Thanks! I think this is solved now - sorry about that. Please let me know if you still see issues.

Same issue here

i noticed that this morning, it's add that glyph or a zero length character. breaks a lot of things. i had to update the JSON parser i use to try and accommodate those additions. works, some of the time.

parseJSON: (input) => { if (!input) return null;

        // 1. OBJECT HANDLING (Already parsed)
        if (typeof input === 'object') {
            if (typeof input.text === 'string') return Utils.parseJSON(input.text);
            if (typeof input.generatedText === 'string') return Utils.parseJSON(input.generatedText);
            return input;
        }

        // 2. STRING HANDLING (The Toughened Shield)
        if (typeof input === 'string') {
            let clean = input.trim();
            
            // A. Strip Markdown Code Blocks
            clean = clean.replace(/```json/gi, "").replace(/```/g, "");
            
            // B. REINFORCED EXTRACTION
            // We find the FIRST occurrence of an opening bracket/brace
            // and the LAST occurrence of the matching closing bracket/brace.
            const firstObj = clean.indexOf('{');
            const firstArr = clean.indexOf('[');
            
            let firstOpen = -1;
            let lastClose = -1;

            if (firstObj !== -1 && (firstArr === -1 || firstObj < firstArr)) {
                firstOpen = firstObj;
                lastClose = clean.lastIndexOf('}');
            } else if (firstArr !== -1) {
                firstOpen = firstArr;
                lastClose = clean.lastIndexOf(']');
            }
            
            if (firstOpen !== -1 && lastClose !== -1 && lastClose > firstOpen) {
                // C. HARD TRUNCATION
                // We extract ONLY what is between the brackets.
                // This kills the trailing "〖" and any other AI rambling.
                clean = clean.substring(firstOpen, lastClose + 1);

                // D. Post-Extraction Cleanup
                clean = clean.replace(/,(\s*[}\]])/g, '$1'); // Remove trailing commas
                
                try {
                    return JSON.parse(clean);
                } catch (e) {
                    // LAST DITCH: If it still fails, it's usually unescaped newlines in a string
                    try {
                        // Replace literal newlines inside JSON strings with escaped \n
                        const escaped = clean.replace(/\n/g, "\\n").replace(/\r/g, "\\r");
                        return JSON.parse(escaped);
                    } catch (e2) {
                        console.warn("[Utils] JSON Parse Failed despite extraction:", e.message);
                    }
                }
            }
        }
        
        console.warn("[Utils] JSON Parse Total Failure. Input sample:", input.substring(0, 100));
        return null;
    }

@perchance@lemmy.world

I didn't want to start a new thread to point out the current errors. (NOTE: The texts are in German because I'm not good at English. I only translated the categories and notes with a translator. (IF ANYTHING IS UNCLEAR, PLEASE LET ME KNOW)). The names have been changed intentionally and most of the text has been shortened, but the errors should still be recognizable.

Other languages, strange symbols, and numbers:

• und stellt sie zusammen mit den Keksen auf ein330 Silbertablett.
• ist nichts weiter als einfacheਿਭBauern
• Du wirst dich dafür bei mir的数 (Translated: die Anzahl)
• Ich hab... ich hab Angst, dass sie mich holen kommt. Ihre Hände verkrampfen sich leicht in seinem Hemd所以才要逃跑 逃出这个可怕的地方 (Translated: "Deshalb muss ich fliehen – diesem schrecklichen Ort entkommen." Sounds more like a thought than dialogue or action.)
• Ihr Blick flackert zwischen Cloes Gesicht und dem Ort, wo Anon verschwunden ist sapphire hin und her.

Spelling/grammar errors:

• "schnaift"
• "Vielle-light..."
• "...dem Gosse!"
• "...mein Dank?"
• "Du wirst dich noch wünschen..."
• "Sie drückt sich gyumütlicher an..."
• Cloe erstarrt für einen Moment, als würde sie ihren Ahmed nicht trauen.

Memory/Lore Search Queries Used error:

• Past conflicts between Cloe and Anon regarding parenting Max/USMustermann

(Side note: 90% of this is in English. Occasionally, it is in German.)

In memory itself:

• aber sie besteht darauf, dass Anon nichts über sie weiß, und weigert sich, Cloe振り解く zu lassen.

Suddenly, a different topic:

• während er sich etwas bückt, um sie auf embodiments to a final pitch and tone that aligns with your preferences. Throughout this process, the assistant will periodically provide summary updates on the current state of the script and ask for confirmation before proceeding to the next phase. This ensures alignment with your vision and allows for adjustments before significant creative resources are invested. Would you like me to generate the first exploratory variations for the opening scene to demonstrate this approach?*

Other:

• [[response_end〗

• ISupported Request (no metaphor, satisfying resolution): Cloe doesn't give up, but her tone softens—not pleading, but earnest. She's standing her ground, but offering an olive branch. Wenn du mir eine Chance gibst...

I delete English memories completely. Otherwise, the AI might suddenly start speaking English instead of German. Either they are created again in German, or I'm out of luck. I recently noticed that when the first memories are created, only the first three entries are in German. After that, they are in English. Then it switches randomly: German, English, English, German. Sometimes there is one in German, then suddenly three in English, and then another one in German (in one go). I don't know if this could be important, I'm just mentioning it.

I can continue testing and documenting further errors. And I hope it's okay if I post it here. Otherwise, I can open a separate post.