Open Studio

Local-first AI content studio for scripts, thumbnails, titles, captions, packages and creator workflows.

Open Studio turns a raw video idea into a publishable content package: script, CTR/SEO title options, thumbnail images, captions/descriptions, assets, critique and exports.

Built for creators who want a real production tool, not a pretty mockup.

---

Preview

More real screenshots

Thumbnails	Titles and captions

Pipeline	Settings

---

What Open Studio Does

Open Studio is a content operations workspace for video creators. The product goal is simple:

Start with a brief. Leave with a complete publishing package.

It currently focuses on:

• Scripts for YouTube, Shorts, tutorials, reviews and creator videos.
• Real thumbnail image generation with batch quantity control.
• 10 CTR/SEO title candidates with a top-3 ranking.
• Captions/descriptions using a high-SEO structure with keywords, hashtags, links and social blocks.
• Content packages that bind script, title, thumbnail, captions, critique and export metadata.
• Assets and exports stored locally.
• Local-first execution with browser settings, BYOK providers and local CLI agents.
• Project/run infrastructure inspired by OpenDesign-style daemon workflows.

Open Studio is not only a UI. The app has provider adapters, local storage, daemon-compatible routes, agent CLI detection, project files, run events, media tasks, research hooks and generation APIs.

---

Why It Exists

Most creator tools are split across too many places:

• one tab for scripts;
• another tool for thumbnails;
• another page for captions;
• another spreadsheet for title ideas;
• another folder for assets.

Open Studio tries to collapse that into one local-first workflow where title, thumbnail, script and caption are treated as one delivery package.

That matters because a video does not win with one isolated output. The title and thumbnail need to match. The script needs to support the promise. The caption needs to reinforce search. The package needs to be exportable.

---

Quick Start

Requirements

• Node.js 20+ recommended.
• npm.
• A provider key for paid providers, or a local CLI / local LLM if you want local-first text generation.
• Optional: Pollinations for quick image testing without an API key.

Install

git clone https://github.com/vivieches/open-studio.git
cd open-studio
npm install

Configure environment

cp .env.example .env.local

You can start with almost nothing configured and use the Settings screen later. Useful environment variables:

# Demo mode: UI + mock/demo outputs
NEXT_PUBLIC_DEMO_MODE=false

# MiniMax
MINIMAX_API_KEY=
MINIMAX_BASE_URL=https://api.minimax.io

# OpenAI
OPENAI_API_KEY=
OPENAI_BASE_URL=https://api.openai.com/v1

# Gemini
GEMINI_API_KEY=
GEMINI_BASE_URL=https://generativelanguage.googleapis.com

# Pollinations: optional for basic image tests
POLLINATIONS_API_KEY=
POLLINATIONS_BASE_URL=https://gen.pollinations.ai

# Local storage
LOCAL_STORAGE_DRIVER=json

Run the app

npm run dev

Open:

http://localhost:3000

Optional local daemon

The standalone daemon is available for local runs, health, logs and run lifecycle experiments:

npm run daemon
npm run daemon:health

Default daemon URL:

http://127.0.0.1:7456

---

Tutorial: From Idea To Package

1. Choose language

Use the globe button or Settings page.

Supported UI and generation languages:

• Portuguese Brazil: pt-BR
• Spanish Spain: es-ES
• English US: en-US

The selected language controls new scripts, titles, captions and visible text instructions.

For image providers, Open Studio writes the technical image prompt in English for better model performance, but enforces this rule:

Any visible text inside the generated image must be in the selected language.

2. Configure execution

Go to Settings.

You can choose:

• CLI local: use detected local agent CLIs for text/agent work.
• BYOK: use your own API keys in the browser/local settings.
• Text defaults: provider and model for script/title/caption/package text.
• Image defaults: provider and model for thumbnail/image generation.

The app can detect local CLIs such as Claude Code, Codex CLI, Gemini CLI, OpenCode and others. Detection scans PATH and local toolchain folders.

Important: local text CLIs do not automatically generate pixels. Image generation still needs an image provider or the media tool contract.

3. Generate a script

Open Roteiros / Scripts.

Provide:

• topic or brief;
• audience;
• tone;
• duration;
• language.

The result is saved as a local asset and can be reused by content/package routes.

4. Generate thumbnails

Open Miniaturas / Thumbnails.

You can configure:

• video topic;
• title;
• impact text;
• audience;
• visual style;
• mood;
• background;
• color preference;
• quantity of thumbnails.

The new batch route generates real images:

POST /api/generate/thumbnails

For quick tests, select Pollinations / flux as image provider. It can work without a key for basic public image generation, though it may be slower or rate-limited.

5. Generate titles and captions

Open Títulos e legendas / Titles and captions.

Current modules:

• generate 10 CTR/SEO title candidates;
• rank the top 3;
• generate SEO captions/descriptions with hashtags, keywords, links and social blocks;
• use a Lucas-style caption pattern by default or a custom pattern.

6. Run the package pipeline

Open Pipeline.

The package flow is:

briefing -> script -> titles -> thumbnail -> captions -> critique -> export

The package is the core object. It keeps the title, thumbnail and script connected instead of treating them as separate random outputs.

7. Review assets and exports

Generated outputs appear in Arquivos / Assets and can be exported as structured package files.

---

Features

Creator workflow

Area	Status	Notes
Dashboard	Working	Real local assets and recent activity.
Scripts	Working	Provider-backed text generation with CLI/BYOK fallback logic.
Thumbnails	Working	Real image generation, batch quantity and generated image cache.
Titles	Working	CTR/SEO candidates with top-3 ranking.
Captions	Working	SEO description/caption format with keywords, hashtags and social links.
Pipeline	Working	Full content package generation flow.
Assets	Working	Local asset records for scripts, thumbnails, prompts and exports.
Exports	Working	Markdown/JSON style package exports and image references.
Critique	Working	Package critique for title/thumbnail/script alignment.
Research	Partial	Tavily-style research route exists; richer outlier analysis is still evolving.
Video/audio	Hidden	Internal media hooks exist, but UI scope is currently text + image.

Local-first platform

Capability	Status	Notes
Browser/local settings	Working	API keys and defaults can be stored locally.
Provider catalog	Working	Text/image providers are listed from manifests.
Agent CLI registry	Working	Detects installed local CLIs and custom binaries.
Standalone daemon	Partial	Health, status, logs, runs and SSE events exist.
Projects/runs	Working foundation	Project files and run lifecycle exist, still being expanded.
Media tasks	Working foundation	Generate/wait/cancel routes exist for media tasks.
Skills registry	Working foundation	Skill scan endpoint exists, deeper UX pending.
Memory/brand kit	Working foundation	Routes exist; deeper prompt composer integration ongoing.
Import/export workspace	Partial	Project import exists; full professional ZIP/PDF/PPTX export pending.

---

Providers And Models

Open Studio uses a BYOK model: bring your own keys when a provider requires one.

Text providers

• Anthropic
• OpenAI
• OpenAI-compatible providers
• Azure OpenAI
• Gemini
• OpenRouter
• Groq
• Together AI
• DeepSeek
• MiniMax
• Ollama
• LM Studio
• vLLM
• Local OpenAI-compatible servers
• Stub placeholder for deterministic local tests

Image providers

• MiniMax
• OpenAI image models
• Azure OpenAI image deployments
• Together / FLUX-style models
• fal.ai
• Replicate
• Pollinations
• Local/OpenAI-compatible image endpoints where supported

Local CLI agents

The agent registry is inspired by OpenDesign-style local power workflows and detects:

• Claude Code
• Codex CLI
• Gemini CLI
• OpenCode
• Devin for Terminal
• Hermes
• Kimi CLI
• Cursor Agent
• Qwen Code
• Qoder CLI
• GitHub Copilot CLI
• Pi
• Kiro CLI
• Kilo
• Mistral Vibe CLI
• DeepSeek TUI

The resolver scans PATH plus known user toolchain directories such as npm global shims, local bin folders, Bun, Volta, Cargo and other common Windows/macOS/Linux install locations.

---

Image Generation: What Actually Happens

Text generation and image generation are separate surfaces.

Gemini CLI / Codex CLI / Claude Code -> text, reasoning, scripts, package logic
Image provider adapters              -> actual pixels / thumbnails
Media tool contract                  -> agent-callable image/video/audio bridge

If you select Gemini CLI and generate a script, that can work through the CLI.

If you generate a thumbnail image, Open Studio needs an image provider. That can be MiniMax, OpenAI, fal.ai, Replicate, Together, Pollinations or another supported image adapter.

For the fastest no-key test:

Settings -> Generation defaults -> Image -> Pollinations -> flux

Then generate from Miniaturas.

The generated files are cached under:

public/generated/thumbnails/

---

Languages

Open Studio supports:

Locale	UI	Generation
pt-BR	Yes	Yes
es-ES	Yes	Yes
en-US	Yes	Yes

Rules:

• UI follows selected language.
• New text generations follow selected language.
• Captions/descriptions follow selected language.
• Image prompts are technical English.
• Any visible text inside images must be in the selected language.

---

Local-First Architecture

flowchart LR
  User["Creator"] --> UI["Next.js dashboard"]
  UI --> Settings["Browser/local settings"]
  UI --> API["Next API compatibility layer"]
  API --> Providers["Provider adapters"]
  API --> Agents["Local CLI agents"]
  API --> Store[".open-studio local store"]
  API --> Projects["Projects, runs and files"]
  API --> Media["Media tasks"]
  Providers --> Text["Text generation"]
  Providers --> Images["Image generation"]
  Agents --> Text
  Media --> Images
  Store --> Assets["Assets and exports"]
  Projects --> Package["ContentPackage"]

Loading

The application currently runs mainly through Next API routes while also exposing a standalone daemon foundation.

Storage layout

.open-studio/
  assets/
  daemon/
    logs/
  exports/
  generated files/
  projects/
  runs/
  settings.json

public/generated/thumbnails/
  cached generated images

---

API Surface

Health, agents and providers

Method	Route	Purpose
GET	/api/health	App health.
GET	/api/agents	Detect local CLI agents.
POST	/api/agents/test	Smoke-test a selected CLI.
GET	/api/providers	Provider catalog.
GET	/api/providers/:providerId/models	Provider model discovery/cache.
GET	/api/providers/:providerId/test	Provider connection test.
POST	/api/proxy/:provider/stream	Normalized provider stream proxy.

Generation

Method	Route	Purpose
POST	/api/generate/text	Text generation.
POST	/api/generate/image	Single image generation.
POST	/api/generate/thumbnails	Batch thumbnail image generation.
POST	/api/generate/titles	CTR/SEO title candidates.
POST	/api/generate/captions	SEO caption/description generation.
POST	/api/generate/package	Full content package.

Workspace

Method	Route	Purpose
GET/POST	/api/projects	List/create projects.
GET/PATCH/DELETE	/api/projects/:projectId	Project lifecycle.
GET/POST	/api/projects/:projectId/files	Project file tree.
GET/DELETE	/api/projects/:projectId/files/:path	File read/delete.
POST	/api/projects/import	Safe project import.

Assets, exports and tools

Method	Route	Purpose
GET/POST	/api/assets	Local asset list/create.
GET/DELETE	/api/assets/:id	Asset detail/delete.
GET/POST	/api/exports	Export list/create.
GET	/api/exports/:id/download	Download export.
POST	/api/media/generate	Media task generation.
GET	/api/media/tasks	Media task list.
POST	/api/media/tasks/:id/wait	Wait for media task.
POST	/api/media/tasks/:id/cancel	Cancel media task.
POST	/api/research/search	Research source search.
GET/PATCH/POST/DELETE	/api/memory	Local memory.
GET/PATCH/POST/DELETE	/api/brand-kit	Brand kit and style memory.

---

CLI Usage

The package exposes two local commands:

node scripts/open-studio.mjs media generate --surface image --prompt "dark studio thumbnail, bold text" --provider pollinations --model flux --aspect 16:9
node scripts/open-studio.mjs media wait <taskId>
node scripts/open-studio.mjs media cancel <taskId>
node scripts/open-studio.mjs research search --query "AI tools outlier titles" --max-sources 6

Daemon:

node scripts/open-studio-daemon.mjs serve --port 7456
node scripts/open-studio-daemon.mjs health --url http://127.0.0.1:7456

Environment:

OS_DAEMON_URL=http://127.0.0.1:3000
OPEN_STUDIO_DAEMON_URL=http://127.0.0.1:7456
OPEN_STUDIO_DATA_DIR=.open-studio
OS_PROJECT_ID=

---

Development

npm run dev          # Start Next dev server
npm run daemon       # Start local daemon
npm run build        # Production build
npm run lint         # ESLint
npm run typecheck    # TypeScript only
npm run test         # Vitest unit tests
npm run test:e2e     # Playwright E2E tests

Project structure

app/
  (dashboard)/        Dashboard routes
  api/                Next API routes
  components/         App UI components

lib/
  daemon/             Agent registry, media tasks, projects, runs, critique
  providers/          Provider manifests, adapters, model catalog, runtime
  prompts/            Script, thumbnail and pipeline prompt builders
  security/           SSRF, origin and input guards
  storage/            Local JSON storage
  validation/         Zod schemas

scripts/
  open-studio.mjs
  open-studio-daemon.mjs

docs/
  specs/
  screenshots/
  providers.md
  roadmap.md
  troubleshooting.md

---

Roadmap

Open Studio is moving fast. The current priority is to make the creator workflow genuinely useful end to end.

Near-term

• Improve thumbnail provider UX and provider-specific quality presets.
• Add stronger title + thumbnail cohesion scoring.
• Make research/outlier analysis more visible inside title generation.
• Add richer package export formats: .zip, .pdf, .pptx, self-contained HTML.
• Add better project browser and file workspace.
• Add a 10-second Remotion product teaser for the repository.

Platform work

• Finish standalone daemon separation from Next API.
• Expand CLI agent streaming parsers.
• Improve run resume/cancel/retry.
• Add richer provider model discovery and cache invalidation.
• Add live artifacts for editable packages.
• Add deeper memory and brand kit injection into all generation surfaces.

Future / hidden media scope

• Video generation adapters.
• Audio/music generation adapters.
• Storyboard/sketch editor for thumbnails and video packages.
• MCP server and connectors for external creative workflows.

---

Current Limitations

• Local CLI agents are strongest for text/agent work. Image generation still needs image/media providers.
• Pollinations is useful for quick image tests, but quality and latency vary.
• Some provider integrations need real keys and quota to be fully verified.
• Standalone daemon exists, but the app still uses Next API routes as the main compatibility layer.
• Video and audio are intentionally hidden from the active UI while the text/image workflow is hardened.
• The professional export pipeline is still evolving.

---

Security Notes

• .env.local is ignored.
• Browser/local settings are intended for local-first use.
• BYOK keys should only be used on trusted local machines.
• SSRF protection is implemented for provider base URLs.
• Local agent CLI mode can use broad permissions by design. Treat it as a power-user local workflow.

See SECURITY.md for responsible disclosure.

---

Documentation

Document	Description
Setup	Installation and configuration details.
Providers	Provider setup and model notes.
Roadmap	Product roadmap.
Troubleshooting	Common issues and fixes.
Complete SDD	Full system design direction.
Sprint specs	Implementation sprint plan.

---

Contributing

Good contribution areas:

• provider adapters;
• local CLI parsing;
• thumbnail prompt and quality presets;
• title/caption SEO modules;
• tests;
• export formats;
• documentation;
• UI polish that preserves the current product direction.

Start with CONTRIBUTING.md.

---

Co-developed by

• Vitória Ferreira — CEO and Creator Open Studio Project
• Lucas Valen — Co-developer and AI Product Engineer

---

License

MIT