I have thought a long time about uploading this model. It is a merge of many models. I have worked hard getting it to the point that it will output consistent photos. I have been working on it for over a year and I can't even remember all of the models that I have merged into it.
It works very well with Loras trained on SD 1.5. It is actually really consistent at creating hands and facial features for an AI model. It still struggles with feet though.
It is geared towards creating great images with simple text commands. More complex commands may or may not yield better results. If you look at the prompts for the sample images that I have uploaded, you will see that they are all very simple prompts.
It is designed to create life like photographic images. It may perform well with other art styles, but I have not tested it, as that was not what I was going for.
I would love for people to try it and upload their images and feedback. I would love for people to merge it and improve it as they see fit. This is for people to have fun with and I look forward to seeing what you do with it.
It will produce NSFW images just as easily if not more so than SFW images.
I generally have had best results with DPM++ 3M SDE Exponential at around 70 steps, using HiRes Fix at 2x with 4xUltraSharp at 40 steps with .04 denoising strength. GFG Scale of 2. Also, VAE is baked in.
Please if you use it, upload your results and comment any improvements that you have found so that we can all enjoy. I'm hoping that many people will find this enjoyable and with collaboration we can all create better art. Thank you so much.
Description
Updated to Photogrametry_V2
Still focused primarily on achieving the highest quality, true to life, photographs of people possible. Large improvements of textures, including skin texture. Produce photographs without the "plastic skin" look.
Performs better with outdoor scenes.
Model tested in Automatic1111 and Forge, (which is based on Automatic1111). Forge is faster with the same or better quality. If you are using Forge, I recommend using sampling method Euler A, and changing HiRes sampler to DPM++ 3M SDE. Euler A works better for composition but will leave a plastic effect. So however you do it, you will get better results with DPM++ 3M SDE as the sampler in your last step, either in HiRes or img2img.
Improvements in quality and speed. Updated model allows for better quality with about 30 fewer steps. This allows for much faster image generation, while also improving output quality.
New recommended settings are Euler A or DPM++ 3M SDE Automatic at around 40 steps, using HiRes Fix at 2x with SwinIR 4x or ESRGAN_4X, with 0.3 -0.5 denoising strength. GFG Scale of 3. Image size 768,768. It also helps to put (malformed hands:1.4), (malformed feet:1.4), in the negative prompt.
Prompt understanding is also slightly improved. No need to use a bunch of "quality" keywords to achieve good results. I do recommend using FastNegativeV2, which can be found here: Fast Negative Embedding (+ FastNegativeV2) - v2 | Stable Diffusion Embedding | Civitai
FAQ
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.