Sign In

Blaziken Lora

273
1.6k
30
Verified:
SafeTensor
Type
LoRA
Stats
1,397
Reviews
Uploaded
Apr 4, 2023
Base Model
SD 1.5
Training
Steps: 100
Epochs: 2
Trigger Words
blaziken
blue iris
yellow sclera
neck tuft
chest tuft
crotch
scuted arms
talons
talon hands
talon feet
+1 more
Hash
AutoV2
3DC5EEA0E5
0
0
0
0

It's a lora, and it makes spicy fire birds. That's about it.

I trained it 768x768 (bucketed) on miscellaneous images from Pixiv, e621, and GelBooru that weren't straight up r34. It isn't as prone to unexpected NSFW as other loras I've messed with, but do make sure to negative weight appropriately; using genders as negative prompts should work to pull out more normal androgenous ones.

I recommend using SevensMix and the AT3 version with a strength of 0.7-0.8. Here's all the tags you can use:

blaziken, nintendo, pokemon, gen3, avian, biped, anthro, red body, feathers, red feathers, tuft, crotch tuft, chest tuft, neck tuft, blue eyes, blue iris, yellow sclera, colored sclera, beak, hair, long hair, blonde hair, talons, talon hands, 3 fingers, scutes, scuted arms, tail feathers

I would recommend checking the images though, some tags are repeated multiple times and using them is similar to overweighting, i.e. (blaziken:1.8), and can fry you generations. Mix and match these as necessary, or use them as additional negatives if you mix a lot of loras together. Use just "scuted arms" and leave out "scutes" for example. Post anything interesting you get from this, or any funny bad gens.

There's 2 versions for 2 different bases that are common in most mixes: AnythingV3 and AbyssOrangeMix3. AT3 works with a lot of stuff and is my favorite, AOM3 is a fallback if a model doesn't work for whatever reason, but if you're using an AOM3 model then you've made a poor decision.