MiniTavern Ultra is a mobile AI tavern for immersive AI roleplay. Start from one character card: everyday company, or a long story. It fits SillyTavern / SillyTaver cards and lorebook habits, and it is a familiar next step if you are bringing a nativetavern-style workflow onto your phone. Models are BYOK. Chats and characters stay on the device first.

1.0.7: write a description, add an optional reference photo, and let a model you already connected build the card. In the preview, keep the fields and world book you like. Revision notes rewrite only the rest. If the character comes from an existing work, the world book can come along and attach when you create the card.

【What's new in 1.0.7】
- Create with AI — a description, an optional reference image, then name, persona, and greeting
- Preview, then create — edit or keep fields and the world book; notes rewrite only what you did not keep
- Bring the setting — a world book from the source work attaches to that character
- The reference image shows the model who this is, and the first successful generation fills it in as the avatar

【Character cards】
- Import PNG from the gallery, JSON from files, or a link. Batch import is supported
- Mainstream character card v2 and v3. Import and start, or export a PNG to share
- Create with AI, or write the name, personality, scenario, and greeting yourself
- Community formats such as Chub.ai, JannyAI, and Character Hub
- Card lorebooks and regex can be copied into your library and still stay inside the card for export

【Lorebook, regex, presets, macros】
- Lorebooks: write your own, or import one you already have. Keywords or always-on entries; several books per chat
- Regex: write rules, or import SillyTavern regex, for what you see and for text sent to the model
- Prompt presets: build a layout, or import one. Change the writing style without rewriting the card. Temperature and other sampling stay in model settings, and a single chat can override them
- Macros and the macro helper: {{user}}, {{char}}, variables
- Personas: a default identity, or a different one for this chat only
- Interactive cards: HTML/CSS and variable status bars (off by default)

【Models. Keys stay on the device】
- Bring your own key for GPT, Claude, Gemini, OpenRouter, Qwen, DeepSeek, and Kimi
- On the same network, connect local models such as Ollama and LM Studio
- Card generation uses a connection you already set up. A reference image needs a model that can read pictures
- Each chat can use its own model, temperature, memory, persona, preset, lorebook, regex, and image setup

【Immersive chat】
- Streaming replies, regenerate, continue, and smart reply
- Immersive mode, visual novel mode, and a background of your own
- Photos you send can be read, and the people in them recognized. The character can reply with a picture that follows the scene
- 7 themes. Set bubble colors and fonts yourself
- Full backup of characters, chats, lorebooks, and regex. You choose whether keys are included
- Chat archives: export one chat, merge it into an existing thread, or import it as a new chat

【Why this AI tavern】
- Immersive AI roleplay: describe a character, then tune the theme, prompts, and how the chat looks
- Close to common SillyTavern / SillyTaver paths. Search nativetavern and you still land on mobile roleplay
- BYOK: you pick the model and the cost
FAQ
Q: Do I need my own API key?
A: Yes. BYOK. OpenRouter-style gateways and local models both work.

Q: How is this related to SillyTavern / SillyTaver?
A: An independent mobile AI tavern and roleplay client. It reads community PNG/JSON. It is not an official app.

Q: Which model creates the card?
A: One you already configured. A reference image needs a model that accepts images.

Q: Do prompt presets change temperature?
A: No. Presets hold the prompt structure. Sampling stays in model settings.

Immersive AI roleplay. Update to 1.0.7: describe a character, keep what you like, and rewrite the rest.