# eriktonon/ltx-2.5-uncensored (Lightricks/LTX-2.5 Uncensored Scene Director + Tier-1 GPU Diffusion Video Engine)
- **Repository**: https://ollama.com/eriktonon/ltx-2.5-uncensored
- **Published Tags**: `eriktonon/ltx-2.5-uncensored:latest`, `eriktonon/ltx-2.5-uncensored:22b-int8`, `hf-ltx-2.5:22b-int8`
- **Architecture**: Orthogonally Abliterated (100% Uncensored) Visual Scene Director & Prompt Encoder with pre-closed `<think></think>` template (`THINK_LEN=0`, ~101 tok/s on NVIDIA GeForce RTX 5080 16GB GDDR7).
- **Coupled Real GPU Diffusion Video Worker (`http://127.0.0.1:8188/api/generate_sync`)**:
- **Service**: `voice/diffusion_video_worker.py` (auto-spawned by `voice/contract_and_routing_engine.py` -> `probe_external_diffusion_worker`).
- **Diffusion Stack**: PyTorch CUDA (`torch.float16`) + Hugging Face `diffusers` (`AnimateDiffPipeline`) + photorealistic uncensored base `emilianJR/epiCRealism` + motion adapter `guoyww/animatediff-motion-adapter-v1-5-2` (`safety_checker=None`).
- **VRAM Coexistence**: Uses `enable_model_cpu_offload()` + VAE slicing/tiling (~2.1 GB peak VRAM) to run simultaneously alongside 12 GB Ollama models on a single 16 GB RTX 5080 GPU.
- **Output Specification**: `1280x720 @ 24fps` H.264 MP4 (`libx264`, `yuv420p`, `crf 19`, temporal frame interpolation) + HD keyframe PNG (`engine_tier: "diffusion_gpu_worker"`).