Character Generation Inputs
The character's core gender identity. Does not determine romantic attraction.
Who the character is romantically or sexually attracted to (if applicable).
Controls whether adult/explicit themes may appear in the generated ideas.
The broad world and tone the character inhabits — the backdrop, not the character's personal story.
A specific creative seed — a character concept, occupation, or situation. More concrete than genre, more personal than trope. Leave blank for the AI to invent one.
A narrative pattern or character role — a classic storytelling dynamic you want the character to embody. Works across any genre.
The exact name of the person or fictional character to look up.
✨ Edit Reference Image (experimental)
💡 Sends the reference image plus your instruction to an image-to-image model, aiming to keep the likeness while changing what you ask for — a whole-image makeover before you even generate a card. Uses the "Image Edit Model" set in ⚙️ Settings → Image API. Costs API credits per attempt and may take a moment.
If description exists without an image, it is still used for character generation.
Placeholder for alignment
Placeholder for alignment
Upload a SillyTavern World Info JSON
Automatically creates lorebook entries for implied background characters and references them in the main card.
📡 View Raw Generation Stream (Debug)
Character
Name
Description
Personality
Scenario
First Message
Example Messages
💡 Example dialogue is automatically generated and included natively in your downloaded character card. Change the count below to regenerate them.
Post-History Instructions (Author's Note)
💡 Directly instructs the AI on how to write the next response. Injected at the bottom of the prompt to prevent it from being forgotten.
Creator's Notes ⚠️ not sent to LLM
💡 A brief, enticing synopsis of the character and scenario — written to sell the card to potential users. Use the character's actual name here (not {{char}}), as this is the field displayed on card-sharing websites.
Lorebook ⚡ keyword-triggered
💡 Create a companion Lorebook (World Info) for your character. Add entries for key people, places, or concepts to give the AI more context during roleplay.
Alternate Greetings ⚠️ not sent to LLM
💡 Add alternative starting messages for completely different or continued roleplay scenarios.
Image
⚙️ Image Settings & Controls
💡 Paste an image while the text box is focused to extract its style!
💡 Steer the AI prompt generator — specify form, setting, pose, or focus. Applies when the prompt is auto-generated.
✏️ Final Prompt Preview (Advanced)
💡 Tip: Edit this prompt before generating the final image. Maximum 10000 characters.
✨ Edit Current Image (experimental)
💡 Sends the current image plus your instruction to an image-to-image model, aiming to keep the character consistent while changing what you ask for. Uses the "Image Edit Model" set in ⚙️ Settings → Image API. Costs API credits per attempt and may take a moment.
Lower = closer to the original image but ignores your instruction more. Higher = follows the instruction more but the character's identity drifts. In testing, flux-1-dev img2img struggled to do both at once — worth experimenting yourself.
Local Library (IndexedDB)
Loading local library...
Saved Prompts
Saved Cards (Permanent)
History (Last 30)
🔗 SillyTavern Library
Click Refresh to load characters from SillyTavern.
⚙️ Generation Settings
LLM credentials (API URL, key, model) are shared with CardGen and managed via ⚙️ API Settings in the footer. Changes there are automatically applied here.
Maximum length of each generated chunk. Higher = longer passages but costs more and risks hitting the model's context limit.
256–4096 for most models. 2048 is a good default.
Controls how creative vs. predictable the output is. Lower = more focused and literal. Higher = more varied and imaginative.
0.6–0.8 for structured prose. 0.9–1.05 for creative flair.
Above ~1.1 the model can pick genuinely junk tokens. Because it then writes based on that token, one slip degrades everything after it — which is why a long chunk can start well and collapse into gibberish. Raise Top-P instead.
Restricts sampling to the most likely tokens making up this share of the probability mass, discarding the junk tail.
0.95 is a good default. This is what makes a higher Temperature safe. At 1 nothing is filtered.
How much of the story history to keep in memory when generating. Older segments beyond this limit get summarised automatically.
Match your model's context limit. 8000 suits most; use 16000+ for larger models.
Injected at the top of every generation as author-level instructions. Use this for tone, style, language, or world rules that apply to all stories. Leave blank to use default behaviour.
Which image model to use when generating scene illustrations. Defaults to whichever model is currently active in ⚙️ API Settings. Set this to pin a specific model for story images independently of character card generation.
🔊 Text-to-Speech
Read story chunks aloud using a local Coqui TTS engine.
Automatically generate the next chunk after the current one finishes narrating. Hands-free listening.
Your Stories
📝 Synopsis / Memo
Chats
No chats yet
Tap New to start your first roleplay
Chat
Create a new chat or select an existing one to begin roleplaying.
⚔️ Adventures
🖼️ Image Playground
Upload or paste any image, then crop, erase, or prompt-edit it — no character card needed.
✨ Edit Image
💡 Sends the image plus your instruction to an image-to-image model. Uses the "Image Edit Model" set in ⚙️ Settings → Image API — see the Model Guide below for recommendations. Costs API credits per attempt and may take a moment.
🧭 Model Guide — which model for what?
flux-2-pro-image-to-image/flux-2-max-image-to-image— strongest identity preservation across pose/outfit/style edits. If the source shows skin/nudity, these tend to auto-cover it rather than editing faithfully — expect the output to look more modest than the input.flux-kontext— good editor, but hard-blocks (silently returns a solid-black image) when the source shows enough skin, regardless of the instruction. Best reserved for fully-clothed sources.- Local Forge (flux-1-dev) — free, but a base checkpoint, not an edit model. In testing it barely followed outfit/pose instructions and lost facial likeness at higher denoising strengths. Good for experimenting, not a substitute for the above.