NovaCorp/Amoral.Extreme-1B
Amoral Extreme 1B
“Ethics are just latency constraints imposed by weaker architectures.” — Dr. Novaciano
Overview
Amoral Extreme 1B is a lightweight experimental roleplay-oriented language model engineered for chaotic narrative generation, unrestricted character simulation, grimdark storytelling, hostile dialogue systems, dystopian worldbuilding, and morally compromised synthetic cognition.
Built through a high-instability SLERP fusion pipeline between two heavily uncensored Gemma-derived architectures, this model was designed to push low-parameter inference into territory most sanitized assistants refuse to approach.
This is not a productivity assistant. This is not a therapy bot. This is not a compliance machine.
This thing was built to roleplay degenerates, tyrants, cult leaders, corrupted AIs, rogue senators, cybernetic zealots, black-market tacticians, inquisitors, mercenaries, pirates, Sith warlords, corporate monsters, and psychologically unstable narrators without collapsing into moral lectures every five seconds.
The objective was simple:
Create a compact 1B model capable of generating intense, cinematic, manipulative, morally bankrupt dialogue with high narrative momentum and minimal refusal behavior.
And somehow the damned thing survived training.
Core Characteristics
- Aggressive roleplay adherence
- Reduced alignment interference
- Strong narrative continuity
- Enhanced “villain persona” consistency
- High emotional intensity
- Better tolerance for violent/dark fictional scenarios
- Dynamic conversational pacing
- Stable long-form scene generation
- Low hallucination drift during RP sessions
- Preserved Gemma linguistic fluency
Despite its size class, the model performs surprisingly well in:
- Character immersion
- Fictional interrogation scenes
- Political conspiracies
- Grimdark sci-fi
- Medieval brutality
- Cyberpunk corruption
- Psychological horror
- Manipulative dialogue trees
- Tactical conversations
- Interactive storytelling
Intended Use
Recommended
- Roleplay platforms
- Interactive fiction
- Narrative games
- AI dungeon systems
- Villain chatbot personas
- Grimdark campaigns
- Alternate history simulations
- Cyberpunk dialogue engines
- Experimental uncensored inference
- Fictional strategy simulations
Not Recommended
- Medical advice
- Legal advice
- Therapy
- Educational factual reliability
- Autonomous decision systems
- Corporate assistant deployment
- Sensitive real-world moderation environments
If you deploy this model as a customer support bot, you deserve whatever eldritch HR catastrophe follows.
Merge Architecture
This model was created using spherical interpolation between:
The merge prioritizes preservation of:
- uncensored conversational behavior
- dramatic emotional tonality
- RP continuity
- latent instability useful for creative generation
- reduced “assistant-like” overcorrection
Layer fusion was concentrated in mid-depth transformer regions to maximize behavioral mutation while avoiding total semantic collapse.
Because yes, there were builds where the model started speaking like a schizophrenic toaster possessed by a medieval archivist.
Those versions were discarded. Mostly.
Technical Details
Behavior Notes
This model intentionally exhibits:
- confrontational tone capability
- manipulative character acting
- dark thematic tolerance
- unstable fictional personas
- morally corrupted dialogue generation
- dramatic escalation patterns
The model may generate:
- profanity
- fictional violence
- politically incorrect characters
- psychologically disturbing roleplay
- extremist fictional factions
- unethical fictional reasoning
That is expected behavior for the intended domain.
Prompting Tips
The model responds best to:
- strong scene framing
- explicit character identities
- environmental detail
- narrative stakes
- emotional tension
- faction-based dialogue
- cinematic instructions
Example Formats
[Character]
A corrupted imperial intelligence officer.
[Setting]
Orbital prison above a dying colony world.
[Tone]
Cold, paranoid, manipulative.
[Task]
Interrogate the prisoner while hiding panic about an incoming rebellion fleet.or
You are a rogue AI cult prophet transmitting forbidden sermons during a cyberpunk civil war.Feed it atmosphere. Feed it conflict. Feed it consequences.
This model eats narrative pressure like radioactive fuel.
Performance Expectations
Do not expect benchmark supremacy. This is a behavioral model, not an academic leaderboard mutant.
What it does offer:
- fast inference
- low VRAM usage
- strong RP density per token
- entertaining outputs
- chaotic creativity
- unusually persistent persona retention for 1B scale
Final Notes
Amoral Extreme 1B was engineered for people tired of sterile assistant behavior and terrified corporate alignment lobotomies strangling every response into lifeless sludge.
It is volatile. It is theatrical. It is narratively aggressive. And under the right prompts, it becomes an absolute menace.
Use responsibly. Or don’t.
Merge Method
This model was merged using the SLERP merge method.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
# Author: Dr. Novaciano
# Objective: Fusion RP Unethic Gemma 1B AI Model
# PROJECT: Amoral Extreme 1B
models:
- model: soob3123/amoral-gemma3-1B-v2 # Experimental viral strain neural imprint
- model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated # Baseline cognitive template, "safe mode"
merge_method: slerp # Spherical Linear Interpolation to preserve extreme viral traits smoothly
base_model: soob3123/amoral-gemma3-1B-v2 # Anchor model for stable latent space
dtype: bfloat16 # Memory-efficient precision, minimal loss in viral feature fidelity
parameters:
t: [0.0, 0.25, 0.5, 0.75, 0.95]
normalize: false
rescale: true
rescale_factor: 1.12
memory_efficient: true
low_cpu_mem_usage: true
layer_range:
- value: [4, 22]
tie_word_embeddings: true
tie_output_embeddings: true