Download
1 variant available
1230 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
3.7K0 1 2 3 4 5 6 7 8 9.0 1 2 3 4 5 6 7 8 9K
8440 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
(43)
Jan 26, 2025
Illustrious
Yo, I used the new dataset format, Shaka as the first stepping stone >.>';;;. Goal is to make sure the LoRA encapsulates/recognize more of the tags in a dataset while remaining flexible
Merged the three core tags into 1 --- I made sure that the core tas describes who the subject focus is, what they are, and what do they represent.
Hidden tags such as "anime screencap, pokemon (creature), crossover, cosplay etc." will still be effective for their uses, especially anime screencap with the style / safety net function <.<
I cannot marge all the character details in 1 trigger word / core tag because of finer details focus. This is to preserve the elements without confusing the AI. Sample: a "blue eyes" tag should not be together with "red lips" tag under a single core tag. This will hinder flexibility and cause artefacts by the oversaturation of elements
50-70 images is optimal without heavy style influence,
1060-1090 steps is still best for me <.<
32/16;ND/NA
You can still remove the helmet by replacing "helmetted male focus" --> "male focus" then adding 1boy,
Am still testing out this dataset format and so far it's 7.5/10 XD, will need to test this with pokemon creature inducing characters as well as the difference if I replace the
" - " with "_"
Show more
License:
Illustrious LicenseVEGAPUNK SHAKA from ONE PIECE!
ββββΰ¨ΰ§ββββ
Reincarnated into an alternate universe where he races for forbidden pleasures
πCORE TAGSπ
Pony = humanoid robot-male focus, ShakaOP, 1humanoid robot, anime screencap
ILXL = shaka-helmetted 1male focus, anime screencapββββ ΰ¨ΰ§ ββββ
πThe Gallery is open. Unique tags are present for your favors.
πDownload for the willing. Interest yourself of its creation
πExperiment freely but logically. For the creative is bound by descriptions to be perceived
ββββ ΰ¨ΰ§ ββββ
π«Berryful Tipsπ«
Use with Illustrious Diffusion XL Based models
If images have rainbow colored static or misplaced colors reduce the LoRA strength
If imagined style is difficult to implement remove the tag "Anime screencap"
If image is stiff at every generation, you may...
Remove the character name within the unique tag
e.g. unique tag for eyes --> color_eyes
Change the unique tag to a general tag
e.g. unique tag for upper clothing --> shirt
Reduce overall LoRA weight but keep a/a sequence of unique tag integrity with (_PARENTHESIS_)
Reduce the weight of a/a sequence of unique tags with [_BRACKETS_]
For better checkpoint infusion, Increase CFG scale and reduce LoRA weight.
For better full body images, Adetailer or face fix can beautify facial mishaps
ββββ ΰ¨ΰ§ ββββ
πLIQUIDITYπ
CHECKPOINTS have their own BIAS
The best checkpoints can be used universally. The special checkpoints are finetuned to the preferred subject/class
ALWAYS remove "Anime screencap" for checkpoint bias to influence LoRA
You can substitute the subject focus in the core tags depending on your imagination
Substitute subject focus with the opposite gender
e.g. male focus --> female focus β female focus --> male focus
Substitute the 1class with preferred class
e.g. 1boy --> 1girl β 1girl --> 1boy :: 1boy --> 1furry male :: ...
ποΈINTERESTING NOTE:
Keep the subject focus and class but add a contradicting body part such as the tag breast on males for larger pectorals or penis on females for futa females
ββββ ΰ¨ΰ§ ββββ
πBAKING OWN OC!π
When MIXING Character LoRAs (mine), reduce weight to 0.6-0.8. while keeping their mother tags in.
When MIXING Character LoRAs (not mine but with mine), reduce my LoRAs weight appropriately while keeping their mother tags in.
When influencing own imagination with my LoRA, reduce weight appropriately
ποΈINTERESTING NOTE:
Differing the color of a unique tag will result with either a successful color change or a complete Checkpoint takeover. Use the (Parenthesis) to keep character likeness integrity
ββββΰ¨ΰ§ββββββββΰ¨ΰ§ββββ
ββββΰ¨ΰ§ββββDon't forget to like, comment, follow or direct post in the gallery βΈ(ο½‘Λ α΅ Λ )βΈβ‘
Daberry generates their showcase images and fanarts in TENSORART. You can too.
π©
NOTE REGARDING REPRINTS( ΛΆΒ°γ Β°) !!
Ask for my permission if you wish to reprint or upload my models on sites I have no knowledge of. I would love it if my LORAs can only be downloaded through CIVIT. Thank you (Λ΅ β’Μ α΄ - Λ΅ ) β§



