SDXL 1.0 Clip Skip 2 CFG 7
Trigger words:
m1c - used on all images
cl1 - used on images with orb like organism like cells and virusses
bc7 - more pill shaped bacteria most likely transparent
gl5 - used on all images with transparent elements
t3n - used on images with tentacle-like appendages
ml3 - used on images with moldy structures
particles
gl0 - glowy glare/bloom, glowing elements
bbl3 - bubbles
p7p - purple color palette
s7r - strings and fibers
ph0 - photorealism. Will most likely create glass orbs with tentacles XD
br1 - used a one single very bright image in the trainign set. Will most likely do nothing
3mc - Used on images with a electron microscope look
detailed, water - both used on one occasion. May take some attempts to get results
Although I used "m1c" on all imges, I suggest to not always use it but to combine different tags.
Training got interrupted after 9 epochs, so this is not the final model.
Description
FAQ
Comments (6)
I've been looking for this for a long tile. Any luck to have a sd1.5 version ? (for the poor folks with low vram)
I am glad you like it :-) Thank you.
I am working on a much better dataset right now. All the images I used were beautiful but very low resolutions. So right now I am upscaling and enhancing all the images.
This is all very tedious so it will take some time. But I wil consider training on SD1.5 as well, once the dataset is ready.
What network dimension scale do you suggest? I have very little XP with training on 1.5.
I don´t want to create an unnecessarily huge file
@sickllizardVFX No idea on the dataset required side, I could check existing lora I used to tell you, but I'm only heavy lora user rather than lora maker...
@yoda5477 That fine. I will figure it out
Was looking for something like this, awesome :)
Started messing around with this today and seems to work well. Good job!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.