A prompt engineering extension for Stable Diffusion Forge Neo. Enhance text prompts with AI, interrogate images with vision models, and manage custom personas — all from a single dashboard.
Important
Compatibility: Built and tested exclusively on Forge Neo. Compatibility with other Forge forks is not guaranteed.
Type a rough idea, pick a persona, and let an LLM transform it into a detailed generation-ready prompt.
- Prose Enhancer: Expands keywords into rich, natural language descriptions.
- Tag Specialist: Converts ideas into Danbooru/e621 tag sequences.
- Results can be sent directly to txt2img or img2img.
Upload an image and scan it with a vision-capable model to extract prompts from existing artwork.
- Descriptive Caption: Outputs a detailed prose description of the image.
- Tag Analysis: Outputs a structured booru-style tag list.
- Results can be appended to or replaced in the enhancer input.
Create, edit, and delete custom system prompts that shape how the AI interprets your intent. Four personas are included out of the box.
Central configuration panel for managing provider connections.
- Multi-Provider Support: Switch between OpenRouter, Hugging Face, Ollama, and LM Studio backend engines.
- VRAM Model Unloader: Select and instantly unload active models from local VRAM for both Ollama and LM Studio backends.
- Request Timeout: Specify the maximum duration allowed for backend network requests.
- Max Output Tokens: Set an upper limit on the length of the generated model responses.
- Diagnostics: Test connections, save authentication keys, and configure custom endpoint URLs.
- Model Synchronization: Query and sync available models dynamically from the active provider.
- Open your Stable Diffusion WebUI (Forge Neo).
- Navigate to the Extensions tab → Install from URL.
- Paste:
https://github.com/SiliconeShojo/ScribeNEO.git - Click Install and restart the WebUI.
| Provider | How to get a key |
|---|---|
| OpenRouter | Create an account and generate a key at openrouter.ai/keys. Gives access to Gemini, Claude, GPT-4o, and more. |
| Hugging Face | Generate an Access Token (read permissions) at huggingface.co/settings/tokens. |
| Ollama | No key needed. Just run the Ollama server locally. |
| LM Studio | No key needed. Just run the LM Studio server locally. |
Click to expand changelog history
- LM Studio Integration: Added support for LM Studio as a first-class service provider backend for prompt enhancement (text LLMs) and image scanning (vision models).
- Asynchronous Request Layer: Migrated backend HTTP request operations from synchronous
requeststo asynchronoushttpxto support concurrent operations and network-level request cancellation. - Instant Request Cancellation: Introduced functional stop buttons (
🛑) that abort active LLM generation or image scanning requests on the network/socket level instantly. - Configurable Request Timeout: Added a global "Request Timeout" setting to allow users to customize maximum API wait times before a connection aborts.
- Configurable Max Output Tokens: Added a global "Max Output Tokens" slider to constrain generated output lengths for both text and vision models.
- Native Ollama API Integration: Migrated all Ollama calls from the OpenAI API compatibility shim to native
/api/chatand/api/generateendpoints. Special thanks to @hirorohi03 for suggesting this improvement! - Ollama VRAM Model Unloading: Added an "Unload" button and model selection manager to instantly release active models from GPU VRAM. Special thanks to @hirorohi03 for the recommendation!
- Configurable Keep-Alive: Added a customizable "Keep-Alive" numeric setting (in seconds) next to the model unloader to control how long models remain loaded in memory. Special thanks to @hirorohi03 for the recommendation!
- Bypass Vision Filters: Added a "Show all models (bypass vision filter)" checkbox in the Vision Toolset to query Ollama models manually even if they aren't on the predefined vision model list.
- Expanded Vision Models: Included broader default support for Ollama vision models (
qwen,gemma,vl,vision,minicpm,paligemma). - Schema-Based Persona Management: Removed legacy name prefixes (
Scribe:andVision:) from thepersonas.jsonfile, replacing them with a structured"type"attribute (scribevs.vision) for cleaner database separation. - Automatic Persona Database Migration: Added startup logic that automatically migrates legacy name prefix schemas to the new type-based database format.
- Persona Lab Category Toggle: Integrated a compact Category selector control inside the Persona Lab UI to separate Prompt Enhancer and Vision personas during editing.
- Ollama Vision Persona Support: Rewrote local Ollama image interrogation to use the native
/api/chatendpoint, adding full support for custom Vision system prompts and personas.
- Vision Query Error Handling: Added graceful HTTP 400 interception. If a text-only model is interrogated with an image, the system now returns a helpful capability suggestion instead of throwing an unhandled exception.
If ScribeNEO has enhanced your creative workflow, consider supporting its development!
Made with 🤍 for the AI Art Community.
