Let your SillyTavern characters see the world you're building. Picture Prompt gives your multimodal LLM a visual reference — how your character looks, how your persona looks, and a bunch of extra images for the scene.
Works with any vision-capable model behind a Chat Completion API: OpenRouter, Ollama, vLLM, KoboldCpp, Featherless, NanoGPT etc.
Anyone using SillyTavern with a vision model who wants richer, more consistent scene/character understanding. Especially useful for detailed world-building, consistent character appearance, or "show, don't tell" prompting.
- Avatar injection: The model sees character and persona avatars as pure images, each inside its description prompt block
- Extra images per persona: Upload reference images for your personas, they pass alongside the avatar
- Gallery images injection: Inject gallery images to the prompt, use them as extra images for character, items and scene references
- Lorebook entry images: Attach images to Lorebook entries, they pass alongside entry text; solid for NPCs, places and artifacts
- Fully local and offline: Images are stored in your browser and never leave your machine if you are using a local model
- Import/Export: Easily back up or migrate your images in just two clicks
- Custom labels: Attach a text hint to each image; it helps to steer the model and get the best results
- Token estimation: The extension shows you token cost overhead in settings before generation
- Per-source quality override: Different quality per source (avatars, extras, gallery, lorebook) to balance detail and token cost
- SillyTavern 1.14+
- A Chat Completion API (Text Completion is not supported)
- A vision-capable model (latest Gemma 4, Qwen, ChatGPT, Grok etc.)
- "Inline image media" enabled in AI Response settings
Install via the SillyTavern extension installer: Extensions → Install Extension:
https://github.com/RetroVioletRed/SillyTavern-PicturePromptThen enable in Extensions panel → Picture Prompt
Open the persona management panel and you'll see a new Extra Images section. Upload as many images as you want per persona. Your browser stores them and injects them into prompts alongside your persona avatar.
Enable/disable injection and cap the max image count per prompt from extension settings. Drag and drop the thumbnails to reorder.
SillyTavern's built-in gallery now has an "𖡡Inject Mode" switch. In inject mode you can select any image to pin it for injection.
Pinned images are per-character and you can label them.
As with Extra Images, extension settings control enable/disable and max image count.
Open any Lorebook entry, scroll to the bottom of the editor. There's a Lorebook Images section. Upload images per entry — they get injected whenever that entry activates.
Add a label if you want to explain the image to the model. Drag and drop the thumbnails to reorder.
The extension tracks active entries fully automatically.
When you hit send, Picture Prompt catches the prompt before the model sees it and injects your images as image_url content blocks into the relevant places across the prompt.
By default:
- Character avatar + gallery images land in the character's personality system message
- Persona avatar + extra images land in the persona description system message
- Lorebook images land in the world info system message, right after their entry's text
Per-source quality lets you set different detail levels per image type (Global | Low | Auto | High), defaulting to ST's global inline image quality
The settings panel estimates token cost live based on your current config. /pp-images shows exactly what'll get injected next. After each generation, a small indicator ("🖼 5 images · ~1,200 tokens") confirms what was sent.
"The images aren't showing up?"
Picture Prompt works silently in the background — it shows your images only to the model, never in the chat UI. To verify what's being injected:
- Check the token estimator in extension settings (it reflects actual injection count)
- Use
/pp-imagesto see a detailed preview of what will be sent on the next message - Look for the small "🖼 N images · ~X tokens" indicator that appears after each generation
"Does this work with my local model?"
If you use OpenAI-compatible Chat Completion, yes. Ollama, vLLM, KoboldCpp, llama.cpp — they all work.
"How many tokens does this cost me?"
Depends on image size and quality. Low quality is ~85 tokens per image. High quality can run 500-2000+ depending on resolution. The estimator in settings gives you the real numbers with your current config.
"How do i add extra images to my persona?"
You go to Persona Management Panel → pick your persona → Extra Images → Upload. Saved per persona, injected every request.
"How to use Lorebook entry images?"
Open any Lorebook entry editor, find the Lorebook Images section at the bottom. Upload an image, give it a label. When the entry activates during chat, the image comes along. Toggle individual images off anytime without deleting them.