fix: improve embedding autocomplete detection and triggers
- Use ComfyUI's native api.getEmbeddings() for proper embedding detection - Add dedicated /kikotools/autocomplete/loras endpoint for LoRA files - Improve trigger detection for "embedding:" and "<lora:" patterns - Context-aware suggestions based on trigger type - Better insertion logic that maintains correct syntax - Sort suggestions by relevance (exact match, starts with, alphabetical) - Fix character matching patterns to include underscores and hyphens
This commit is contained in:
Vito Sansevero
+14
@@ -21,6 +21,8 @@ def setup_autocomplete_api():
|
||||
try:
|
||||
from aiohttp import web
|
||||
from server import PromptServer
|
||||
import folder_paths
|
||||
import os
|
||||
from kikotools.tools.embedding_autocomplete.node import KikoEmbeddingAutocompleteAPI
|
||||
|
||||
@PromptServer.instance.routes.get("/kikotools/autocomplete/suggestions")
|
||||
@@ -41,6 +43,18 @@ def setup_autocomplete_api():
|
||||
)
|
||||
|
||||
return web.json_response(suggestions)
|
||||
|
||||
@PromptServer.instance.routes.get("/kikotools/autocomplete/loras")
|
||||
async def get_loras(request):
|
||||
"""API endpoint for getting list of LoRAs."""
|
||||
try:
|
||||
lora_files = folder_paths.get_filename_list("loras")
|
||||
# Return just the names without extensions
|
||||
loras = [os.path.splitext(f)[0] for f in lora_files]
|
||||
return web.json_response(loras)
|
||||
except Exception as e:
|
||||
print(f"[KikoTools] Error getting LoRAs: {e}")
|
||||
return web.json_response([])
|
||||
|
||||
except ImportError:
|
||||
pass # Server not available when not in ComfyUI
|
||||
|
||||
@@ -1,4 +1,5 @@
|
||||
import { app } from "../../scripts/app.js";
|
||||
import { api } from "../../scripts/api.js";
|
||||
import { ComfyWidgets } from "../../scripts/widgets.js";
|
||||
|
||||
// KikoEmbeddingAutocomplete - Provides autocomplete for embeddings and LoRAs in text widgets
|
||||
@@ -45,33 +46,38 @@ class KikoEmbeddingAutocomplete {
|
||||
|
||||
async fetchResources() {
|
||||
try {
|
||||
// Fetch embeddings
|
||||
// Fetch embeddings using ComfyUI's API
|
||||
if (this.settings.showEmbeddings) {
|
||||
const embResponse = await fetch("/embeddings");
|
||||
if (embResponse.ok) {
|
||||
const data = await embResponse.json();
|
||||
this.embeddings = Object.keys(data || {}).map(name => ({
|
||||
name: name,
|
||||
type: "embedding",
|
||||
display: `embedding:${name}`
|
||||
}));
|
||||
}
|
||||
const embeddings = await api.getEmbeddings();
|
||||
this.embeddings = embeddings.map(name => ({
|
||||
name: name,
|
||||
type: "embedding",
|
||||
display: `embedding:${name}`,
|
||||
value: `embedding:${name}`
|
||||
}));
|
||||
console.log(`[KikoTools] Loaded ${this.embeddings.length} embeddings`);
|
||||
}
|
||||
|
||||
// Fetch LoRAs
|
||||
// Fetch LoRAs using our custom endpoint
|
||||
if (this.settings.showLoras) {
|
||||
const loraResponse = await fetch("/models/loras");
|
||||
if (loraResponse.ok) {
|
||||
const data = await loraResponse.json();
|
||||
this.loras = (data || []).map(name => ({
|
||||
name: name,
|
||||
type: "lora",
|
||||
display: `<lora:${name}:1.0>`
|
||||
}));
|
||||
try {
|
||||
const response = await fetch("/kikotools/autocomplete/loras");
|
||||
if (response.ok) {
|
||||
const loras = await response.json();
|
||||
this.loras = loras.map(name => ({
|
||||
name: name,
|
||||
type: "lora",
|
||||
display: `<lora:${name}:1.0>`,
|
||||
value: `<lora:${name}:1.0>`
|
||||
}));
|
||||
console.log(`[KikoTools] Loaded ${this.loras.length} LoRAs`);
|
||||
}
|
||||
} catch (e) {
|
||||
console.log("[KikoTools] Could not load LoRAs, endpoint may not be available");
|
||||
}
|
||||
}
|
||||
|
||||
console.log(`[KikoTools] Loaded ${this.embeddings.length} embeddings and ${this.loras.length} LoRAs`);
|
||||
console.log(`[KikoTools] Total: ${this.embeddings.length} embeddings and ${this.loras.length} LoRAs`);
|
||||
} catch (error) {
|
||||
console.error("[KikoTools] Error fetching resources:", error);
|
||||
}
|
||||
@@ -173,11 +179,38 @@ class KikoEmbeddingAutocomplete {
|
||||
|
||||
const insertSuggestion = (suggestion) => {
|
||||
const text = textarea.value;
|
||||
const before = text.substring(0, prefixStart);
|
||||
const after = text.substring(textarea.selectionStart);
|
||||
const cursor = textarea.selectionStart;
|
||||
const textBefore = text.substring(0, cursor);
|
||||
|
||||
textarea.value = before + suggestion.display + after;
|
||||
textarea.selectionStart = textarea.selectionEnd = prefixStart + suggestion.display.length;
|
||||
// Determine what to insert based on context
|
||||
let insertText = "";
|
||||
let replaceLength = currentPrefix.length;
|
||||
|
||||
// Check if we're completing an embedding
|
||||
if (textBefore.match(/embedding:([a-zA-Z0-9_-]*)$/)) {
|
||||
insertText = suggestion.name; // Just the name, not the full "embedding:name"
|
||||
}
|
||||
// Check if we're completing a lora
|
||||
else if (textBefore.match(/<lora:([a-zA-Z0-9_-]*)$/)) {
|
||||
insertText = suggestion.name + ":1.0>"; // Complete the lora syntax
|
||||
}
|
||||
// General insertion
|
||||
else {
|
||||
insertText = suggestion.value || suggestion.display;
|
||||
// For general insertion, replace the whole word
|
||||
let wordStart = cursor - currentPrefix.length;
|
||||
while (wordStart > 0 && /[a-zA-Z0-9_-]/.test(text[wordStart - 1])) {
|
||||
wordStart--;
|
||||
}
|
||||
prefixStart = wordStart;
|
||||
replaceLength = cursor - wordStart;
|
||||
}
|
||||
|
||||
const before = text.substring(0, prefixStart);
|
||||
const after = text.substring(prefixStart + replaceLength);
|
||||
|
||||
textarea.value = before + insertText + after;
|
||||
textarea.selectionStart = textarea.selectionEnd = prefixStart + insertText.length;
|
||||
|
||||
hideSuggestions();
|
||||
|
||||
@@ -189,42 +222,89 @@ class KikoEmbeddingAutocomplete {
|
||||
const findPrefix = () => {
|
||||
const text = textarea.value;
|
||||
const cursor = textarea.selectionStart;
|
||||
const textBefore = text.substring(0, cursor);
|
||||
|
||||
// Look for embedding or lora prefix
|
||||
// Check for embedding: trigger
|
||||
const embeddingMatch = textBefore.match(/embedding:([a-zA-Z0-9_-]*)$/);
|
||||
if (embeddingMatch) {
|
||||
currentPrefix = embeddingMatch[1].toLowerCase();
|
||||
prefixStart = cursor - embeddingMatch[1].length;
|
||||
return "embedding";
|
||||
}
|
||||
|
||||
// Check for <lora: trigger
|
||||
const loraMatch = textBefore.match(/<lora:([a-zA-Z0-9_-]*)$/);
|
||||
if (loraMatch) {
|
||||
currentPrefix = loraMatch[1].toLowerCase();
|
||||
prefixStart = cursor - loraMatch[1].length;
|
||||
return "lora";
|
||||
}
|
||||
|
||||
// Check for general word to match (after minimum chars)
|
||||
let start = cursor - 1;
|
||||
while (start >= 0 && text[start] !== " " && text[start] !== "\n" && text[start] !== ",") {
|
||||
while (start >= 0 && /[a-zA-Z0-9_-]/.test(text[start])) {
|
||||
start--;
|
||||
}
|
||||
start++;
|
||||
|
||||
const prefix = text.substring(start, cursor).toLowerCase();
|
||||
const prefix = text.substring(start, cursor);
|
||||
|
||||
// Check if we should show suggestions
|
||||
// Check if we should show suggestions for general matching
|
||||
if (prefix.length >= this.settings.triggerChars) {
|
||||
currentPrefix = prefix;
|
||||
currentPrefix = prefix.toLowerCase();
|
||||
prefixStart = start;
|
||||
return true;
|
||||
return "general";
|
||||
}
|
||||
|
||||
return false;
|
||||
};
|
||||
|
||||
const getSuggestions = (prefix) => {
|
||||
const getSuggestions = (prefix, type) => {
|
||||
const suggestions = [];
|
||||
|
||||
if (this.settings.showEmbeddings) {
|
||||
// If type is embedding, only show embeddings
|
||||
if (type === "embedding" && this.settings.showEmbeddings) {
|
||||
suggestions.push(...this.embeddings.filter(e =>
|
||||
e.name.toLowerCase().includes(prefix) ||
|
||||
e.display.toLowerCase().includes(prefix)
|
||||
prefix === "" || e.name.toLowerCase().includes(prefix)
|
||||
));
|
||||
}
|
||||
// If type is lora, only show loras
|
||||
else if (type === "lora" && this.settings.showLoras) {
|
||||
suggestions.push(...this.loras.filter(l =>
|
||||
prefix === "" || l.name.toLowerCase().includes(prefix)
|
||||
));
|
||||
}
|
||||
// For general matching, show both
|
||||
else if (type === "general") {
|
||||
if (this.settings.showEmbeddings) {
|
||||
suggestions.push(...this.embeddings.filter(e =>
|
||||
e.name.toLowerCase().includes(prefix)
|
||||
));
|
||||
}
|
||||
|
||||
if (this.settings.showLoras) {
|
||||
suggestions.push(...this.loras.filter(l =>
|
||||
l.name.toLowerCase().includes(prefix)
|
||||
));
|
||||
}
|
||||
}
|
||||
|
||||
if (this.settings.showLoras) {
|
||||
suggestions.push(...this.loras.filter(l =>
|
||||
l.name.toLowerCase().includes(prefix) ||
|
||||
l.display.toLowerCase().includes(prefix)
|
||||
));
|
||||
}
|
||||
// Sort suggestions - prioritize exact matches and starts-with matches
|
||||
suggestions.sort((a, b) => {
|
||||
const aName = a.name.toLowerCase();
|
||||
const bName = b.name.toLowerCase();
|
||||
|
||||
// Exact match
|
||||
if (aName === prefix) return -1;
|
||||
if (bName === prefix) return 1;
|
||||
|
||||
// Starts with
|
||||
if (aName.startsWith(prefix) && !bName.startsWith(prefix)) return -1;
|
||||
if (!aName.startsWith(prefix) && bName.startsWith(prefix)) return 1;
|
||||
|
||||
// Alphabetical
|
||||
return aName.localeCompare(bName);
|
||||
});
|
||||
|
||||
return suggestions.slice(0, this.settings.maxSuggestions);
|
||||
};
|
||||
@@ -236,8 +316,9 @@ class KikoEmbeddingAutocomplete {
|
||||
return;
|
||||
}
|
||||
|
||||
if (findPrefix()) {
|
||||
currentSuggestions = getSuggestions(currentPrefix);
|
||||
const triggerType = findPrefix();
|
||||
if (triggerType) {
|
||||
currentSuggestions = getSuggestions(currentPrefix, triggerType);
|
||||
|
||||
if (currentSuggestions.length > 0) {
|
||||
// Get cursor position for suggestion placement
|
||||
|
||||
Reference in New Issue
Block a user