DEV Community

Yashasvi Yadav
Yashasvi Yadav

Posted on AI-assisted

I Created It for Creators: An Autonomous, $0 Open-Source AI Content Studio πŸš€

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission πŸ’Έ

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

I Created It for Creators: An Autonomous, $0 Open-Source AI Content Studio πŸš€

What I Built

Independent YouTubers, freelancers, and social media managers suffer from two massive bottlenecks: creative burnout from manually repurposing long-form videos across multiple platforms, and SaaS subscription fatigue from paying $50–$150+/month for closed tools like Copy.ai, Jasper, and Midjourney.

I created it for Creators (Creator Studio Local) to solve both problems at once. It is an autonomous, 3-agent AI content production workspace that takes a single video topic, raw creator notes, and a local folder of subject photos, and outputs a complete multi-platform publishing package in seconds:

  1. πŸ“± Agent 01 β€” Social Media Captions Agent: Converts raw script notes into platform-native posts for Instagram, X (Twitter), and LinkedIn, complete with scroll-stopping hooks, clean formatting, and targeted hashtags[cite: 1].

  2. 🎨 Agent 02 β€” Visual Selection & Thumbnail Strategy Agent: Scans the creator's local asset directory of facial expressions/poses, selects the exact image filename with the highest psychological Click-Through Rate (CTR) potential, crafts a punchy 2–5 word text overlay, and writes a detailed background generation prompt optimized for FLUX.1 / Stable Diffusion XL in strict JSON format[cite: 1, 2].

  3. ▢️ Agent 03 β€” YouTube SEO & Copywriting Agent: Generates 3 high-CTR video titles (<65 characters), a search-optimized YouTube description with automatic chapter timestamp placeholders, and 10 high-intent SEO tags[cite: 1].

πŸ’° Side Hustle & Monetization Angle

Instead of forcing solo creators into another $49/month cloud subscription, Creator Studio is designed as a $29–$49 one-time purchase self-hosted desktop/web app (or an ultra-low-cost BYOK hosted workspace)[cite: 1]. Freelance social media managers and solo YouTubers get a private AI production team that costs $0/month to run locally, immediately replacing multiple recurring SaaS bills[cite: 1].


Demo

πŸŽ₯ Watch the Full Demo on YouTube:


Code

πŸš€ I Created It for Creators β€” Open-Source AI Creator Studio

An autonomous, 3-agent AI content production studio powered locally by Gemma 4 & Ollama (with seamless OpenRouter cloud support).


πŸ’‘ Why "I Created It for Creators"

Independent YouTubers, streamers, and solo creators face two massive bottlenecks:

  1. Creative Burnout: Turning a single long-form video into tailored Instagram, X, and LinkedIn posts, designing high-CTR thumbnails, and writing YouTube SEO metadata takes hours of manual work.
  2. SaaS Subscription Fatigue: Paying for Copy.ai, Jasper, and thumbnail tools easily costs $50–$150+/month, while requiring creators to upload unreleased scripts and personal face photos to third-party servers.

I created it for Creators to solve both problems at once. By orchestrating Gemma 4 locally via Ollama and FastAPI, this studio gives every creator their own private, zero-marginal-cost AI production team.


🧠 3-Agent Architecture

When a creator enters a single Video Topic, Raw Notes…


How I Built It

I designed the system around a lightweight, framework-free Python multi-agent orchestrator paired with a modern FastAPI web dashboard[cite: 1, 2]:

  • Open-Weight Model Engine: Powered primarily by Google Gemma 4 (gemma4) running locally via Ollama (CreatorStudio.py & ollama_main.py)[cite: 1, 2].
  • Dual-Engine Flexibility (Local + Cloud): To ensure accessibility on any hardware, I also built a drop-in OpenRouter engine (creatorStudio_openRouterVersion.py & main.py) controlled via .env variables (OPENROUTER_MODEL="google/gemma-4-26b-a4b-it:free" or google/gemma-3-27b-it)[cite: 2].
  • Sequential 3-Agent Orchestration: Built in Python 3 (requests, json, pathlib, pydantic, python-dotenv) without heavy, brittle agent wrappers[cite: 1, 2]. Each agent uses tailored system personas, sampling temperatures (0.7 for creative social copy, 0.4 with strict JSON schema enforcement for thumbnail selection, and 0.6 for YouTube SEO), and robust JSON parsing (parse_llm_json) that handles markdown code-fence stripping[cite: 2].
  • Frontend Workspace: A responsive FastAPI + Jinja2 web application (templates/index.html) styled with Tailwind CSS, custom glassmorphism, Marked.js, and DOMPurify to safely render Markdown copy and structured thumbnail strategy cards with one-click clipboard actions[cite: 2].

Why Does Open Innovation Matter?

Building this on open-weight models (Gemma 4) and open local inference (Ollama) made three critical things possible that closed, proprietary APIs cannot offer[cite: 1]:

  1. Zero Marginal Cost for High-Volume Creators: Content ideation requires iteration. With closed APIs, regenerating hooks, titles, and thumbnail concepts dozens of times a week racks up API bills or hits subscription tiers. Running Gemma 4 locally via Ollama reduces the marginal cost of daily content orchestration to $0.00[cite: 1].
  2. 100% Data & Biometric Privacy: Creators work with sensitive intellectual propertyβ€”unreleased video scripts, embargoed sponsor integrations, and folders of their own personal face photos (creator_assets/)[cite: 1, 2]. Local open-source inference guarantees that personal likenesses and unreleased scripts never leave the creator's machine[cite: 1].
  3. Model Customization & Zero Vendor Lock-In: Creators aren't locked into a single provider's pricing or censorship changes[cite: 1]. They can run gemma4 locally offline, hot-swap open-weight models via .env, or fine-tune a local model on their own past viral videos and brand voice[cite: 1, 2].

My First User

I tested Creator Studio on my own YouTube workflowβ€”using it to generate the exact titles, description, and social promotion posts for the demo video embedded above! I also shared it with a fellow student content creator who highlighted how Agent 2's JSON thumbnail breakdown saved them 30 minutes of staring at a blank canvas before opening Photoshop.


Top comments (0)