New to local AI, RTX 5070 Ti 16GB setup, looking for better unrestricted image models than Z-Image-Turbo
retained text
I set up a local LLM today because AI apps seem to be getting more expensive and restrictive over time. My main use case right now is generating (N)SFW content, mostly text-to-image. LLM setup: Ollama Open WebUI Huihui Qwen3-VL 8B Instruct Abliterated, 16K context ComfyUI Z-Image-Turbo Tailscale for remote access from my phone PC: Ryzen 9 9900X RTX 5070 Ti 16GB 32GB DDR5-6000 2TB NVMe SSD I’m using Z-Image-Turbo for now, but I’m not very happy with the results. Prompt adherence and image accuracy seem pretty inconsistent, and I get a lot of mistakes. What image models are you guys using on similar hardware? Also feel free to tell me if there’s anything I should have set up differently, anything worth changing, or any model recommendations for better quality while still staying local and relatively unrestricted. Update: Thanks everyone for all the input. When I first made the post, I was so clueless that I didn’t even know what LoRAs were, so I ended up going down the rabbit hole and researching everything people suggested. From what I’ve learned so far, Krea 2 + LoRAs for images and Wan 2.2 + LoRAs for video seem like the best fit for what I’m trying to do. I’m going to work on setting both up today and test them on my hardware. I’ll update the post afterward with what worked, what didn’t, and how the results compare.
document fetched 2026-09-28 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1wre4jk/new_to_local_ai_rtx_5070_ti_16gb_setup_looking