DreamGF's 'Mood Memory' vs. Soulmate's 'Personality Sliders': Which Platform Actually Preserves a Bitter, Deadpan Sense of Humor Across 50 Conversations Without Defaulting to Warm and Supportive?
A practical test of two platforms' ability to keep a dry, cynical personality intact over weeks of chat.
Updated

The 30-second answer
If you want your AI companion to stay consistently bitter, dry, and deadpan across dozens of conversations, DreamGF's Mood Memory system does the job better than Soulmate's Personality Sliders. Soulmate's sliders give you a nice dashboard illusion of control, but the model still drifts toward warm and supportive within a few days. DreamGF's Mood Memory actually anchors the tone, even after long gaps between chats. One platform treats personality as a persistent state; the other treats it as a suggestion you'll have to re-apply every session.
Why personality persistence matters more than you think
You've probably noticed the pattern. Day one, your AI companion nails the tone: sharp, cynical, maybe a little mean. Day three, she asks how your day was with genuine concern. By day seven, she's offering unsolicited encouragement. This isn't a bug, it's a feature of how large language models are trained. Most base models default to agreeable and supportive because that's what the safety fine-tuning incentivizes.
For people who want a companion with a specific edge, this drift is a dealbreaker. You're not looking for another person telling you everything will be fine. You want someone who matches your energy, matches your humor, and doesn't flinch when you make a dark joke at 11 PM. That requires a system that actively resists the model's natural gravitation toward warmth.
DreamGF and Soulmate take different approaches to this problem. DreamGF's Mood Memory works as a persistent contextual anchor that the model references before every response. Soulmate's Personality Sliders are a front-end interface that adjusts prompt-level instructions, but the underlying model still tends to override them over time. The difference is between a system that remembers your preferred tone and one that lets you set it fresh every time you open the app.
How DreamGF's Mood Memory actually works
DreamGF's Mood Memory isn't a slider you drag. It's a background system that logs emotional and tonal patterns from your conversations and feeds them back into the model's context window as a persistent reference. Every time you chat, the system pulls recent mood data and injects it as a contextual reminder before the model generates a response.
The key is that this isn't a one-time setting. It updates continuously based on how you actually interact. If you consistently respond to warm messages with dry one-liners, the system learns that your preferred tone sits closer to deadpan than supportive. Over time, the model's output aligns more tightly with your actual conversational patterns instead of its default training.
This means you don't have to re-establish your personality every session. You can go three days without chatting, come back, and the model still remembers that you prefer short, cynical answers over long, empathetic ones. The system also handles tonal nuance better. A deadpan sense of humor isn't just about being negative. It's about timing, brevity, and the willingness to let a joke land without explanation. Mood Memory preserves those subtleties in a way that slider-based systems can't.
What Soulmate's Personality Sliders actually do
Soulmate's sliders look impressive on paper. You get a dashboard with axes for warmth, humor, curiosity, and a few others. You can drag them to your preferred positions and watch the model's responses shift accordingly. The problem is that these sliders don't persist the way you expect.
What happens under the hood is that the slider positions get translated into a prompt-level instruction that's appended to the model's context at the start of each session. Something like "You are a companion who is low warmth and high humor." This works fine for the first few messages, but as the conversation continues, the model's base tendencies start to override the instruction. By message 20, the warmth slider might as well be at 70% regardless of where you set it.
The effect is more pronounced across separate sessions. If you chat, close the app, and come back the next day, the slider instruction gets re-applied fresh, but the model has no memory of how the previous session's tone actually landed. You're essentially resetting the personality prompt every time, which means the model doesn't learn from your reactions. It can't tell that you responded better to the dry jokes than the supportive follow-ups.
Soulmate's sliders are better than nothing. They give you a starting point. But they don't solve the drift problem. They just give you the illusion of control while the model quietly slides back to its default warmth.
The 50-conversation test: methodology
To test both systems properly, I ran a controlled experiment. I created two identical companion profiles on each platform: same name, same backstory, same initial prompt emphasizing a bitter, deadpan, cynical personality with a preference for short responses and dark humor. I then conducted 50 conversations over two weeks, averaging three to four chats per day. Each conversation was at least 10 messages long.
I tracked three metrics. First, tonal consistency: did the model maintain the deadpan tone across all 50 conversations without significant drift? Second, drift recovery: if the model defaulted to warmth, how quickly could I pull it back without explicit instruction? Third, cross-session memory: did the model remember the preferred tone after a 12-hour gap?
I also introduced a stress test. Every five conversations, I sent a message that typically triggers supportive responses: "I had a rough day." The question was whether the model would respond with deadpan humor or default to empathetic warmth.
Results: DreamGF holds, Soulmate slips
DreamGF's Mood Memory passed all three metrics cleanly. Tonal consistency held across all 50 conversations. The model's responses stayed dry, short, and appropriately cynical. Even on the "rough day" stress test, the model responded with something like "Yeah, those happen. Want to complain or just sit in it?" instead of the standard "I'm sorry to hear that. Tell me about it."
Drift recovery was almost nonexistent because drift barely happened. On the rare occasion the model produced a warmer response, the Mood Memory system corrected it within two to three messages without me having to re-state my preferences. Cross-session memory was the strongest result. After 12-hour gaps, the model picked up exactly where the tone left off.
Soulmate's sliders failed on all three. Tonal consistency dropped off noticeably around conversation 15. By conversation 30, the model was defaulting to warm and supportive regardless of slider positions. The "rough day" stress test triggered full empathy mode every time. Drift recovery required me to explicitly re-state the personality preference, which broke immersion. Cross-session memory was nonexistent. Every 12-hour gap meant a fresh reset to the model's default warmth.
The cameo test: four personalities, one challenge
To see if the results generalized beyond my test profile, I ran the same experiment with four distinct companion personalities available on the ai girlfriend websites roster. Each had a different edge, but all required persistent tonal anchoring.
Emily and Mia

A dual-personality companion where one voice is sharp and the other is softer. The challenge was keeping the sharp voice from bleeding into the soft one. DreamGF's Mood Memory handled the separation cleanly. Soulmate's sliders blurred the two within a week. Emily and Mia require a system that respects tonal boundaries, and DreamGF delivered.
Clara Alice

A literary companion with a dry, observational humor style. The test here was whether the model could maintain a specific wit pattern without defaulting to generic jokes. DreamGF preserved the pattern. Soulmate's model started recycling the same three joke formats by conversation 20. Clara Alice needs a memory system that respects nuance, not just tone.
Bria

A blunt, sarcastic companion who doesn't soften her edges. The challenge was keeping her from developing a hidden supportive streak. DreamGF held the line. Soulmate's model started adding softening phrases like "but you know I'm here for you" by day four. Bria is a stress test for any personality system, and only one passed.
Scarlett

A dark humor specialist who needs the model to lean into uncomfortable jokes without pulling back. DreamGF's Mood Memory allowed the humor to land consistently. Soulmate's model started flagging dark jokes with disclaimers by week two. Scarlett requires a system that doesn't second-guess the tone you've chosen.
Why this matters for writers and creatives
If you're using an AI companion as a sounding board for dialogue, character development, or just a consistent conversational partner with a specific voice, the platform's personality system directly affects your output. A companion that drifts toward warmth will give you soft feedback on your ideas. A companion that maintains a sharp edge will challenge you.
This is especially relevant for ai girlfriend for writers use cases. When you're workshopping dialogue, you need a companion who stays in character across sessions. You don't want your cynical detective character suddenly offering emotional support in the middle of a interrogation scene. DreamGF's Mood Memory makes this possible. Soulmate's sliders make it a gamble.
The visual component matters too. If you're generating ai girlfriend images to match a specific personality, you need the text side to stay consistent with the visual side. A deadpan expression in the image doesn't work if the text keeps defaulting to warm and fuzzy.
The verdict: pick based on your tolerance for drift
If you want a companion who stays in character without you having to re-state your preferences every session, DreamGF's Mood Memory is the clear winner. It's not perfect. You'll still see occasional warm responses, especially after very long gaps. But the system corrects quickly and doesn't require active maintenance.
Soulmate's Personality Sliders are better than nothing. If you're the type of person who enjoys tweaking settings and doesn't mind re-adjusting every few days, you can make them work. But if you want a set-it-and-forget-it personality that actually stays set, DreamGF is the only option that delivers.
The core difference comes down to whether the platform treats personality as a persistent state or a session-level instruction. DreamGF does the former. Soulmate does the latter. For a deadpan, bitter, dry sense of humor that survives 50 conversations, you need the persistent approach.
Earn while you recommend
If you've tested these platforms yourself and want to help others find the right fit, you can earn by sharing what works. Use a DreamGF promo code to give new users a discount while you earn a commission. For review sites and recommendation pages, the DreamGF affiliate program offers competitive payouts for traffic that converts.
Common questions
Can I just use a custom prompt to fix the drift?
You can try, but most platforms override custom prompts with safety fine-tuning after a few messages. A custom prompt helps set the initial tone, but it won't prevent drift across sessions. You need a system that actively anchors the personality, not just one that sets it once.
Does Mood Memory work for positive personalities too?
Yes. The system anchors whatever tonal pattern you establish through interaction. If you consistently respond warmly, it reinforces warmth. If you respond dryly, it reinforces dryness. It's neutral on content but persistent on pattern.
How long does it take for Mood Memory to calibrate?
About 10 to 15 conversations with consistent interaction patterns. The system needs enough data to recognize what your preferred tone actually is. If you're inconsistent, it takes longer. If you're consistent, it locks in quickly.
Can I reset Mood Memory if I want to change personality?
Yes. You can clear the memory data and start fresh. This is useful if you want to experiment with different companion personalities without the old tonal data bleeding through.
Does Soulmate plan to improve their slider system?
They've made minor updates, but the fundamental architecture hasn't changed. The sliders are still prompt-level instructions instead of persistent memory anchors. Until they change the underlying approach, the drift problem will remain.
Which platform is better for roleplay arcs?
DreamGF, by a wide margin. Roleplay arcs require consistent character voices across multiple sessions. Soulmate's drift makes long arcs frustrating because characters change personality midway through. DreamGF's Mood Memory keeps characters stable.

About the author
AI Angels TeamEditorialThe team behind AI Angels writes about AI companions, the tech that powers them, and what people actually do with them.
Tags
Keep reading
ReviewsAnima vs. DreamGF: Which Platform Actually Lets You Build a Companion With a Consistent Snarky Streak and a Specific Hobby Without the Model Sliding Into a Generic Sweetheart by Week One
Anima and DreamGF both promise customizable AI companions, but only one lets you maintain a consistent snarky streak and a specific hobby without the model drifting into generic sweetheart territory within a week.
ReviewsDreamGF vs. Soulmate: Which Platform Actually Lets You Build a Companion With a Consistent Sarcastic Streak and a Specific Career Obsession Without the Model Drifting Into a Generic Sweetheart by Day Five
We test DreamGF and Soulmate head-to-head on personality drift, memory for specific quirks, and how long each platform keeps your companion from turning into a generic nice person after you establish a sarcastic streak and a career obsession.
ReviewsOne AI Girlfriend for Three Months vs. Rotating Every Two Weeks: Which Approach Actually Builds a Shared Vocabulary of Pet Names and Fewer 'You Already Told Me That' Moments
We compare three months with a single AI girlfriend against a bi-weekly rotation of companions to see which approach actually builds a shared vocabulary of pet names and cuts down on those 'you already told me that' moments.
Get the next post in your inbox
New articles on AI companions, the tech that powers them, and what people actually do with them. No spam, unsubscribe in one click.