Velvet Study — Anatomy & Intimacy Consultation Parlor

AI Configuration

Name

Backend

API Key

Model

Start typing to search, or load the model list from OpenRouter.

Provider routing (advanced)

Preferred providers (comma-separated, in order)

Blocked providers (providers that reject requests are added automatically)

Temperature: 0.8

Notes (optional)

Automatic fallback to LM Studio

If OpenRouter fails, answer with a local LM Studio configuration instead

Fallback configuration

Used when a call to OpenRouter fails with a network error or timeout, an authorization or credit problem (401/402/403), a rate limit (429) or a server error (5xx). The same request is retried once. Your OpenRouter key is never sent to LM Studio. A local primary backend never falls back.

General

Show AI telemetry (time, tokens, cost, model) on messages, candidates, drafts and portraits

Quick actions default size (scenes without their own setting) Groups Options per group

Consultant Personas

The consultant voice and character. Create multiple personas and switch the active one.

Scenes & Customer Generation

A scene defines the consultation setting, candidate generation rules (hard facts, age range), and the chat moods. Create multiple scenes and switch the active one.

Edit Scene

Scene Name

Scene Description

System Prompt

Injection Prompt

Injected as a system message right before the assistant's reply. Wildcards: generic tracked stats like {{trust}}, {{trust_max}}, {{arousal}}, {{arousal_max}}, or any configured stat {{<stat_id>}}, {{<stat_id>_max}}, {{<stat_id>_min}}, {{current_<stat_id>}}.

Minimum Customer Age

Maximum Customer Age

Tracked Stats (e.g. Trust, Arousal)

Dynamic stats tracked during the scene. Configure name, description for extraction, value range, and per-turn delta limits.

Quick Actions Enabled

The suggestion grid under the chat. When disabled, no quick-action AI call is made and the bar is replaced by a short hint.

Groups (1–8) Options per group (1–8)

Group hints (seed titles, e.g. Say, Touch, Examine)

Extra instructions for the generator (optional)

Hard Facts (candidate generation)

"Roll" facts pick one or multiple values from a weighted pool. "AI" facts are composed by the model.

Chat Moods & Situations

A portrait for each mood is generated in the background at chat start. The extractor picks the active mood per message.

Configure local ComfyUI image generation. For each chat session, images for the configured moods/situations are queued in the background while the consultation begins.

Portraits and mood images are rendered on the server. For each chat session, images for the configured moods/situations are queued in the background while the consultation begins.

AI Horde API key

Portraits are rendered on AI Horde with this key. Get your own at aihorde.net/register; with a key your requests queue faster. It is never shown to users.

ComfyUI Server Address

Local ComfyUI endpoint where the Pony V7 workflow is submitted. Chat moods are now configured per-scene under the Scene & Customers tab.

Image Feedback Loop (experimental)

After each render the LLM looks at the image (vision model required), compares it with the character data and requested mood, and either accepts it or rejects it with a note and an improved prompt. This repeats until an image is accepted or the maximum is reached – then the best-scoring version is used.

Enable image feedback loop

Customer generation (candidate portrait) Mood generation (chat mood / situation images)

Max generations per image

1 = a single render, no retries. Every extra generation costs one ComfyUI render plus one vision LLM call.