Documentation/The studio guide

A field guide to PotionUI

Your studio,
explained.

From a first idea to a collection worth keeping. Learn the tools, find your workflow, and make PotionUI your own.

Images · Video · Music · 3DOpen source · Alpha

One workspace, many ways to create

PotionUI is a self-hosted AI studio. Open it in your browser, choose a model-specific preset, and work with a form designed for that task. Generation can run on your own machine or on a configured remote backend.

A real First Light product study: model controls on the left, prompt composition in the center, and the result in the Workbench. The generation workspace
A real First Light product study: model controls on the left, prompt composition in the center, and the result in the Workbench.

Find your way around

Read from the beginning or use the section menu to jump to a feature. Screenshots open in a viewer with a full-size option, so you can inspect the actual controls. The First Light campaign appears throughout as a worked example.

An evolving studio

PotionUI is in alpha. Available controls depend on your preset, enabled plugins, and account access. Screenshots show real app captures; your installed version may look a little different.

01 / Start here

Your first generation

A preset, a prompt, and somewhere to run it. That’s the starting point.

Open your studio

If you haven’t installed PotionUI, begin with the installation options and hardware guide ↗. Open the address printed by the launcher in your browser.

On a fresh installation, create the owner account. The first account becomes an administrator. If you connect from another machine, enter the one-time claim code printed by the server. Guided setup then helps you choose a model directory, configure a backend, and approve any required downloads.

Joining an existing studio? Sign in with the account your administrator created, or register if the owner has enabled registration.

Make an image

  1. Choose a preset and mode

    Open Generate. Pick an available image preset, then its text-to-image mode. The form changes to show that preset’s controls.

  2. Describe what you want

    Start with one prompt segment: a subject, a setting, and the light. Add negative direction if your preset supports it.

  3. Keep the first run simple

    Use the preset’s defaults, select a supported resolution, and generate one image. Required model files must be installed and available to your account.

  4. Generate, then refine

    Press Generate. Watch progress in the Workbench, inspect the finished image, and change one part of your idea for the next run. The result and its recorded settings are saved to History.

A preset is ready-made; its models download separately.

Installing a preset gives you the workflow and controls. Guided setup or model management supplies the model files it needs.

Full getting-started guide ↗

02 / Create

Generation workspace

Keep the idea, the controls, and the result in view.

Read the workspace

Preset & mode
Choose the tool and the kind of generation. Switching presets resets the mode and form to match the new tool.
Prompt editor
Compose positive and negative directions as ordered segment cards. Some presets use multiple prompt lists or Video Director.
Generation form
Choose models, dimensions, seed, quantity, and any optional controls the preset exposes.
Workbench
Inspect previews and results. Resize the preview area and step through batch items or video outputs.

Follow a run

Generate starts the job; while it runs, the control becomes Cancel. Progress reports the current pipeline stage, such as loading models, encoding the prompt, generating, or upscaling. Intermediate previews appear when the pipeline provides them.

Jobs run on the server. Switching tabs, leaving the page, or reloading does not cancel a run. PotionUI reconnects to its state when you return. The workspace identifies a busy tab so you can jump back to it.

Inspect and compare

Open a result at full size or send it back to the Workbench. Compare two images with a movable divider to inspect differences. A run may also expose artifacts: the actual seed, applied models, or a before-and-after image from an enhancement step.

The Amber and Moonlight studies compared with a draggable divider in the actual app. Comparing results in the Workbench
The Amber and Moonlight studies compared with a draggable divider in the actual app.

Generation reference ↗

03 / Create

Presets & forms

The right controls for the model you’re using.

What a preset gives you

A preset connects a generation pipeline to a purpose-built form. It defines supported modes, required models, controls, and defaults. Choose an installed preset to start creating; authoring your own is optional.

ModeUse it to…
Text to imageCreate an image from a written description.
Image to imageTransform an input image using a prompt and the available strength controls.
InpaintEdit a masked region of an image.
Enhance / upscaleIncrease resolution or refine an existing result.
Video, audio, or 3DWork with a preset designed for that medium.

Each preset offers its own subset of modes. A field may appear only after you enable a related option.

Understand common controls

  • Resolution: image dimensions or a named aspect ratio supported by the preset.
  • Steps: how many sampling steps a run uses. Follow the preset’s defaults before tuning.
  • Guidance / CFG: how strongly the model follows the prompt, where supported.
  • Seed: a starting value for randomness. -1 requests a fresh seed; History records the actual value used.
  • LoRAs: compatible add-on models for a style, character, or subject, with individual strength controls.

Shape a preset’s form

Administrators can change defaults, lock settings, or hide fields in an installed preset. For a custom tool, authors define fields in form.yml and connect their values to processing steps in pipeline.yml.

Preset administration exposes defaults and field overrides for the Krea-2 generation form. Customizing a preset form
Preset administration exposes defaults and field overrides for the Krea-2 generation form.

Presets & forms guide ↗Preset authoring ↗

04 / Create

Video Director

Think in shots. Give each moment its own direction.

A four-shot First Light film plan, with timed direction and reviewed campaign images as starting frames. Video Director
A four-shot First Light film plan, with timed direction and reviewed campaign images as starting frames.

Choose how a clip begins

Compatible native video presets can offer text-to-video, image-to-video, first-and-last-frame, and Director modes. Use a written brief, a starting image, or a pair of frames as your starting point. The selected preset determines which modes and controls appear.

Build the direction

Use the familiar segment editor for global direction and individual beats. In a timeline-style preset such as LTX, each shot has its own duration, timed prompts, and supported keyframe or audio inputs. Shots generate separately in order.

Chain-style presets such as Wan connect sequential segments using a tail-frame handoff. A continuation in the LTX editor instead uses the previous shot’s rendered output as a starting frame; that output needs to exist before the next shot can run.

Capabilities belong to the preset.

Keyframes, audio, reference conditioning, and duration limits vary by model. LTX and MiniMax-H3 integrations are experimental. Check the model guide before planning a render.

Video Director reference ↗

05 / Create

Beyond the image

Explore sound and shape through the same preset-driven studio.

Music Experimental

The MiniMax-Music3 song preset combines lyrics in the prompt editor with a Style description in the form. Describe genre, tempo, mood, vocal character, and arrangement. Keep verses and choruses in reusable prompt segments, or enable Instrumental for a track without vocals.

The duration control sets a maximum, rather than an exact target. Song structure and lyrics influence the natural ending. Required audio models and their hardware demands are separate from image generation.

3D from a reference

The TRELLIS.2 preset turns a reference image into a 3D mesh. Open mesh results in History to inspect materials and wireframes. This is a separate generation task with its own models and input requirements.

Music preset ↗TRELLIS.2 preset ↗

06 / Compose

Prompts & segments

Build an idea in pieces. Keep the pieces that work.

Four related tools

The Prompts workspace contains four libraries. They share the same rich segment format, including text, phrasebook chips, enabled state, and optional names, colors, and descriptions.

Prompts
Complete ordered compositions. Save a finished product brief or a negative prompt for another generation.
Segments
Single named, categorized cards. Reuse a character description or a lighting direction by replacing one editor card.
Segment Templates
Ordered sets of slots, with optional starter content. Start a new brief with subject, setting, lighting, and camera cards already arranged.
Segment Categories
Names and colors that organize saved Segments, such as Product, Lighting, or Camera.
An ordered product-editorial template gives a new brief a repeatable structure. Segment Templates
An ordered product-editorial template gives a new brief a repeatable structure.

Save and apply a composition

Use Save as Prompt on a whole list or Save as Segment on one card. When applying a Prompt or Template, choose Append, Prepend, or Replace. Replacing meaningful content asks for confirmation.

Inserted cards are independent copies. Editing them doesn’t update the saved item. Applying a library item changes only the prompt list you target; use a session to preserve the preset, mode, and form as well.

Arrange the idea

Reorder, duplicate, or disable cards as you refine a prompt. Collapsing a card only folds the editor; disabling excludes it from the generated prompt. Positive and negative lists remain independent.

Prompts workspace guide ↗

07 / Compose

Phrasebook

A vocabulary for your work, always within reach.

Reusable campaign language organized into lighting, locations, camera, palette, treatments, and motion. The First Light phrasebook
Reusable campaign language organized into lighting, locations, camera, palette, treatments, and motion.

Build your vocabulary

Create categories for terms you use often, with nested categories for a larger collection. Each value has a short label and the actual text inserted into your prompt. For example, a “Soft daylight” label can stand for a more detailed lighting description.

Manage categories and values in Phrasebook. Mark entries active or inactive, and filter the tree to focus on the ones you use. A configured assistant can help refine the inserted text.

Insert, pin, or shuffle

Use # in a prompt to find phrasebook suggestions. Accept a suggestion to insert a chip, then choose a specific value or enable shuffle to explore alternatives from that category across generations.

Saved Prompts, Segments, and Templates retain chip selections and shuffle settings. Generation details record the resolved composition, so you can see which wording a result used.

Phrasebook guide ↗

08 / Compose

Variables & choices

Keep a detail consistent, or leave room for a useful surprise.

Define a detail once

Variables are named values scoped to a Generate tab. Put a repeated product description in a variable, then insert its chip wherever you need that description. Updating the variable changes what those references resolve to.

The Variables editor stores reusable named values for the current generation workspace. Prompt variables
The Variables editor stores reusable named values for the current generation workspace.

Explore alternatives

Choice chips select from alternatives, including weighted choices and multiple selections. The editor manages these as chips; the underlying syntax also works when importing or copying prompts.

Prompt syntax · examples
${product} in {daylight|moonlight}
{0.7::warm light|0.3::cool light}

In the first example, ${product} inserts a defined variable and the choice supplies one lighting direction. The second example gives warm light a higher selection weight. Check variable names: an undefined variable resolves to empty text.

Prompt expansion reference ↗

09 / Compose

AI assistant & MCP

An optional collaborator beside your creative tools.

Work with the assistant

Once an administrator configures a language model and gives you access, open the chat panel to brainstorm, expand a prompt, or refine a segment. Choose among available models and modes, or attach an image when the model supports vision.

Enable Tools to let the assistant propose actions, such as updating prompt text, form settings, or phrasebook values. Changes in the built-in chat wait for your approval. With Tools off, the assistant replies with text. Pin a conversation to a tab to keep its actions aimed at that workspace.

Conversations have their own saved sessions. Use /help for built-in commands or /tools to inspect available tools.

Connect an external client

MCP, the Model Context Protocol, lets a compatible external client work with your PotionUI account. An administrator must enable MCP for the instance and your user. Create a named token in Settings → MCP, copy it when shown, and connect to your server’s /api/mcp endpoint using bearer authentication.

A token carries your account’s access.

Keep it private and revoke it in Settings when no longer needed. External clients handle their own action approvals; changes do not open a second approval prompt in PotionUI. Tools requiring a live Generate editor are unavailable through MCP.

Assistant guide ↗MCP setup ↗

10 / Organize

Workspaces & sessions

Leave an idea open. Come back with a fresh eye.

Keep ideas side by side

Each workspace tab holds its own preset, mode, prompt composition, form values, and results. Add a tab with the plus button and switch between ideas without replacing their setups. Tabs are remembered between visits.

Save a setup by name

Use a session to save the whole creative setup: preset, mode, prompts, and form values. Name sessions for a direction you want to return to, such as “Product / daylight” or “Launch film.” Restore one when you want to continue that setup.

Two saved Lumen Atelier creative directions available from the workspace’s Sessions menu. Saved generation sessions
Two saved Lumen Atelier creative directions available from the workspace’s Sessions menu.
Choose what you want to keep.

A Prompt saves composition. A Session saves the generation setup. History keeps the actual run and its recorded result.

11 / Organize

History & collections

The result is only half the story. Keep its recipe, too.

First Light image studies organized into project collections, with tags and reviewed selections. History and project collections
First Light image studies organized into project collections, with tags and reviewed selections.

Find the right run

History automatically records generations. Search prompt text and narrow by date, media type, status, preset, mode, or model. Phrasebook filtering helps find runs that used a particular saved value.

Use tags for labels that cross projects, such as “select” or “needs review.” Use collections to gather related work into a project structure. Combine filters to find a particular direction without paging through every experiment.

Recover the recipe

Open a generation’s details to inspect the recorded prompt composition, resolved phrasebook values, seed, dimensions, preset version, model files, and parameters. Reuse that setup as a starting point for another variation.

The original First Light product study alongside its recorded prompt, seed, model files, and settings. Generation details
The original First Light product study alongside its recorded prompt, seed, model files, and settings.

Keep the useful work

Select results for bulk tagging or cleanup. Deleting generations is permanent and asks for confirmation. Check the selection and filters before confirming, especially when deleting by tag.

History & tags guide ↗

12 / Organize

Library & Inspirations

Give your next idea a useful starting point.

Build a reference library

Keep reusable media in the Library, upload reference files, and arrange them in collections. Gather product images, locations, character references, or the starting frames for a film in one place.

Selected campaign images grouped as product, location, people, and art-direction references. The creative Library
Selected campaign images grouped as product, location, people, and art-direction references.

Share the thinking

Inspirations brings selected work and its creative direction into a browsable studio collection. Inspect a useful example, save it as a reference, or take its generation recipe into a new exploration.

A selected First Light study with its original settings and an art-direction note. Studio Inspirations
A selected First Light study with its original settings and an art-direction note.

13 / Make it yours

Models & backends

Choose the tools. Choose where the work happens.

Know what’s installed

Models is the inventory of actual model files: base checkpoints, LoRAs, text encoders, VAEs, and upscalers. Filter by type, tag, or name. Open a model to inspect its size, details, previews, and generations made with it.

A preset’s model pickers show compatible files available to your account. Administrators manage downloads and assignments. Provider plugins can add metadata and download links from external catalogs.

Model files assigned to the example studio for image and video generation, text encoding, and decoding. Model management
Model files assigned to the example studio for image and video generation, text encoding, and decoding.

Connect the compute

An engine is the kind of pipeline a preset runs. A backend is a configured place that runs that engine. Native presets can run locally or on a compatible remote native worker; ComfyUI presets require a configured ComfyUI server.

Administrators configure and enable backends, connection details, and applicable device settings. A preset needs a backend that speaks its engine’s protocol. A remote setup lets the browser-facing studio and GPU worker live on different machines.

Plan for the model you want to run.

Memory needs vary by model, resolution, and workflow. Weights download separately and carry their own licenses. Use the hardware guide for model-specific requirements and validation status.

Hardware guide ↗Backend reference ↗

14 / Make it yours

Studio administration

One installation. The right tools for each person.

Producer, stills, and motion accounts in the example Northline Studio, with roles and resource access controls. Users and access
Producer, stills, and motion accounts in the example Northline Studio, with roles and resource access controls.

Manage people and access

Administrators create accounts, set roles, and organize people into groups. Assign presets, model access, and assistant configurations to individuals or groups. A shared group makes it easier to give a team the same tools.

Regular users see the resources assigned to them. Administration is available to accounts with admin rights. The exact features a person sees also depend on enabled plugins and server configuration.

Maintain the installation

  • Presets: install tools, manage access, and customize form defaults and visibility.
  • Models & backends: manage files, downloads, compute connections, and device options.
  • LLM configuration: connect supported providers or native checkpoints, then assign access.
  • System settings: configure storage, thumbnail profiles, housekeeping, and backups.

Back up your studio

Use Administration → System Settings → Backups to configure backup options and run a backup. Restoration is a command-line task with the app stopped. Check the backup guide for what each tier includes, model-file handling, and restoring onto a fresh installation.

Administration guide ↗Backup & restore ↗

15 / Make it yours

Automations

Set up the routine around the creative work.

Connect a trigger to an action

The visual automation editor connects triggers through conditions to actions. Triggers can include a schedule, a manual run, a file watcher, GPU memory thresholds, or application events. Actions can organize results, assign resources, index models, control backends, or send notifications.

TriggerFile addedActionIndex modelsActionNotify in app
A paused automation draft connects a file watcher to model indexing and an in-app notification. The automation editor
A paused automation draft connects a file watcher to model indexing and an in-app notification.

Configure, then inspect

Start from a template or build a graph. Configure each node’s inputs and conditions before enabling it. Use execution history and logs to understand what ran and where a step failed. Automations can be exported and imported as JSON; plugins can contribute more templates.

Automation overview ↗

16 / Make it yours

Plugins & ComfyUI

Add the capabilities your studio needs.

Extend your installation

Plugins can add pages, provider integrations, generation backends, sidebar widgets, quick actions, and automation templates. Examples include model downloading, external catalog providers, system monitoring, and GPU memory cleanup.

Open Administration → Plugins to inspect installed plugins, configure their settings, and enable or disable them. Their pages and controls appear only when enabled.

Installed plugin capabilities and configuration in the actual PotionUI administration interface. Plugin management
Installed plugin capabilities and configuration in the actual PotionUI administration interface.

Bring a ComfyUI workflow

  1. Connect ComfyUI

    Enable the ComfyUI Backend plugin and configure a reachable server in Administration → Backends.

  2. Export the runnable workflow

    In ComfyUI, use Export (API). Enable Dev mode options if that export is missing. The regular editor export is a different format.

  3. Build the preset

    Open Administration → Plugins → ComfyUI Backend → Import workflow. Paste or upload the JSON, design the form, select fields to record in History, and review requirements.

  4. Check dependencies and generate

    The importer can create a preset with missing-node warnings, but the workflow needs its required custom nodes and models installed before it can run.

Build your own tools

For a tailored generation tool, start with preset authoring. For capabilities around or beneath generation, use the plugin API. The app’s Developer reference includes form field types, template functions, icons, and live previews.

ComfyUI import guide ↗Plugin API ↗

The tools are here. The idea is yours.

Go make something.

Get PotionUI
Guide reviewed September 15, 2026Back to top ↑