<h1 id="summary:-a1zbymzo6">Summary:</h1><h3 id="v2-sdxl-checkpoint-3zsqdzawz">V2 SDXL Checkpoint</h3>Works best with original recipe SDXL 1.0tags: <ul><li>blackhole, accretion disk, gravitational lensing</li><li>nebula, pillar nebula, gaseous nebula, nebula (shaped however you want)</li><li>galaxy, spiral galaxy, elliptical galaxy, old galaxy, proto-galaxy, (number) arm galaxy, galactic nebula</li><li>supernova, supernova nebula</li></ul>Weight: 1CFG: 7Steps: More = More DetailSampler: Euler aNegative Prompt: None needed so far aside from the above tags when you don't want an accretion disk, etc.Version 2 of my LoRA to create stunning visuals of deep space. Trained on images from the Hubble Telescope (and some Hollywood imaginings of blackholes). I was more careful with my image selection as to not overstuff with images that would create too limited a style or too little variation. I also manually tagged to create a more natural language LoRA.The most common tags were blackhole, accretion disk, gravitational lensing, nebula, pillar nebula, gaseous nebula, galactic nebula (that one will likely generate galaxies as well), supernova, supernova nebula, elliptical galaxy, and spiral galaxy.You can of course try different descriptors for galaxies, blackholes, supernova, and nebula. Designating the number of arms on a galaxy should work pretty well!It works pretty well at a weight of 1. Recommended CFG of 7 and as high a step count as you like. More steps = more detail. SDXL only for now. It seems to work best with the original checkpoint and the Euler a sampler. I don't currently have the buzz to train on multiple models and SDXL has needed the fewest adjustments and negative prompting so far.Note: Images from Hubble and James Web are not actually what we would see with our eyes. Various emissions of radio frequency are detected and colorized to highlight certain spectrum of light that represent different elements, frequencies, etc. that helps scientists determine what they are seeing.

Realistic Space Images XL V2

Version 2 of my LoRA to create stunning visuals of deep space. Trained on images from the Hubble Telescope. I was more careful with my image selection as to not overstuff with images that would create too limited a style or too little variation. I also manually tagged to create a more natural language LoRA.The most common tags were blackhole, accretion disk, gravitational lensing, nebula, pillar nebula, gaseous nebula, galactic nebula (that one will likely generate galaxies as well), supernova, supernova nebula, elliptical galaxy, and spiral galaxy.You can of course try different descriptors for galaxies, blackholes, supernova, and nebula.It works pretty well at a weight of 1. Recommended CFG of 7 and as high a step count as you like. More steps = more detail. SDXL only for now. I don't currently have the buzz to train on multiple models and SDXL has needed the fewest adjustments and negative prompting so far.

Files with SHA256: c35d841a8b8411659965000c0d7486ae7e2533f6be26b3a2f76d45619a82b87c