Settings

Service Providers, Workflow Packs, Prompt Templates, and System Settings.

Service Providers

Settings → Service Providers: name, type, address, capabilities, key, and enabled. Use http:// for the address (writing https:// locally causes a protocol error). The API key for ComfyUI can be left empty. If speech sets LIBER_SPEECH_API_KEY, enter the same key in the provider.

The search box on the left filters only provider names and descriptions; it is not a login box. Open an entry to fill in the API address and key. The type name for Liber Depth is "Liber Depth Depth Map", the default address is http://127.0.0.1:5556, and its capability is depth map. Enabling the local plugin creates this entry automatically; to connect to another depth service on the LAN, change the address here instead of starting a remote process from the Plugins page.

The image / video / speech / recognition service dropdowns on each page list only enabled entries with matching capabilities (such as image, video, tts, asr). The Depth Map page lists only enabled Liber Depth providers.

Workflow Packs

Settings → Workflow Packs can import:

1. Full LiberAstra pack (with workflow + inputs and other fields)
2. ComfyUI API Format (Save (API Format))
3. ComfyUI canvas JSON (with nodes / links, optionally with subgraphs): after importing a file or pasting it, click "Convert canvas JSON" to convert it to API Format and heuristically generate capabilities, parameter mappings, output nodes, and optional accel / fps drafts — verify the node / field mappings before importing.

Location	Description
backend/workflows/	Built-in (shipped with the release)
backend/data/workflows/	User packs; same-name packs override built-ins

Capability vocabulary: t2i text-to-image, i2i image-to-image, upscale image enhance, v2v video enhance, t2v text-to-video (Quick Create · Generate Video), i2v image-to-video / asset animation, avatar digital human, motion motion transfer, tts speech, analysis analyze media and write the script. The three concat tools under Tools are local native tools, not workflow packs, and do not appear here.

Optional top-level accel: the frontend shows a Turbo toggle and submits __accel. The pack dropdown first shows the capability name and a one-line description; uncensored packs are marked "Uncensored". The lightbulb on the right shows the full description from the manifest.

Function nodes on the canvas choose packs by their "Group". Groups are maintained on this page under the "Canvas Groups" tab: assign packs to groups, rename them, or delete them. See Canvas → Workflow Groups.

Prompt Templates

Settings → Prompt Templates. When you click ✨ to optimize, the system prompt sent to the default LLM is assembled from two parts: the function template explains what this task is doing, and the scene preset adds a layer of style after it. Built-in entries ship with the release; user files store only overrides and custom entries. If the saved content matches the built-in content, it stays "default" and does not show "Customized"; only actual edits show that. Click "Reset to default" to remove the override and return to the text shipped with the release. The next optimization uses the new content immediately after saving.

Tabs and buttons:

• Function Templates / Scene Presets: select an entry on the left and edit its content on the right; the save button is disabled when nothing has changed.
• New: enter an ID. The ID cannot be changed after saving. Saving with the same ID as a built-in entry only creates a custom override when the content differs.
• Import: paste JSON. The structure is { "tasks": { "some-id": { "name", "system", "capabilities" } } } or { "scenes": { "some-id": { "name", "append" } } }. A same-name entry only becomes custom when its content differs from the built-in.
• Delete: user-created "My Templates / My Presets" can be deleted; built-in entries cannot be deleted, only reset to default.
• Export: download the current merged result as prompts-catalog.json.

You can also edit data/prompts/tasks.yaml and scenes.yaml directly. Built-in templates ship with the app package: the source is in backend/prompts/, and in the installer it is under liberastra/prompts/. data/prompts/ holds only user overrides and custom entries; full copies left over from older versions are compared with the built-ins, and content identical to the old defaults is cleaned up automatically and not mistakenly marked as custom. Container deployments do not need to copy built-in files manually. After an upgrade, if an old copy differs from the new built-in content, it is kept and marked "Customized"; click "Reset to default" to adopt the new built-in content.

Function Templates

Each page and the canvas pass a fixed template ID to the optimization endpoint. The table below lists the built-in IDs. The "Applies to" field on a function template is a comma-separated list of capabilities, used to show which capability it corresponds to on this page; when empty, this page does not list that capability.

ID	Name	Used by
t2i	Text to Image	Text to Image page; canvas function nodes with capability t2i
i2i	Image to Image / Inpainting	Image to Image & Edit
upscale	Image Enhance	Tools → Image Enhance; canvas upscale packs
i2v	Image to Video	Image to Video, Custom Assets animation prompts; canvas i2v packs
t2v	Text to Video	Canvas t2v packs
quick_visual	Quick Create · Visual Prompt	Quick Create visual prompts
motion	Motion Transfer	Motion Transfer page
quick_topic	Quick Create · Topic Expansion	Quick Create topics
assets_topic	Custom Assets · Topic Expansion	Custom Assets topics
character_sheet	Characters · Four-Panel Character Sheet	Characters "Generate from description"

Canvas function nodes use the template with the same name as the selected pack's capability. When there is no matching template, they fall back to t2i. Visual analysis does not use this prompt field; the analysis angle is written in the pack parameters. Characters always use character_sheet and does not share t2i with the Text to Image page. The upscale template only applies to custom upscale packs that declare a prompt slot; the built-in upscale_basic and SeedVR2 image enhance have no prompt field.

Scene Presets

A scene only appends style; it does not replace the function template above. The PromptField scene dropdown comes from here. There are nine built-in entries:

ID	Name	Tendency
auto	Auto / General	Does not force poster, portrait, or other fixed scenes; still follows the function template
photo	Photography & Portraits	Camera body and film texture, focal length, candid framing
poster	Posters & Layout	Main title, info bar, limited palette; in-image text given word by word
illustration	Illustration & Art	Line work, coloring, material
product	Product & E-commerce	Hero image, material close-up, selling-point copy
character	Character & Persona	Standing art, outfit, expression
infographic	Infographic & Visualization	Numbered sections and short labels
architecture	Architecture & Space	Material, light and shadow, perspective
cinematic_video	Cinematic Storyboard Video	Shot size, camera movement, temporal action

Choosing "Auto" still includes the auto overlay text, which explicitly says not to force any particular visual type.

PromptField on Each Page

Text to Image, Image to Image, Image to Video, Motion Transfer, Quick Create, Custom Assets, Characters, and canvas function nodes with prompt slots all use the same prompt entry: switch between the original input and the optimized draft, scene presets, ✨, and "Use optimized prompt". The original draft, optimized draft, and toggle are stored in the local browser and survive a refresh.

SeedVR2 image/video enhance and regular upscale packs have no prompt slot and do not show this field.

✨ submits the prompt.optimize job: POST /api/v1/prompts/optimize returns a job ID and the frontend polls progress. The server streams from the default LLM in System Settings. After a refresh, the job ID is enough to keep showing completion. If a workflow pack name is included during optimization, the pack's capability, description, and model family are passed to the LLM as well. ?wait=true waits synchronously. The legacy path /api/v1/images/optimize-prompt still returns synchronously.

API: GET /api/v1/prompts lists the merged templates; PUT /api/v1/prompts/tasks/{id} and PUT /api/v1/prompts/scenes/{id} save overrides; the corresponding DELETE resets to default; POST /api/v1/prompts/import imports.

System Settings

Change account password.

Default LLM (scripts and prompt optimization): choose the provider and model used by Quick Create, Custom Assets, each function page, and ✨ on canvas function nodes. Edit the template text in the previous section, "Prompt Templates".

Depth Map Generation: choose the default depth provider and model for the Depth Map page and canvas depth generation. When empty, the first enabled Liber Depth and dpt_swin2_large_384 are used.

Speech synthesis (Liber Speech): choose the text-to-speech model, speech-to-text model, and voice conversion model separately. Text-to-speech defaults to multilingual_v3 (Chatterbox-Multilingual V3); speech-to-text and voice conversion follow the speech service configuration (.env) by default, and you can also specify them here. Text to Speech, Speech to Text, Voice Conversion, Quick Create, Custom Assets, and Digital Human all use these defaults; function pages can still override them.

Default Character Generation: used only by Characters "Generate from description". The image provider defaults to the first enabled image service; the three packs below correspond to no reference image, one reference image, and two to three reference images:

• Character sheet workflow pack: lists only t2i packs; defaults to qwen_image_21.
• Single-reference workflow: lists only i2i packs; defaults to qwen_image_21_edit.
• Multi-reference workflow: lists only i2i packs; defaults to qwen_image_21_edit_multi.

Change official or uncensored image models here instead of going back to the Characters page.

Runtime settings: ComfyUI wait timeout, synchronous API wait, and LLM timeout; "Serialize services on the same host" is off by default (parallel). When enabled, multiple services on the same host run only one job at a time. Changes take effect on the next job after saving.

---

Related: Canvas → Workflow Groups; Works & Tasks → Plugins; Image / Video / Speech.