# DesignCat Studio - Web App Agent Guide

Purpose: concise operating instructions for AI agents using Studio. This is a UI workflow guide, not an API payload reference.

Runtime guide URL: /app/studio/llms.txt
Studio routes: /app/studio/create_image, /app/studio/image_edit, /app/studio/create_video, /app/studio/video_edit, /app/studio/voice_design, /app/studio/create_voice, /app/studio/sound_effects, /app/studio/create_music

## Core mental model

Studio has eight creation surfaces: Image, Image Edit, Video, Video Edit, Voice Design, Text to Speech, Sound Effects, and Music. Voice Design includes Design, Remix, and Instant Voice Clone modes. Voice collections live in the right Library rail.

Sign-in is required to generate or use private libraries. Image creation and image editing do not require a project; video, speech, sound effects, and music generation use the project selected in the global header. Voice creation and saving use separate voice-resource workflows. Generate is a paid action.

## End-to-end workflow

1. Select the intended project when generating video or audio media.
2. Choose a capability from the left rail.
3. Choose a model and set only its visible controls. Image/video model, aspect ratio, resolution, and duration controls live in the composer; their right inspector is for attachments.
4. Write the prompt. In Image Edit and Video Edit, add the required source on the central canvas; add optional references through Attachments.
5. Read the current credit charge and disabled Generate reason. The visible form is authoritative for model limits.
6. Click Generate only when paid output is intended. Generation starts after uploads and validation, without a countdown. Repeated deliberate clicks can create separate paid requests.
7. Inspect progress and results through Generations. Ready generated and uploaded media also appears in the account-wide Library.

## Shared workspace

Library and Attachments:

- The layout runs from the left capability rail through the central workspace and optional Attachments inspector to Library on the far right.
- Library starts collapsed in a new browser tab and remembers its state for that tab. Use the control beside its title or click the narrow Library rail to expand it. Attachments collapses independently and preserves its inputs.
- Library source filters are All, Creations, Uploaded, and Exports. Media filters include Image, Video, Drafts, Music, Narration, and Sound FX; My Voice opens saved voices. Drafts shows video previews generated in draft mode; All and Video also include them. Use search, sorting, and scrolling to find more items.
- Library is account-wide, not limited to the selected project. Preview controls open the same file details as Motion Creations: media navigation, editable name/note, prompt, settings, source inputs, file information, download, and confirmed deletion. Draft details show the completion expiry when known; Unknown does not mean the draft never expires.
- Drag an image to Image Edit or a video to Video Edit to use it as the canvas source. Desktop files can also be dropped onto edit sources.
- Attachment pickers offer Character, Assets, Creations, Uploaded, and Upload file. Character and Assets open kits; files in active kits remain in those kit choices.

Prompts and inputs:

- Reference inputs receive mention labels such as `#Image1`, `#Video1`, `#Audio1`, or `#Element1`. Type `#` to insert available mentions. Missing mentions disable generation.
- Source Image, First Frame, Last Frame, and Source Video are source roles, not prompt mentions. Image Edit's plural reference-image input is an exception: the canvas image remains `#Image1`, with inspector images beginning at `#Image2`.
- Source is the media being transformed or animated. References steer appearance, identity, motion, sound, or composition.
- An Element represents one coherent subject with 2-4 images. Its additional angles do not consume separate slots in the shared reference budget. Some models also accept an element video.
- Source and Reference headers show used/maximum counts. At capacity, empty add controls disappear; existing inputs can still be inspected or removed.
- Switching models or source/reference routes can change fields, requirements, and limits. Audio references need visual references where the selected capability requires them.

Generations:

- The shared Generations control lists requests from the previous 72 hours with All and This tool scopes and pagination.
- Selecting a request opens its owning surface with `?generation={request_id}` and restores its submitted parameters and pending, ready, or failed result. Voice requests reopen in the correct mode.
- Back to draft restores the surface's previous in-memory prompt, attachments, and controls. Changing surfaces preserves their in-memory drafts; do not assume unsent drafts survive a full reload.

## Image and video

- Image supports prompt-only creation and model-dependent image references or Elements.
- Image Edit requires an image. Its canvas shows `#Image1` for plural image references or `Source Image` for a singular source. Before / After switches between source and generated output.
- Video supports model-dependent prompt-only generation, First Frame and optional Last Frame, reference media, or source-video transformation. Source/Reference and frame/video header actions switch input routes when offered.
- With Seedance 2.5 selected, DRAFT at the top right toggles draft mode and changes color. Clicking the toggle does not generate. Generate submits a paid 480p preview while it is on; turn it off to use the chosen resolution. The credit estimate follows the mode. Drafts appear in Generations and Library; no action to finalize a draft is exposed yet.
- Multi-prompt is available only for supporting Studio capabilities. It replaces the single prompt and duration with timed prompt items; obey the visible item and total-duration limits. Switching to an unsupported route removes the control.
- Video Edit requires a Source Video on the canvas. Describe a transformation. The output is a new creation; Before / After switches between source and result.
- Studio video has no Sound Design mixer, playback-mute setting, or generation-audio toggle. Audio references are separate from output-audio policy.
- Visible image/video determines the canvas's native aspect ratio without cropping. A blank canvas follows the selected ratio.
- Replacing an edit source while a result is displayed requires confirmation. It clears only the displayed result; the creation stays in Library. Wait for an in-progress generation before replacing its source.

## Voice Design, Remix, and Clone

- Voice Design: describe the voice, use automatic or custom preview text, and adjust the visible voice settings. An optional audio reference can be uploaded or recorded when supported. Generate previews, listen, then explicitly save the chosen voice with its identity metadata.
- Remix: choose a compatible custom voice from My Voices, describe the change, and choose Prompt Strength. Generate candidate previews and save the desired result. Community/library voices cannot be remixed.
- Instant Voice Clone: switch mode, upload or record samples, enter the voice name and optional metadata, and confirm rights or consent. Each sample must be at least five seconds and the total at least ten seconds; obey the displayed file limits. Remove noise starts enabled. Create voice saves the ready voice directly to My Voice, without a second candidate Save step.
- Voice Library supports browsing, previewing, and adding community voices. My Voice provides saved voices and actions such as Speak, Remix where supported, and Remove. Speak opens Text to Speech with that voice selected.
- Unavailable saved voices cannot be used for new speech. Removing a saved library voice removes the user's save.

## Text to Speech, Sound Effects, and Music

Text to Speech:

- Choose an actual saved/library voice and write the speech text; a gender selector does not choose the voice.
- Eleven v3 supports Add speaker for dialogue, up to ten turns. Additional voices must be compatible with the first voice's provider. Mixed-provider requests cannot generate.
- Follow the visible character limit, model settings, and output formats. Current limits are 5,000 characters for single-speaker speech and 2,000 for dialogue.

Sound Effects:

- Describe the sound, set Looping/Off, Prompt influence, and Duration. The current model is Eleven Text to Sound v2, with a 450-character prompt limit and 0.5-22 second duration.
- One Generate action produces four alternatives grouped in one Generations item. Preview or download the alternatives.

Music:

- Choose the model and describe the music. Controls vary: Music v2 offers Lyrics/Instruments, duration, and output; Stable Audio offers duration; Lyria models show fixed or automatic duration.
- Only use controls shown for the selected model. Generated audio appears in the canvas, Generations, and Library.

## Safe operation

- Confirm the selected project and displayed charge before generating media. Do not generate merely to inspect the UI.
- Uploads, saved voices, and generated media are durable account resources, even when used only as attachments or previews.
- Leaving a surface does not cancel accepted work. Local cancellation, where offered, applies before dispatch and does not imply provider cancellation.
- Do not confirm project, voice, or media deletion unless intended. Follow the visible model fields, limits, disabled reason, and credit balance when they differ from this guide.