Sign In

Refined

1.5k
11k
39
Type
Checkpoint Merge
Stats
3,724
Reviews
Uploaded
May 19, 2023
Base Model
SD 1.5
Hash
AutoV2
70F3CD399A
0
0
0
0

Like photorealism? Try my new fine-tune 'Lomostyle'

Merged Model with slight fine-tuning focused on human anatomy and skin texture, can also be used as a general purpose model. v10 is a foundation that will be further built upon with more mergers + fine-tuning.

List of models used in merge can be found in the version description.

Limitations (as of now)

  • Photos of Men

  • Anime

  • SFW generations without specific tags

Trigger Words

Trigger words are not required.

However, the following tags can be used to alter the generation.

analog style, modelshoot style, wa-vy style, synthwave, sci-fi, cyborgdiffusion, erect penis, flaccid penis, postapocalypse, 3d, render, nsfw, sfw

For tags like "nsfw", "sfw", "3d", and "render", you can swap these between your positive and negative prompts to get more specific outputs.

For example, putting "sfw" in your positive prompt, and "nsfw" in your negative prompte will push the model in the direction of generating sfw outputs. You can also add/decrease attention using ()and []

Textual Inversion (TI)

Many of sample images include a negative textual inversion that helps fix things like hands, ect.. This isn't a requirement, I've tested the model without this TI and didn't have any issues. I do recommend using it though, as it does help.

Bad Prompt TI - https://huggingface.co/datasets/Nerfgun3/bad_prompt

I recommend using Version 2 and setting it's weight to 0.8 in the negative prompt; (bad_prompt_version2:0.8)

For webui, bad_prompt_version2.pt goes in:

sd/stable-diffusion-webui/embeddings

Hires Fix

I used Hires Fix for many of the sample images. This is not a requirement but it does seem to improve results. I mainly use latent(bicubic antialiased) with Denoising strength set to 0.65.


Check out my other models

SDXL

SD1.5

LoRA

Questions or Feedback?

Visit my thread on the Unstable Diffusion Discord Server