Settings

Find current workspace, creation, connection and privacy preferences

Last updated: 2026-11-11

Open Settings from the app navigation. Use a category to focus its controls, or All to see the available cards. App preferences are separate from the current render recipe; Workflow save/apply belongs to the header.

Categories

Group Category Use it for
Workspace Profile Choose public identity, biography/links and optional avatar/automatic credits. Review the displayed effective identity before saving or sharing.
Workspace UI Interface preferences and Learning Mode.
Workspace Hotkeys Inspect/customize supported shortcuts, resolve conflicts and restore defaults.
Workspace Projects Project preferences; current output location still follows the active Project.
Workspace Storage Data root, global output override and storage usage/cleanup.
Create Render Completion behavior, navigation and tile-history retention.
Create Plugins Available source/plugin switches and readiness/refusal messages. The category is Plugins, under the Create group.
Create Performance Supported precision/placement/performance choices; inspect hardware guidance.
Create Vision Local image-description model and its settings.
Connections Gallery Explicit account connection/consent, connection status and disconnect.
Connections HuggingFace Credentials/access for selected model downloads.
Connections API Keys Your chosen AI provider credentials and discovered models.
Connections MCP Connect external local agents and inspect availability.
System Privacy Separate optional usage tracking and crash reporting.
System Developer Advanced controls; use only when a specific workflow/support instruction calls for them.

Software-update channel choices are in Software Update, not a generic Settings category. See Updates. The License… dialog is opened from Help; see License Activation.

Profile and public credits

Choose Local, Gallery or None, review the saved identity and configure future Shader crediting/avatar inclusion. Gallery overrides affect local credits, not the online account profile. Normal valid edits save on field exit/discrete changes; read Saved/error status, and reconcile an uncertain save before retrying. See Profile and Credits for the complete identity, image, account-change and repair task.

UI and Learning Mode

UI owns presentation and viewer preferences, and Learning Mode. Keyboard Shortcuts explains focus, fullscreen and control levels; Interface explains workspace/viewer behavior. Learning Mode is off by default and displays bundled guidance, independently of Copilot or an AI provider. Help > Learning Mode changes the same preference.

GPU Information

Use GPU Management and System Requirements for device/memory guidance. Model catalogue fit is an estimate; it does not reserve VRAM or guarantee a workload fits. Installed driver/runtime information does not require a separate CUDA Toolkit installation.

VRAM Requirements by Pipeline

Read the current selected model's hardware/precision guidance. There is no one memory minimum for every model or every operation, and fixed “recommended VRAM” tables do not account for your input, add-ons, offload or other GPU work.

Vision and Describe

Settings > Vision controls the local models used by Describe, separately from the diffusion model that renders an image. Florence-2 serves Fast/Quality; Qwen2.5-VL 3B serves Best. Qwen2.5-VL is not Qwen-Image diffusion.

In the model table, use Install for the model you need and wait for the download/ready result. Resolve the selected entry's access requirement through its offered authentication action. A queued download is not a loaded model. Choose the Describe mode where that action is offered; installing a model does not automatically describe an image.

Choose the model-lifetime option for your workload:

  • Auto (recommended) loads for Describe and keeps the model only while GPU room permits. A later GPU job can release it when that job needs the space.
  • Keep resident keeps it after Describe until GPU pressure or manual Unload; it does not reserve memory against every later task.
  • Load on demand loads for Describe and unloads afterward, trading repeated loading time for lower retained VRAM use.

Unload releases the loaded model without deleting downloaded weights. Deleting cached weights is a separate action and can require unloading first; do not delete them merely to free current GPU memory. See Model Management for download/repair and GPU Management for memory inspection.

Output Folder Configuration

Output Directory

Settings → Storage distinguishes the data root, global output override and the active Project's output. An open Project can override the global location. Inspect the effective path before a job/export; do not assume it is a folder beside the installed app.

Choose only the offered location/move/reset action and follow its active-work guards. Changing a default is not proof that every existing file moved. Keep important work backed up before relocating storage.

For the actual Change Location, output override and disposable-cache controls, follow Storage and Backup. Data-root and new-render destination changes do not move existing files.

Folder Structure

Use the app's locate/reveal actions to find the actual item. Managed renders, sequences, generated-source history, user presets and installed weights have different storage owners. A saved Workflow is a recipe, not an archive of these files. See Projects for per-Project output and migration.

Render Options

Auto-cleanup Tiles After Render

The current control is Keep tile history in Settings → Render. Choose Keep 0 (clean up after render), Keep last 3, Keep last 5 or Keep all. It governs retained tile files from recent diffusion steps, not universal deletion of all graph history or exported output. Choose the retention you need before relying on later tile inspection/reuse.

Auto-open Folder on Completion

Open folder when render completes controls opening the output location in Explorer. It is separate from a successful export and can interrupt your desktop workflow.

Sequence export names

Rename exported files to sequential names and Filename Prefix govern whether Copy/Convert applies the export dialog's numbering choice. By default, names/stems are retained. With this global option on, packing starts at 0000; see Exporting Sequences before relying on the dialog's numbering examples.

Save Config with Renders

There is no separate switch with this old label. Use Save as Workflow… to save a recipe deliberately. Existing managed render data and exported delivery files are different objects.

HuggingFace Authentication

For a model that reports gated access, accept that exact repository's terms using your model-host account and provide the required token through the app's authentication dialog/settings. Authentication does not by itself grant access to every repository.

Setting Up Authentication

Choose the model first, follow its access message, accept the applicable terms and test/save your token through Settings → HuggingFace or the download's offered authentication action. Keep keys/tokens out of public reports. See Model Management.

Authentication Status

Read the actual validation/access result. A valid token, accepted repository access, downloaded weights and runtime readiness are separate conditions.

Gated Models

Some SD 3.5 models need accepted gated access. Do not assume every Qwen model is gated: Qwen-Image diffusion and Qwen2.5-VL vision/Describe are different models with different catalogue entries. Follow the selected entry's requirements rather than substituting their repositories.

Quick Access

Use the model/source's requirements action when offered to install or resolve access without searching for a retired Settings → Models screen.

License Activation

Open Help → License… for activation and device management. Initial online verification and the seven-day offline window are distinct from local signature checks.

Activating Your License

Use account approval or a full license key in that dialog and wait for operational status. A pending verification is not unlocked rendering.

Deactivating Your License

Use the dialog or your account's device management. Local deactivation and confirmed remote slot release can have different outcomes; inspect the result. See License Activation.

Desktop activation does not automatically connect Gallery. In Settings → Gallery, choose Connect, complete the browser account/terms approval and return to confirm connection status. Resolve an expired/refused connection through the offered reconnect flow. Disconnect is separate from desktop license deactivation.

The publication dialog owns the chosen image, title/tags/collection, output choices and recipe-sharing decision; these are not generic upload defaults in Gallery Settings. See Gallery Publishing.

API Keys and AI Providers

Use API Keys for the selected service's credentials: Claude, OpenAI, xAI and Civitai have separate entries. Civitai access for selected downloads is not an AI conversation provider. Hugging Face authentication lives in its own category.

Enter your key, use Test if desired, then Save and check the stored status. A successful Test does not itself store the key. Keep credentials out of prompts, shared Workflows and support screenshots/reports.

Provider preference, discovered models and Copilot choices apply only to features that support them; a key alone does not grant every model or tool permission. Read availability/access errors and any refresh cooldown before retrying. AI requests can send selected context to the provider and incur charges. See Copilot and Privacy for the feature-specific paths rather than assuming every service has the same fallback.

Config Manager

Workflow save/apply is in the header's identity caret, not a separate Config Manager.

Saving Configurations

Use Save as Workflow… for the current recipe and choose its filename/location.

Loading Configurations

With an open render, use Apply Workflow to…, select .hextile.json and review the offered replace/fork target. Resolve referenced input/model requirements before rendering. It does not include images or weights.

Recent Configurations

Use the actual offered file/history choices. Presets and Workflows explains sharing and the distinction from Worlds, presets and scenes.

Keyboard Shortcuts

Use Settings → Hotkeys for current assignments and Keyboard Shortcuts for contextual behavior. Single-key hotkeys (switch Enable single-key hotkeys) starts on. It controls bare character keys such as F, C, R, and W/A/S/D. Shader playback Space is not one of those keys; video Space is (see Video). A shortcut's action depends on focus and the current tool; do not assume a global file command is every Workshop's Save.

Privacy and online services

Settings → Privacy separates usage tracking from crash reporting. Turning these off does not stop license checks or user-selected online features. AI provider use can carry provider charges and Gallery publication shares chosen content. See Privacy and Telemetry.

Plugin availability

Settings → Plugins offers switches for configurable bundled sources such as Pattern, 360 LoRA, Outpaint, Fractal and Shader. Other bundled contributions may have no switch. Read the displayed unavailable/partial message rather than trying to force-enable unsupported work. Disabling fresh work does not erase previously saved images/scenes or prove an admitted job was cancelled.

The Shader row is in the Create group. Its switch is Enable Shader. The row always shows: "When off, new Shader authoring and output are blocked. Saved Shaders and accepted jobs remain available." With the switch off, the Shader page says Shader is paused. and "Your draft stays here until you enable Shader or choose to discard it. Saved Shaders and accepted jobs remain available." Enable Shader turns the plugin back on. Return to CREATE leaves; an unsaved draft asks "Discard the unsaved Shader draft and return to CREATE?" If that enable fails, the page says "Could not enable Shader. Try again."

When a contribution cannot run, the row says "Hextile can't use this plugin right now." If this version of the app has not checked the plugin, it says "Hextile hasn't checked this plugin against this version of the app yet, so it stays off." The Shader page then says Shader is unavailable. or The workshop is unavailable right now. A failed settings save says "Could not update Shader" and "Your previous setting is still in effect. Try again."

MCP connection and plugin/source enablement are separate decisions. See Agent Mode for agent setup and Create for source-specific requirements.

Esc
Searching...
No results found.
Type to search the documentation
↑↓ Navigate ↵ Select Esc Close

We use cookies for analytics to understand how visitors use our site. Privacy Policy