FULLY_REAL_XL - TRINITY
TRINITY is the current version of my FULLY REAL XL checkpoints. While newer models exceed SDXL in various areas (nsfw, anime, etc.), SDXL still continues to give some amazing interpretations of prompts that are often much better than expected. TRINITY brings that creativity to life with a strong lean toward realism.
TRINITY is the 16th FULLY_REAL_XL checkpoint publicly released, and the culmination of 1000's of hours of training. It does still have some SDXL weaknesses (hands I am looking at you), but minimized as best as possible. It excels at creative prompting, portraits, and detail. TRINITY should work fine with most SDXL style LoRA's. However, it will change the appearance of any character LoRA that was trained on other checkpoints (most of the time).
TRINITY was not geared toward NSFW and for that I recommend a PONY model (or LUSTIFY). However, it will do some NSFW if prompted that direction.
Recommended Settings:
CFG 3-6 (I use 4 mainly.)
40-60 steps = this is not an accelerated version
You can use a lot of samplers. I normally use DPM++ 2M SDE Karras or DPM++ 3M SDE Expo.
VAE is baked in
Hires Fix is helpful
Occasionally Adetailer may be needed, but I use it less with this model than most.
NOTE: This is not a PONY model. It does better with slightly longer (less choppy) prompts and does not understand "score=".
Disclaimer: This model is provided for educational purposes and for the advancement of AI art in general. The creator of this checkpoint assumes no responsibility or liability for the end user's outputs. Any illegal, unethical, or deep fake use of this model is the sole responsibility of the user engaged in those activities. The end user assumes all liabilities from use of the checkpoint.
No permissions are given for anyone to distribute this model on other sites. I do not give permission for it to be used on any image generation service other than CIVITAI.
Description
V1 was a test of concept. V2 takes the training dataset up to 6000 images, adds more diversity of content, and has about 4 times the training hours baked into it. I may run a V3 to dive into facial details and to put some curves back in where FLUX tends to remove them.
FAQ
Comments (11)
ok, let's use Flux, if you say so ... ;)
Thank you for bringing this model back, it's one of the best XL models I've ever used.
leave this up for at least a week! i won't be home to download it properly until then
Personally I think its a bad idea to use AI outputs to make AI inputs, it introduces artifacts. I like the detail on this but don't like the AI issues it has, or anomalies. Floating objects, warped shelves, off center bottles etc, it has that AI look, and doesn't really have artistic shading. Its detailed, but doesn't look like a photo. And the NSFW is weirdly really blurry.
is it commercially usable?
If you are curious what this training technique might produce in a PONY model: https://civitai.com/posts/5814039
Clip Skip 1 or 2?
Image animation was done locally using LTX.
сукпздц, лучше бы триллюстриос тренил дальше. в 2к25 кому нужна сдохля обычная
That is, it turns out that FullyReal 2 and 3 were trained on images of Flux.dev? May be v1 not?
ok pain... i'm out of disk space here, so I'm gonna delete Trinity for now while I'm trying out Illustrious for the first time now that it has some photographic models... don't you dare remove it while I don't have it downloaded! I'm coming back for it! It better be here!
P.S. I still have the old (pain)FULLY_REAL V10 for this exact reason lmao
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.