{"id":51305602,"url":"https://github.com/cognis-digital/creatorforge","last_synced_at":"2026-07-01T00:00:40.720Z","repository":{"id":368010284,"uuid":"1283079236","full_name":"cognis-digital/creatorforge","owner":"cognis-digital","description":"The AI content team you own: voice-trained hooks, scripts, captions, thumbnails \u0026 multi-platform packaging. Offline by default, your model when you want it.","archived":false,"fork":false,"pushed_at":"2026-06-28T17:37:37.000Z","size":116,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-06-28T18:05:20.977Z","etag":null,"topics":["ai","content-creation","creator-tools","marketing","mcp","scriptwriting","self-hosted","social-media"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/cognis-digital.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-28T14:21:35.000Z","updated_at":"2026-06-28T17:37:40.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/cognis-digital/creatorforge","commit_stats":null,"previous_names":["cognis-digital/creatorforge"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/cognis-digital/creatorforge","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cognis-digital%2Fcreatorforge","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cognis-digital%2Fcreatorforge/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cognis-digital%2Fcreatorforge/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cognis-digital%2Fcreatorforge/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/cognis-digital","download_url":"https://codeload.github.com/cognis-digital/creatorforge/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cognis-digital%2Fcreatorforge/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34987610,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-30T02:00:05.919Z","response_time":92,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","content-creation","creator-tools","marketing","mcp","scriptwriting","self-hosted","social-media"],"created_at":"2026-07-01T00:00:23.678Z","updated_at":"2026-07-01T00:00:40.700Z","avatar_url":"https://github.com/cognis-digital.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# creatorforge\r\n\r\n**The AI content team you _own_ — not one you rent for $15K/month.**\r\n\r\n---\r\n\r\nBusiness owner? Creator? Ask yourself:\r\n\r\n- Are you paying (or about to pay) **five figures a month** for a \"done-for-you AI content team\"?\r\n- Does that team's \"proprietary system\" run on **someone else's servers**, with your voice, audience, and data living there?\r\n- When the retainer ends, do you **keep anything** — or does the system walk out the door with them?\r\n- Do you actually want to film 2–4 hours a week and have everything else handled — **without renting the engine that does it**?\r\n\r\nIf you nodded, here's the deal: `creatorforge` is that content team, as **open software you own and run yourself**. No retainer. No 50-spots-only. No \"proprietary\" black box. It's free, COCL (Cognis Open Collaboration License), and runs on *your* hardware and *your* model.\r\n\r\n\r\n## Watch the walkthrough\r\n\r\nA full narrated tour — setup, the tool in action, and every demo scenario:\r\n\r\n[![Watch the creatorforge walkthrough](media/walkthrough-thumb.png)](https://github.com/cognis-digital/creatorforge/releases/download/walkthrough-v1/walkthrough.mp4)\r\n\r\n▶ **[Watch the walkthrough (MP4)](https://github.com/cognis-digital/creatorforge/releases/download/walkthrough-v1/walkthrough.mp4)**\r\n\r\n## What you get\r\n\r\nThe same deliverables an agency installs — generated from one command:\r\n\r\n- **Voice profile** — it learns your style from your past posts (no model training, nothing uploaded — just measurable style features).\r\n- **Niche research brief** — content pillars, audience pains, keyword targets.\r\n- **Content ideas** — a steady pipeline of format × angle ideas so you never face a blank page.\r\n- **Hooks** — proven scroll-stopping formulas, written in your voice.\r\n- **Scripts** — full, structured scripts sized to each platform.\r\n- **Captions** — on-screen overlay text + timed SRT subtitles.\r\n- **Thumbnail concepts** — headline / visual / emotion / layout, **rendered as actual SVG mockups**.\r\n- **Multi-platform packaging** — one idea, tailored for YouTube, Shorts, TikTok, Reels, X, and LinkedIn (each platform's limits respected).\r\n- **Posting calendar** — your ideas scheduled at your cadence.\r\n\r\nYou film. creatorforge does the rest.\r\n\r\n## How it's different (the honest version)\r\n\r\n| | The agency offer | **creatorforge** |\r\n|---|---|---|\r\n| Price | ~$15K/month retainer | **free, open source** |\r\n| Where it runs | their cloud | **your hardware** |\r\n| Your voice/data | lives on their system | **never leaves your machine** |\r\n| When it ends | you keep nothing | **you own it forever** |\r\n| The \"system\" | proprietary black box | **readable code you can audit** |\r\n| Results promised | \"1 billion views\" | **a great engine — the views are on you** |\r\n\r\nNo inflated view-count promises here. creatorforge gives you a genuinely strong content *system*; what it can't do is guarantee virality, and anyone who guarantees that is selling something. It runs **fully offline** on a deterministic engine, and gets sharper the moment you point it at a local model (Ollama) or a cloud one — your call.\r\n\r\n## Quick start\r\n\r\n```bash\r\npip install -e .\r\n\r\n# 1. Learn your voice from a folder of your past posts/scripts\r\ncreatorforge profile ./my_posts/ --out voice.json\r\n\r\n# 2. Run the whole team on a topic, across your platforms\r\ncreatorforge pipeline --topic \"why you should own your AI stack\" \\\r\n    --niche \"AI for business\" --platforms youtube,tiktok,x,linkedin \\\r\n    --voice voice.json --start 2026-07-06 --out plan.json\r\n\r\n# Or one piece at a time\r\ncreatorforge hooks    --topic \"cold email that converts\" --voice voice.json\r\ncreatorforge script   --topic \"cold email that converts\" --platform youtube_shorts --voice voice.json\r\ncreatorforge thumbnail --topic \"cold email that converts\" --svg thumb     # writes thumb-1.svg …\r\n\r\n# Use your own local model to sharpen the prose (nothing leaves your machine)\r\ncreatorforge script --topic \"...\" --provider ollama --model llama3\r\n```\r\n\r\nRun `python demo.py` to watch it learn a voice and generate a full multi-platform plan end to end.\r\n\r\n## Platforms\r\n\r\n| Platform | Aspect | Sweet spot | Tailored output |\r\n|----------|--------|-----------|-----------------|\r\n| YouTube | 16:9 | ~10 min | title + description + script |\r\n| YouTube Shorts | 9:16 | ≤60 s | short script + captions |\r\n| TikTok | 9:16 | ~27 s | caption + hashtags + script |\r\n| Reels | 9:16 | ~60 s | caption + hashtags + script |\r\n| X | 16:9 | — | ≤280-char post |\r\n| LinkedIn | 1:1 | ~90 s | long-form post |\r\n\r\n## Local models — every modality, on your hardware\r\n\r\ncreatorforge drives the best **open-source, local, free** model your machine can run, in every modality — and degrades cleanly when you don't have the big one. Ask it what it can do:\r\n\r\n```bash\r\ncreatorforge capabilities\r\n```\r\n\r\n| Modality | What it does | Open model (GPU) | Runs with no GPU |\r\n|----------|-------------|------------------|------------------|\r\n| **Text** | hooks, scripts, ideas in your voice | any Ollama model (auto-picks your best) | ✅ via Ollama / template engine |\r\n| **Transcription** | footage → text → repurposed posts | faster-whisper `large-v3` | ✅ `base`/`small` on CPU |\r\n| **Voice** | voiceover + **voice cloning** | XTTS-v2 (clone) | ✅ Piper (CPU) |\r\n| **Image** | photorealistic thumbnails | FLUX.1-schnell / SDXL-Turbo / SD 1.5 | ✅ real raster PNG (PIL) → SVG |\r\n| **Video** | finished short, captions timed | LTX-Video / CogVideoX (text-to-video) | ✅ assembled MP4 (ffmpeg) / animated GIF |\r\n| **Audio** | voiceover + music, leveled | MusicGen beds | ✅ ffmpeg mix / stdlib WAV |\r\n\r\nThe recommended model in each row is chosen to fit your VRAM (`recommend()` ladders from a 24 GB GPU down to CPU). Photorealistic images, cloned voices, MP4s, and generated b-roll need the model/tool installed — but there's always a **real, local fallback** so nothing hard-fails: a composited raster thumbnail, an animated GIF, a synthesized WAV. Nothing is ever sent to a cloud API.\r\n\r\n```bash\r\n# transcribe footage with local Whisper\r\ncreatorforge transcribe talk.mp4\r\n\r\n# generate a thumbnail (diffusion if you have it, else a real raster PNG)\r\ncreatorforge image --topic \"owning your AI stack\" --voice voice.json --out thumb\r\n\r\n# produce a finished short from a script (MP4 with ffmpeg, else animated video)\r\ncreatorforge video --topic \"owning your AI stack\" --platform youtube_shorts --out short\r\n\r\n# voiceover (Piper, or clone your voice with XTTS) and an audio track\r\ncreatorforge voiceover --from-script script.json --out vo.wav --speaker my_voice.wav\r\ncreatorforge audio --voiceover vo.wav --music bed.mp3 --out track\r\n\r\n# the whole production in one command: plan + thumbnail + video + audio + outbox\r\ncreatorforge produce --topic \"owning your AI stack\" --niche \"AI for business\" \\\r\n    --provider ollama --out ./production/\r\n```\r\n\r\n## Long-form \u0026 cinematic production (5–15 min)\r\n\r\nShort clips are one mode. creatorforge also plans and assembles **long-form** video — documentaries, video essays, dev logs, promos — structured the way the industry actually does it. It fuses four things:\r\n\r\n- a **format** (`creatorforge formats`) — the proven beat order for documentary / video-essay / devlog / promo\r\n- a **cinematic style** (`creatorforge styles`) — pacing, shot vocabulary, color, and music mood drawn from the grammar of prestige docs, kinetic vlogs, trailer-cut blockbusters, slow-burn arthouse, true-crime, and more\r\n- the **algorithm playbook** — what each platform actually rewards (watch time / AVD, completion rate, the first-30s hook, re-hooks every ~40s, chapters) turned into concrete production directives\r\n- your **voice** (+ an optional local model to polish the narration)\r\n\r\n…into a full plan: acts → scenes → timed shots → narration → chapters → a music \u0026 SFX cue sheet → titles and thumbnails, sized to your target runtime.\r\n\r\n```bash\r\ncreatorforge formats        # documentary, video_essay, devlog, promotional\r\ncreatorforge styles         # epic_doc, true_crime, kinetic_vlog, blockbuster, arthouse_slowburn, …\r\ncreatorforge longform --topic \"owning your AI stack\" --format documentary --style epic_doc --minutes 12 --out plan.json\r\ncreatorforge studio   --topic \"owning your AI stack\" --format documentary --minutes 10 --provider ollama --out ./film/\r\n```\r\n\r\nGenerative **music and sound effects** come from local open models (MusicGen / AudioGen) when your GPU can run them, with a synthesized bed as a fallback so a cut always has a track.\r\n\r\n\u003e **On \"Netflix-level\":** creatorforge plans and directs at that structural level — story structure, shot lists, pacing, sound design, retention engineering — and assembles the cut with whatever render models you have. Cinema-grade *footage and audio* come from the heavy open models (FLUX / LTX-Video / MusicGen / XTTS) on a capable GPU, which the engine drives; on a laptop you still get a complete, correctly-structured cut with real generated assets. The structure is studio-grade everywhere; the render fidelity scales with your hardware.\r\n\r\n## Punch above your weight class: real photos, designed\r\n\r\nYou don't beat a giant cloud image model on a laptop by out-rendering it — you **out-source and out-design** it. creatorforge pulls *real, no-watermark* photography and composites it into pro visuals, which reads as more professional than weak CPU diffusion ever will.\r\n\r\n- **Your own library, offline.** Index image folders you already have and search them by keyword — multiple related shots per scene, your images, zero watermarks, zero licensing questions. Optionally caption them with local `llava` for smarter matching.\r\n- **Free, no-key, no-watermark CC stock.** Openverse / Wikimedia Commons, each result carrying its license + attribution so you stay compliant. Watermarked sources (Getty/Shutterstock previews) are never queried.\r\n- **Designed, not dumped.** A sourced hero photo gets cover-cropped, color-graded, given a legibility scrim, and set with a bold headline — a real editor's thumbnail.\r\n\r\n```bash\r\n# build an offline, searchable index of your image folders (your own = no watermark)\r\ncreatorforge assets index ~/Pictures ~/b-roll --out assets.json\r\n\r\n# pull multiple related, key shots for a scene\r\ncreatorforge assets search \"datacenter server racks\" --index assets.json -k 6\r\ncreatorforge assets search \"founder at laptop\" --index assets.json --online   # + free CC stock\r\n\r\n# build a thumbnail composited over a real sourced photo\r\ncreatorforge image --topic \"owning your AI stack\" --assets assets.json --out thumb\r\n# studio/produce auto-source a hero from your library:\r\ncreatorforge studio --topic \"...\" --format documentary --assets assets.json --out ./film/\r\n```\r\n\r\n\u003e Licensing, honestly: your own library is yours. Openverse/Wikimedia results are Creative Commons / public-domain and **may require attribution** — creatorforge surfaces the license and attribution string with every result so you can credit correctly. It never pulls watermarked or paid-preview imagery.\r\n\r\n## Direction \u0026 engagement craft (multi-cam, cool shots, retention)\r\n\r\nLong-form plans don't just have scenes — they have **coverage and engagement built in**:\r\n\r\n- **Multi-camera shot generation.** Every beat gets a shot list across A/B/C/detail/drone/POV cameras with motivated **camera moves** — dolly-in, push-in, crane, gimbal track, **drone reveal**, **dolly-zoom (vertigo)**, orbit, speed-ramp — and the showy \"cool shots\" land where they matter (cold opens, turns, climaxes). The plan tells you exactly which cameras a production needs.\r\n- **Engagement craft, encoded.** Each beat carries a **retention move** drawn from two lineages: great-filmmaker grammar (in-media-res, but/therefore causality, show-don't-tell, escalating stakes, tension-release, match cuts, the Kuleshov effect) and modern retention tactics (front-load the payoff, escalate every segment, reset the hook on a cadence, concrete stakes/numbers, no dead air, tease what's coming, emotional payoff). Structure isn't just correct — it's *sticky*.\r\n\r\nIt's all in the `longform`/`studio` plan: `scenes[].shots` (cams + moves), `scenes[].retention_move`, `multicam`, and `engagement_plan`.\r\n\r\n## Make content for your repos — and grow like the AI companies that made it\r\n\r\nPoint creatorforge at a repository and it writes the content **for** it; point it at an owner and it does the whole catalog.\r\n\r\n```bash\r\n# one repo -\u003e multi-platform content (reads the README, derives what it is)\r\ncreatorforge repo ./codegraph-mcp --format promotional --out plan.json\r\ncreatorforge repo cognis-digital/agentledger --longform --format documentary   # via gh\r\n\r\n# every repo of an owner, batched\r\ncreatorforge repos --owner cognis-digital --format promotional --out ./repo_content/\r\n\r\n# a 30-day launch strategy that mirrors how the big AI companies actually grew\r\ncreatorforge growth ./codegraph-mcp\r\n```\r\n\r\nThe **growth playbook** encodes the plays that repeatedly worked for the AI companies that broke out — a runnable demo over a pitch deck, building in public, a benchmark moment, developer-first distribution, founder-led content, free/open core, naming the category — and turns them into a concrete day-by-day launch calendar for your project. The edge was never the secret; it's executing these consistently, which is exactly what the engine makes cheap.\r\n\r\n## Wire it into your stack (MCP)\r\n\r\ncreatorforge ships an **MCP server**, so Claude, an internal orchestrator, or any MCP-capable agent can drive it directly:\r\n\r\n```bash\r\ncreatorforge serve     # JSON-RPC over stdio\r\n```\r\n\r\nTools: `profile_voice`, `generate_ideas`, `write_hooks`, `write_script`, `thumbnail_concepts`, `package_for_platform`, `run_pipeline`. Hand the packaged posts to your scheduler/poster of choice and the loop is closed.\r\n\r\n## The pipeline\r\n\r\n```mermaid\r\nflowchart LR\r\n    posts[Your past posts] --\u003e voice[VoiceProfile]\r\n    topic[Topic + niche] --\u003e pipe{{run_pipeline}}\r\n    voice --\u003e pipe\r\n    prov[Provider\u003cbr/\u003eTemplate / Ollama] -.optional polish.-\u003e pipe\r\n    pipe --\u003e ideas[ideas]\r\n    pipe --\u003e hooks[hooks]\r\n    pipe --\u003e script[script] --\u003e caps[captions + SRT]\r\n    pipe --\u003e thumbs[thumbnails + SVG]\r\n    pipe --\u003e pkg[per-platform packages]\r\n    ideas --\u003e cal[calendar]\r\n    pkg --\u003e out[(Multi-platform deliverable)]\r\n    caps --\u003e out\r\n    thumbs --\u003e out\r\n    cal --\u003e out\r\n    topic --\u003e lf[[longform.build_longform]] --\u003e studio[(Cinematic plan:\u003cbr/\u003escenes · shots · cues)]\r\n    voice --\u003e lf\r\n    classDef hot stroke:#F4B400,stroke-width:3px;\r\n    class pipe,lf hot;\r\n```\r\n\r\nSee [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for the full walkthrough.\r\n\r\n## Demos\r\n\r\nFive runnable scenarios in [`demos/`](demos/), each for a different audience.\r\nEvery one runs **fully offline** — no network, GPU, or model download — on the\r\ndeterministic engine, so they double as smoke tests. Details in\r\n[`docs/DEMOS.md`](docs/DEMOS.md).\r\n\r\n```bash\r\n# cp1252 consoles (Windows): PYTHONUTF8=1 makes the emoji output render\r\nPYTHONUTF8=1 python demos/run_all.py            # all five, end to end\r\nPYTHONUTF8=1 python demos/02_agency_full_pipeline.py   # or just one\r\n```\r\n\r\n| # | Scenario | Audience | What it shows |\r\n|---|----------|----------|---------------|\r\n| 1 | `01_creator_voice_to_post.py` | Solo creators | Learn a voice from past posts, then write hooks, a script, overlays, and SRT in that voice. |\r\n| 2 | `02_agency_full_pipeline.py` | Business owners | The whole agency bundle from one `run_pipeline` call — brief, ideas, hooks, script, captions, packages, calendar. |\r\n| 3 | `03_devrel_repo_launch.py` | DevRel \u0026 marketing | Point it at a repo (reads the README), get multi-platform content + a 30-day launch calendar. |\r\n| 4 | `04_multiplatform_repurpose.py` | Multi-platform publishers | One idea packaged for all six channels with every limit honored, plus the algorithm playbook. |\r\n| 5 | `05_longform_studio.py` | Long-form showrunners | A 12-minute documentary plan: scenes, multi-cam shot lists, retention moves, chapters, titles. |\r\n\r\n## Testing\r\n\r\n```bash\r\npip install -e \".[dev]\"\r\npytest -q          # 49 tests\r\n```\r\n\r\n## License\r\n\r\nCOCL (Cognis Open Collaboration License). © Cognis Digital. The whole engine is open — read it, fork it, run it on your own models. You own your content team.\r\n\r\n\u003e Status: v0.1 — runnable and tested. Short-form + long-form (5–15 min) production; text, transcription, voice, image, video, music/SFX backends wired with hardware-aware selection and CPU fallbacks; format/style/algorithm intelligence baked in. Roadmap: ComfyUI image/video backend, true text-to-video assembly, A/B title scoring, scheduled auto-publish.\r\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcognis-digital%2Fcreatorforge","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fcognis-digital%2Fcreatorforge","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcognis-digital%2Fcreatorforge/lists"}