes, then process), get_job_status, list_clips, get_quota, add_subtitles, recut_clip, publish_clip. A prompt like "clip this podcast and schedule the best 3 to TikTok" is now a one-liner in your agent of choice.
Hosted accounts can mint osk_... API keys (account page). A key authenticates as you everywhere — same plan, same minutes, same job ownership:
curl -X POST https://api.openshorts.app/api/process \
-H "Authorization: Bearer osk_..." -H "Content-Type: application/json" \
-d '{"url": "https://youtube.com/watch?v=...", "acknowledged": true,
"webhook_url": "https://your-server.com/hooks/openshorts"}'Interactive docs at /docs (OpenAPI) on any instance.
Pass webhook_url (and optionally webhook_secret) to POST /api/process and you get exactly one POST when the job reaches a terminal state — no polling loops in your n8n / Zapier / cron pipelines:
{"event": "job.completed", "job_id": "…",
"clips": [{"index": 0, "title": "…", "video_url": "…", "download_url": "…"}]}With a secret, the body is signed: X-OpenShorts-Signature: sha256=<hmac-sha256(body)>.
The same API from the terminal, zero dependencies (cli/):
pip install openshorts # or: uvx openshorts
export OPENSHORTS_API_KEY=osk_... # hosted
# export OPENSHORTS_API_URL=http://localhost:8000 # self-hosted, no key
openshorts process "https://youtube.com/watch?v=..." --wait
openshorts clips <job_id>
openshorts publish <job_id> 0 --platforms tiktok,youtubeskills/openshorts/SKILL.md follows the open
Agent Skills standard, so it works in any
skill-capable agent:
# Claude Code (and most agents): copy the folder into the skills directory
cp -r skills/openshorts ~/.claude/skills/
# Hermes Agent: install straight from this repo
hermes skills install mutonby/openshorts/skills/openshorts
# OpenClaw: from ClawHub
openclaw skills install @mutonby/openshortsAn importable workflow (video URL in, published-ready clips out, no polling)
lives in examples/n8n/.
| Layer | Technology |
|---|---|
| Backend | Python 3.11, FastAPI, google-genai, faster-whisper, ultralytics (YOLOv8), mediapipe, opencv-python, yt-dlp, FFmpeg, httpx |
| Frontend | React 18, Vite 4, Tailwind CSS 3.4 |
| AI APIs | Google Gemini, fal.ai (Flux, Hailuo, VEED, Kling), ElevenLabs |
| Infrastructure | Docker + Docker Compose, AWS S3 |
| Publishing | Upload-Post API (TikTok, Instagram, YouTube) |
Server-side (.env):
| Variable | Description |
|---|---|
AWS_ACCESS_KEY_ID |
AWS access key for S3 |
AWS_SECRET_ACCESS_KEY |
AWS secret key |
AWS_REGION |
AWS region (default: us-east-1) |
AWS_S3_BUCKET |
Private bucket for clip backup |
AWS_S3_PUBLIC_BUCKET |
Public bucket for gallery/avatars |
MAX_CONCURRENT_JOBS |
Concurrent processing limit (default: 5) |
LLM_BASE_URL |
OpenAI-compatible server for the moment picker (Ollama, vLLM, LM Studio...). Set it and the Gemini key becomes optional |
LLM_MODEL |
Model name on that server (default llama3.1:8b) |
LLM_API_KEY |
Bearer token for that server, if it checks one |
LLM_SCORE_BATCH |
Transcript windows per scoring call (default 3 local, 8 Gemini) |
GEMINI_MODEL / OPENAI_MODEL |
Fallback models when the dashboard doesn't send one (the Settings/provider selectors win) |
DEEP_AI_PROVIDER |
Fallback deep-analysis provider when the job doesn't specify one (same = follow the job's provider; the dashboard's deep selector always wins) |
Client-side (encrypted in localStorage):
| Key | Description |
|---|---|
GEMINI_API_KEY |
Google Gemini — required unless LLM_BASE_URL is set (then only for layout picking and silent videos) |
FAL_KEY |
fal.ai — required for AI Shorts |
ELEVENLABS_API_KEY |
ElevenLabs — required for voiceover/dubbing |
UPLOAD_POST_API_KEY |
Upload-Post — required, for social posting |
- Non-Root Execution: Containers run as dedicated
appuser - Concurrency Control: Semaphore-based job queue (
MAX_CONCURRENT_JOBS) - Auto-Cleanup: Automatic purging of old jobs (1h retention)
- Encrypted Keys: API keys encrypted client-side, never stored server-side
- Upload Validation: Image uploads validated for format and minimum size
- File Limits: 2GB upload limit protection
- Register: app.upload-post.com/login
- Create Profile: Go to Manage Users
- Connect Accounts: Link TikTok, Instagram, and/or YouTube
- Get API Key: Navigate to API Keys
- Use in OpenShorts: Paste the key in Settings
Contributions are welcome! Whether it's adding new AI models, improving the lip-sync pipeline, or building new features — feel free to open a PR.
MIT License for the core application — OpenShorts is yours to use, modify, and scale.
Exception: the cloud/ directory (billing, managed keys, and the hosted-service infrastructure behind the optional BILLING_ENABLED flag) is source-available under the OpenShorts Commercial License. You can read it, modify it, and self-host it for personal or internal use, but you can't offer it to third parties as a paid/hosted service. Self-hosting the core app never requires this directory.