Tanque Studio

A native macOS companion app for Draw Things, providing a focused AI image generation workspace, Draw Things project browsing, StoryFlow/Story Studio authoring, batch render queues, and LLM-assisted prompt enhancement.

Screenshots

Dashboard	Focus Room

DT Project Browser	Render Queue

Features

Dashboard + Focus Rooms

The app opens on a Dashboard home screen rather than dropping straight into the generation workspace:

Continue card — resume the last in-progress session with one click
System card — live Draw Things connection, LLM Assist status, storage used
Quick Start — preset prompts/models that jump straight into a fresh session
Recent Generations strip and Projects / Labs mini-lists
A persistent top nav (Project Browser · Labs · Settings) reachable from every screen, not just from the Dashboard

Opening a session (or clicking any image) enters a Focus Room: a full-bleed canvas + filmstrip below it, with all generation controls tucked into a single collapsible drawer on the right (Prompt, Assist, Model, Parameters, LoRAs, img2img & Moodboard, Actions) instead of stacked side panels. The feature set is the same one detailed below — same GenerateViewModel, same gRPC transport — just delivered through the drawer instead of dedicated left/right panels.

Generate

The generation feature set — prompt/config, canvas, gallery, LLM Assist, actions — described panel-by-panel below. In the running app this now lives inside a Focus Room's accordion drawer (see above) rather than a dedicated four-panel screen, but every capability listed here still applies.

Left panel

Prompt and negative prompt fields
Model picker with full filename display
Config import — load presets from Draw Things custom_configs.json
Canvas size presets (S / M / L) that preserve the current aspect ratio
Aspect ratio tiles (1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9, 1:2, 2:1)
Full parameter set: sampler, steps, CFG, shift, seed, seed mode, stochastic sampling gamma, batch count, strength, refiner model and start
LoRA list with per-LoRA weight sliders and +/− buttons; add LoRAs from the picker
img2img source image drop zone
Moodboard strip: reference images with per-image weight sliders (0–1); drag from Finder

Canvas

Pinch to zoom (0.5×–6.0×), drag to pan, double-tap to reset
Zoom percentage indicator
Drag-and-drop PNG onto canvas to inspect metadata

Right panel — Metadata tab

Generation parameters, model, LoRAs, dimensions, seed

Right panel — Assist tab

LLM operations system. Operations are Markdown files with YAML frontmatter stored in ~/Library/Application Support/TanqueStudio/LLMOperations/. Built-in operations are bundled with the app; users can add custom operations by dropping .md files into that folder.

Operations selector
Prompt input seeded from the currently selected gallery image's metadata
Result preview with Apply / Discard
Supports Ollama, LM Studio, and Jan

Right panel — Actions tab

Send All — apply prompt, negative, and full config to the left panel
Send Prompt — apply prompt and negative prompt only
Send Config — apply model, sampler, steps, CFG, seed, dimensions, LoRAs
Send to img2img — set current image as img2img source (uses visible crop when zoomed)
Add to Moodboard — add current image to the moodboard strip

Gallery strip

Saved images shown newest-first
Color-coded border (green = generated, gray = imported)
Tap to load image and metadata
Context menu: Reveal in Finder, Copy, Delete
Keyboard navigation in immersive mode (arrow keys, Escape)

Moodboard

Add reference images to influence generation via gRPC shuffle hints.

Drag image files from Finder directly into the moodboard strip
Per-image weight sliders (0.0–1.0)
Remove individual images or clear all
Works with models that support reference/shuffle hints (Qwen Image Edit, Flux, etc.)

DT Project Browser

Browse Draw Things project databases directly from the app.

Add folders containing .sqlite3 project files (local, external drives, network volumes)
Security-scoped bookmarks for persistent folder access across launches
Thumbnail grid with prompt preview, date, and dimensions
Search and pagination (50 entries per page)
Send to Generate — applies full config: prompt, negative, model, dimensions, steps, CFG, seed, sampler, seed mode, strength, shift, LoRAs; sets thumbnail as img2img source

StoryFlow (Labs)

A visual workflow editor for authoring Draw Things automation without hand-writing JSON.

Step types: accumulators (config, prompt, concat, wildcard, sweep), canvas ops (size, adaptSize, clear canvas), flow control (Loop/EndLoop), a human-in-the-loop approve step, and passthrough support for the rest of Draw Things' 34-entry instruction set
Most instructions run natively in-app — sweep, wildcard, concat, and inpaint tools execute without leaving Tanque Studio; the remainder export as Draw Things' own pipeline script format
Round-trips with Draw Things' own Editor: import an Editor-authored project, or export a Tanque Studio-authored one and paste it directly into a Draw Things StoryFlow script
Reusable config/prompt/image/LoRA/wildcard variables shared across steps

Story Studio (Labs)

Chapter-and-scene project structure for multi-render stories with assembled prompts.

Projects hold characters, settings, chapters, and scenes; each scene's prompt assembles live from its character/setting fragments plus its own description
Compiles to a StoryFlow workflow per scene or per chapter, with variant approval on each render
Per-field LLM enhance, plus a one-shot narrative writer for scene text
Chapter contact-sheet export — image sequence, storyboard, or comic grid, as PNG or PDF
Base Config is a saved Draw Things config (or a sensible built-in default) that's easy to change: "Use a saved config…" in Project Info, and a Settings toggle for what new projects start from

Render Queue (Labs)

Batch rendering across a matrix of prompts, models, LoRAs, and settings — the thing Draw Things can otherwise only do by hand-scripting.

Define axes (prompt, negative prompt, model, sampler, seed mode, LoRA sets, steps, seed, guidance scale, strength, shift) over a base prompt/config
Expand into a flat, editable list of concrete standalone jobs — prune, reorder, or leave any of them before running
Run, pause between jobs, and resume; a job that fails (unknown model, a server hiccup) never aborts the ones after it
Every job's config and prompt are captured as a full standalone snapshot, so a row stays reproducible even after the axes that produced it change

Settings

Draw Things connection — host, port, shared secret, history dropdown, test connection (gRPC)
LLM provider — Ollama / LM Studio / Jan; host with history dropdown, model, max tokens, test connection
Save folder — default save location (security-scoped bookmark)
Appearance — panel width defaults

Requirements

macOS 14.0 or later
Draw Things with API Server enabled
Optional: Ollama, LM Studio, or Jan for Assist tab features

Getting Started

Install Draw Things from the Mac App Store
Enable the API Server in Draw Things: Settings → API Server → Enable
Launch Tanque Studio — opens on the Dashboard
Configure connection via Settings in the top nav → Draw Things Connection
Default: localhost:7859 (gRPC)
Test Connection to verify connectivity
Pick a Quick Start preset (or the Continue card to resume a prior session) to enter a Focus Room, type a prompt, and click Generate

For Assist tab features (optional)

Install Ollama, LM Studio, or Jan
Configure the LLM provider in Settings → LLM Provider
Test the connection
In a Focus Room, click the ✨ Assist button next to the prompt field — it expands the Assist section in the drawer and runs the default operation automatically

For DT Project Browsing

Click Project Browser in the persistent top nav (reachable from any screen)
Click Add Folder and select a folder containing .sqlite3 project files
- Default Draw Things location: ~/Library/Containers/com.liuliu.draw-things/Data/Documents/
- External drives: navigate to any mounted volume under /Volumes/
Select a project database to browse with thumbnails and metadata

Bundled Config Presets

DrawThingsStudio/Resources/community_models_configs.json contains all 49 of Draw Things' built-in model configurations, pulled from the official drawthingsai/community-models repo and merged into the custom_configs.json array format. These ship inside the app bundle and appear automatically in the config picker under Built-in, alongside any configs you import yourself.

Source retrieved 2026-06-12. Note: each preset references a specific model file (e.g. flux_1_dev_q5p.ckpt) and applies fully only if that model is downloaded in Draw Things.

Architecture

DrawThingsStudio/
├── App & Navigation
│   ├── TanqueStudioApp.swift          # App entry, ModelContainer, migrations — WindowGroup root is DashboardRootView
│   ├── ContentView.swift              # Classic NavigationSplitView shell, sidebar (no longer the app root; kept for reference)
│   └── AppSettings.swift              # @Observable settings singleton (UserDefaults)
│
├── Dashboard (default navigation as of v0.9.25)
│   ├── DashboardRootView.swift        # WindowGroup root: mode switch (dashboard/focus/projects/labs/settings)
│   ├── DashboardTopBar.swift          # Wordmark, breadcrumb, persistent Project Browser/Labs/Settings nav
│   ├── DashboardHomeView.swift        # Continue/System/Quick Start/Recent Generations/Projects+Labs cards
│   ├── FocusRoomView.swift            # Full-bleed canvas + filmstrip; Paint/Crop/Color-Draw edit modes, scrubber
│   ├── DashboardFocusPanels.swift     # Focus Room's accordion drawer (Prompt/Assist/Model/Parameters/LoRAs/img2img/Actions)
│   ├── DashboardLabsPage.swift        # Labs pill-tabs (StoryFlow / Story Studio / Render Queue)
│   └── DashboardDS.swift              # Isolated light "paper" design tokens for this navigation
│
├── Generate (business logic + classic four-panel view, reused by both navigations)
│   ├── GenerateView.swift             # Classic four-panel root layout (config left, canvas center, gallery, inspect right)
│   ├── GenerateLeftPanel.swift        # Config: prompt, params, LoRAs, moodboard
│   ├── GenerateRightPanel.swift       # Metadata / Assist / Actions tabs
│   ├── GalleryStripView.swift         # Resizable gallery column
│   ├── GenerateViewModel.swift        # @MainActor @Observable ViewModel — shared by Dashboard and the classic view
│   └── ImageStorageManager.swift      # Writes PNG + thumbnail, creates TSImage
│
├── DT Project Browser
│   ├── DTProjectDatabase.swift        # SQLite + FlatBuffer reader
│   ├── DTProjectBrowserView.swift     # 3-column HSplitView browser
│   └── DTProjectBrowserViewModel.swift
│
├── StoryFlow (Labs)
│   ├── StoryFlowEngine.swift          # Accumulator engine: config + prompt, Loop/EndLoop, canvas ops
│   ├── StoryFlowModels.swift          # Variables, steps, workflows
│   ├── StoryFlowStorage.swift         # JSON-per-file storage, output folders, canvas PNG I/O
│   ├── StoryFlowProjectCodec.swift    # Lossless project export/import + DT pipeline export
│   └── StoryFlowView/ViewModel/*Panel.swift
│
├── Story Studio (Labs)
│   ├── StoryStudioModels.swift        # SwiftData schema v2: project/character/setting/chapter/scene
│   ├── StoryStudioView.swift          # Library ↔ workspace shell, outline column
│   ├── StoryStudioEditors.swift       # Project Info / chapter / character / setting editors
│   ├── StorySceneEditor.swift         # Scene editor + LLM-assist bar
│   └── StoryStudioRenderController.swift  # Compiles a scene/chapter to StoryFlow, runs it, routes results
│
├── Render Queue (Labs)
│   ├── RenderQueueModels.swift        # SwiftData axis/job models + UserDefaults-backed base settings
│   ├── RenderQueueExpander.swift      # Pure matrix → job-list expansion (axes × base config/prompt)
│   ├── RenderQueueController.swift    # Runs pending jobs, one direct render call per job
│   └── RenderQueueView.swift          # Base config, axis editor, job list (prune/reorder/pause/run)
│
├── Settings
│   └── SettingsView.swift
│
├── Support
│   ├── TanqueDS.swift                 # Tanque Design System tokens
│   ├── ImageFolderAccess.swift        # Security-scoped reads for custom folders
│   ├── LLMService.swift               # Ollama / LM Studio / Jan client
│   ├── LLMOperationLoader.swift       # File-based LLM operations
│   └── DTConfigImporter.swift         # Draw Things custom_configs.json import
│
├── Data & Persistence
│   └── DataModels.swift               # TSImage SwiftData model, ImageSource
│
└── Draw Things Integration (ported, do not modify)
    ├── DrawThingsProvider.swift        # Protocol + DrawThingsGenerationConfig
    ├── DrawThingsGRPCClient.swift      # gRPC transport (port 7859)
    ├── DrawThingsAssetManager.swift    # Local model/LoRA management
    ├── CloudModelCatalog.swift         # ~400 models from Draw Things GitHub
    ├── PNGMetadataParser.swift         # DTS, DT native, A1111, ComfyUI metadata
    └── RequestLogger.swift             # Debug request log

SwiftData schema — the core gallery model plus one schema group per Labs feature (TanqueStudioApp.swift, additive-only version bumps, no destructive wipe since v1):

@Model final class TSImage {
    var id: UUID
    var filePath: String
    var createdAt: Date
    var source: ImageSource       // .generated | .imported | .dtProject
    var configJSON: String?
    var collection: String?
    var batchID: UUID?
    var batchIndex: Int?
    var thumbnailData: Data?
}

Story Studio adds StoryProject / StoryCharacter / StorySetting / StoryChapter / StoryScene / SceneCharacterPresence; the Render Queue adds RenderQueueAxis / RenderQueueJob.

Roadmap

Completed

Upcoming

In priority order — all three scoped 2026-08-01, ready to build without further research:

StoryFlow: native enhance execution — the Phase 3 maturity gate that deferred all of Phase 4 is now satisfied. enhance reuses the exact LLMService call StorySceneLLMAssist.enhance(field:text:operation:apply:) already makes for Story Studio scene fields: one engine case assigning the LLM's answer to currentPrompt, plus adding "enhance" to StoryFlowRunPreflight.nativelyExecuted so preflight stops warning about it. interrogate stays deferred — it needs a multimodal call LLMService doesn't have (the image itself is already sitting in currentCanvasImage, just no API path to send it)
StoryFlow: Vision-framework removeBkgd / maskFG — via VNGenerateForegroundInstanceMaskRequest on currentCanvasImage, following the engine's existing .crop case pattern exactly (StoryFlowEngine.swift:359-371). The rest of Phase 5 stays deferred: maskBody has no single Vision API, depth ops carry real fidelity risk against DT's own model, poseExtract needs a joint-naming mapping DT doesn't document, and askZoom/maskAsk turned out to be Phase-4-shaped LLM work mislabeled as Vision work
StoryFlow polish: image-variable drag-drop import — turns out there's currently no way to attach an image to a .image-type StoryFlow variable from the UI at all, not just "no drag-drop." Ports the exact drag-drop + file-picker pattern already shipped for Story Studio characters (StoryStudioEditors.swift:405-479) into StoryFlowVariablesPanel.swift's .image case — imageData/imageFileName are already the fields the engine reads, only the editor UI is missing

Dropped, not deferred: promptInstruction's "replace mode" toggle. StoryFlow already has a standalone Clear Prompt instruction (WorkflowStepType.clearPrompt), and Story Studio's own compiler already uses exactly this composition — a Clear Prompt step immediately before a scene's prompt — to reset rather than append. A toggle would just be a second way to do what two existing instructions already do. One real caveat carried forward: Clear Prompt runs correctly in TanqueStudio's own engine but is silently dropped on export to DT's actual pipeline format (StoryFlowProjectCodec.swift:350 — DT only clears its accumulator as a side effect of an empty-prompt render), so a TS-authored "replace" sequence won't round-trip its clearing behavior back into DT's own scripting tool. Pre-existing limitation, unaffected by this decision either way.

Backlog

README demo GIF — a short GIF of one real workflow end to end. The four stills (Dashboard, Focus Room, DT Project Browser, Render Queue) are done — Screen Recording permission was granted to Claude 2026-08-01, screencapture works fine now
StoryFlow: native interrogate execution — needs LLMService extended with a multimodal call (Ollama's images field or an OpenAI-compatible image_url content part) plus making interrogate authorable (currently deliberately excluded from the instruction schema). Separable second piece after enhance ships
StoryFlow: remaining Vision-framework ops — maskBody (no single Vision API covers per-body-part segmentation), depthExtract/depthCanvas/depthToCanvas (fidelity risk vs. DT's own depth model), poseExtract (Vision's joint names need mapping to DT's undocumented pose JSON shape), askZoom/maskAsk (actually gated on the same multimodal-LLM gap as interrogate, not Vision-framework work)
Patterns Studio integration — WKWebView panel or PNG export feeding img2img
Gallery collections / organization
Soft-edged inpaint brush (mask transport is binary today)

Known Limitations

~~Model list may be empty on a shared-secret-protected server.~~ Fixed. The gRPC Echo call now sends the shared secret (upstream PR #16, merged 2026-07-03), so the model/LoRA inventory loads correctly on protected servers.

Acknowledgments

Draw Things by Liu Liu
DT-gRPC-Swift-Client — Swift gRPC client library for Draw Things
dtm by KC Jerrell — FlatBuffer schemas and database parsing approach that informed the DT Project Browser; its clip playback design (decode once, blit one frame per display tick, index derived from elapsed time); and its .mp4 export, which is where the audio-tensor layout, the AAC and BT.709 choices, and embedding the generation config in the exported file all come from. dtm reaches those through ffmpeg and Tanque Studio uses AVFoundation, so no code is shared — what was borrowed is the knowledge of what to do

License

MIT License

Name		Name	Last commit message	Last commit date
Latest commit History 661 Commits
Docs		Docs
DrawThingsStudio		DrawThingsStudio
DrawThingsStudioUITests		DrawThingsStudioUITests
Scripts		Scripts
TanqueStudio.xcodeproj		TanqueStudio.xcodeproj
TanqueStudioTests		TanqueStudioTests
design_handoff_dashboard_final		design_handoff_dashboard_final
design_handoff_palette_stage		design_handoff_palette_stage
.gitignore		.gitignore
README.md		README.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Tanque Studio

Screenshots

Features

Dashboard + Focus Rooms

Generate

Moodboard

DT Project Browser

StoryFlow (Labs)

Story Studio (Labs)

Render Queue (Labs)

Settings

Requirements

Getting Started

For Assist tab features (optional)

For DT Project Browsing

Bundled Config Presets

Architecture

Roadmap

Completed

Upcoming

Backlog

Known Limitations

Acknowledgments

License

About

Releases

Packages

Used by

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

Tanque Studio

Screenshots

Features

Dashboard + Focus Rooms

Generate

Moodboard

DT Project Browser

StoryFlow (Labs)

Story Studio (Labs)

Render Queue (Labs)

Settings

Requirements

Getting Started

For Assist tab features (optional)

For DT Project Browsing

Bundled Config Presets

Architecture

Roadmap

Completed

Upcoming

Backlog

Known Limitations

Acknowledgments

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages