# Looking for a fast, smart uncensored LLM for prompt generation on RTX 5070 Ti (16GB VRAM)? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/c3df9554950af577

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-11-02 23:49 ingested 2026-11-03 02:40

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xmdjjy/looking_for_a_fast_smart_uncensored_llm_for/)

## retained text



```
Hi everyone,

I need recommendations for a local LLM to run via **Ollama** (integrated with the **Page Assist** Chrome extension) specifically to generate uncensored prompts.

My hardware setup:

* **GPU:** NVIDIA RTX 5070 Ti (16GB VRAM)
* **RAM:** 64GB DDR4
* **OS:** Windows 10

**What I've tried so far:**

* **Gemma (7B/9B):** Extremely fast (instant responses), but lacks the intelligence/detail quality I need for complex prompts.
* **Qwen2.5-VL-32B:** smart, but... -.- way too slow for daily use because it overflows 16GB VRAM and offloads to System RAM... :\\

**What I'm looking for:** I'm searching for a "sweet spot" model in the **12B to 14B parameter range** (or a lightweight \~8B Vision model) that fits 100% inside 16GB VRAM at Q4/Q5 quantization. It needs to be:

1. Uncensored / Abliterated (no refusal for complex/adult media prompts).
2. Fast (50+ tokens/sec).
3. Available on Ollama or easily converted to GGUF for Page Assist.

Which 12B/14B uncensored models are you currently using for prompt expansion and generation?

Thanks!!! 
```



document fetched 2026-11-03 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xmdjjy/looking_for_a_fast_smart_uncensored_llm_for
