Replace the tag based image selector with a caption based one

In most places. Images are selected now by passing the LLM a list of
photo captions and the words spoken, and asking it which image best fits
the sentiment.

4 files changed, 40 insertions, 36 deletions

diff --git a/scripts/toy.groovy b/scripts/toy.groovy
index 8b21069..e21f2c8 100644
--- a/scripts/toy.groovy
+++ b/scripts/toy.groovy
@@ -50,7 +50,11 @@ return new Object() {
 
 	final getDomme = { domme, set = null ->
 		def d = parseJsonFile("images/toy/$domme/person.json");
+		if (new File("images/toy/$domme/$set/set.json").exists()) {
+			d += parseJsonFile("images/toy/$domme/$set/set.json");
+		}
 		d.id = domme;
+		d.set = set;
 		return d;
 	};
 	final loadDomme = { domme, set = null ->
@@ -241,8 +245,8 @@ return new Object() {
 		return null;
 	};
 	final llmPresent = { text ->
+		final image = showImageForText(text);
 		show(text);
-		def image = showImage(callAPI_imageMatch(text));
 		return [
 			message: text,
 			image: image
@@ -430,6 +434,7 @@ Speak in the first person only.
 Forbidden from using quotes.
 You are permanently in role-play mode as ${DOMME.title} ${DOMME.fullName}.
 ${DOMME.prompt}
+${DOMME.promptSet?:""}
 ${loadString("intro.name")} is your slave.
 ${extra.join('\n')}
 ${eachModule("systemPrompt").findAll().join('\n')};
@@ -446,24 +451,6 @@ ${eachModule("userIntroPrompt").findAll().join('\n')};
 		return messages;
 	}
 
-	final callAPI_imageMatch = { message ->
-		def tags = [DRESSED, TITS, PUSSY, LINGERIE, TEASE, SIT, BOOTS, KNEEL, STOOD, MEAN, CROP, ASS, SQUAT, SSSH];
-		def request = chatInitialMessages();
-		request += [role: "assistant", content: message]
-		request += [role: "user", content: """
-Write only a JSON object.
-Complete the following JSON object, scoring each keyword between -10 and 10 against the previous messsage:
-{
-	${ tags.collect { tag -> "\"$tag\": number" }.join(',\n') }
-}
-				""" ];
-		return parseJsonString(callAPI("chat/completions", [
-			model: loadString("toy.chatModel"),
-			stop: "<",
-			messages: request
-		])?.choices?[0]?.message?.content);
-	}
-
 	String main()
 	{
 		// GO!
@@ -472,7 +459,7 @@ Complete the following JSON object, scoring each keyword between -10 and 10 agai
 		setDefault("toy.chatModel", "Lewdiculous/Erosumika-7B-v3-0.2-GGUF-IQ-Imatrix:IMAT");
 
 		OWNER = loadString("toy.owner");
-		loadDomme(OWNER);
+		loadDomme(OWNER, loadString("toy.owner.outfit"));
 		modules = loadModules(this);
 		execEvents(false);
 		eachModule("setup");
diff --git a/scripts/toy/imagery.groovy b/scripts/toy/imagery.groovy
index af4e9e7..18113cc 100644
--- a/scripts/toy/imagery.groovy
+++ b/scripts/toy/imagery.groovy
@@ -8,22 +8,12 @@
 					domme: d.getName(),
 					sets: d.listFiles()
 						.findAll { s -> s.isDirectory(); }
-						.collect { s -> [ sub: s, f: new File(s.getPath() + "/tags")] }
+						.collect { s -> [ sub: s, f: new File(s.getPath() + "/set.json")] }
 						.findAll { s -> s.f.canRead() }
 						.collectEntries { s -> [
 							s.sub.getName(), [
 								set: s.sub.getName(),
-								images: s.f
-									.readLines()
-									.collect { l ->
-										def fields = l.split(":");
-										return [
-											image: fields[0],
-											tags: fields[1]
-												.split(",")
-												.findAll { t -> !t.isEmpty() }
-										];
-									}
+								images: new groovy.json.JsonSlurper().parse(s.f).captions
 							]
 						]}
 					]
@@ -47,6 +37,34 @@
 		return matches[getRandom(matches.size())].image;
 	};
 
+	toy.metaClass.showImageForText { text ->
+		final captionList = DOMME.captions.collect{ id, caption -> "* ${id} : ${caption}" }.join("\n");
+		final request = [
+			model: loadString("toy.chatModel"),
+			grammar: "root ::= ( ${DOMME.captions.keySet().collect{ id -> "\"${id}\"" }.join(" | ")} )",
+			messages: [
+				[role: "system", content: """
+You are ${DOMME.title} ${DOMME.fullName}.
+${DOMME.prompt}
+${DOMME.promptSet?:""}
+Given a list of images with captions and some spoken words, you select the image which best captures the sentiments of the words.
+The available images are:
+${captionList}"""
+				],
+				[role: "user", content: "Write only an image Id from the list as an unformatted string, select an image for the text: ${text}"]
+			]
+		];
+		while (true) {
+			final imageName = callAPI("chat/completions", request)?.choices?[0]?.message?.content;
+			if (DOMME.captions[imageName]) {
+				setImage("toy/${DOMME.id}/${DOMME.set}/${imageName}.jpg");
+				return imageName;
+			}
+			showPopup("Given:\n${requestMessages}\n\nGot: ${imageName}");
+		}
+		return null;
+	}
+
 	toy.metaClass.showImage { spec ->
 		def outfit = loadString("toy.owner.outfit");
 		def image = selectImage(DOMME.id, outfit, spec);
diff --git a/scripts/toy/intro.groovy b/scripts/toy/intro.groovy
index 3e719cc..bfe8b68 100644
--- a/scripts/toy/intro.groovy
+++ b/scripts/toy/intro.groovy
@@ -10,10 +10,8 @@
 			showButton("OK");
 
 			// Welcome
-			def defaultImageSpec = [ "${DRESSED}": 10, "${TEASE}": 5 ];
-			dress(defaultImageSpec);
-			showImage(defaultImageSpec);
-			show("...");
+			dress([]);
+			llmPresent("...");
 			def messages = chatInitialMessages();
 
 			messages += [role: "user", content: "I am ${loadString("intro.name")}. Please briefly introduce yourself."];
diff --git a/scripts/toy/llmSession.groovy b/scripts/toy/llmSession.groovy
index f140d75..2689cdf 100644
--- a/scripts/toy/llmSession.groovy
+++ b/scripts/toy/llmSession.groovy
@@ -27,6 +27,7 @@
 			playSchedule();
 			return;
 		}
+		dress([]);
 		def messages = chatInitialMessages("You want to play with ${loadString("intro.name")}. You can dismiss ${loadString("intro.name")} when you wish by writing \"DISMISSED.\" only.");
 		if (summon(messages)) {
 			while (!messages.last().content.toUpperCase().startsWith("DISMISSED") && !messages.last().content.toUpperCase().contains("YOU ARE DISMISSED")) {