Set Context and Ask LLM fashioning

Fashioning is the way you can tailor (fashion) the prompts created in preview panel, None option will skip any fashioning

Set Context

Set a context fashioning simply means that when generating images from preview panel every prompt will be prefixed by your chosen context. That way you can set a style, mood or atmosphere in much more elaborate way than just modifiers. Of course you can use both context and modifiers where context will play principal role fashioning your prompts but modifiers will give the flexibility to vary them in a scan. That way you can "fine tune" the dominant function of the context with variability of modifiers.

A combo box with example contexts (10 so far) allows you to pick your context (prefix).

  • If you like to modify existing context, just select it, edit it and click on update (top on the left) button.
  • If you like to add your original context, write it in the edit box and click on big plus button.
  • Minus button will remove currently selected context from the list.

If you would like to include a comment (as title or description) the context, place # in front of that line and the comment text will be saved but ignored when context is used.
The context fashioning works well but you need to balance the lengths of the context text and the cue text. In order not to overwhelm the prompt with the context, rule of thumb is: on average the length of the context should be less than the cue length.  

 Operations over the list of contexts and the list of questions for LLM as update, add and remove are exactly the same for both.

Ask LLM

Much more flexible (and sophisticated) way to tailor your prompts is by asking LLM to rewrite your prompts in a particular way. The idea is very simple: generate a prompt list (with or without modifiers) in Scan Preview panel. Then ask LLM to tailor your prompts (only the cue part) in a distinctive way. You may see from examples different ways to instruct LLM how to rewrite your cues. Let's borrow some of the accumulated skills of LLM into your creative process. 

First you need to launch (Run LLM server) LM Studio. Then you can ask LLM to fashion selected prompt (green button) or all the checked ones (Auto Ask checkbox). Show both checkbox refers to showing both: the original cue and the fashioned one. Temperature control allows you to introduce (or not) some variability/randomness. The max number of tokens suggests the maximum length of the LLM response.

Before start using LLM you need to install and setup one. At the present only LM Studio is available as local LLM server (later Ollama).

  • Download and install LM Studio from here . After installation, you don't need to switch to Developer mode, only if you need more control and information about current state of LM Studio. When you try to launch LM Studio from configuration or from preview panel, Scripthea will try to start LM Studio in server mode using command line argument --server. In some versions of LM Studio that is not enough to run the server, you need to go to the server page of LM Studio and run the server from there.
  • Download a LLM model from within LM Studio. For a start I would recommend  "ibm/granite-4-h-tiny", it's a relatively small one (5.7Gb) which means that it should run on 16Gb RAM machine and very good (quick and adequate) for its size.
  • Set a LM Studio executable location in Scripthea configuration dialog (gear button top left). After you set the path to executable, launch it. That will give you indication that the LM Studio is there and its API is open for business.
  • Launching LM Studio will set the combo-box of LM Studio model with all available models. Choose one and load it.
  • Setting a system prompt (context) will instruct the model to act (as in a role) from a specific perspective (e.g. you are an assistant helping with text-to-image generation) 
  • At the end you may try to test it by asking random question and check the response.
  • Occasionally some models will give you an additional information in their response you don't really need. First, try to instruct LLM (system prompt) not to do that. If that does not work go to Config folder of Scripthea and find file "omit-llm.txt" which is the list of all pieces of text you need Scripthea to omit when reading the LLM response.