V2:

Retrained with original dataset from V1+ generated images from V1 baked together.

Trigger word: fate_crest_worms

Quantity & coverage triggers (still random even in V2)

• several → just a few worms crawling on skin

• multiple → medium amount, clearly visible but body still recognizable

• heavily_covered_in_worms → almost cocoon-level coverage

Environment triggers

• worm_pit → character standing/lying in a thick layer of worms (floor completely covered)

• endless_worms_sea, infinite_worms, → pure infinite ocean of worms, can be combined with any trigger for support basis.

• lying_in_worms → can be combined with above triggers as support basis

Internal & transformation triggers

• bloated_belly → worms inside, swollen abdomen (no x-ray)

• x-ray → see-through view showing worms inside the body

• vaginal_penetration, anal_penetration, oral_penetration, navel_penetration → worms actively entering (in this version it is also became random at generation)

Important note for local generations users:

Even in this updated version txt2img can sometimes be conservative with quantity and quality.
For maximum control and most accurate results use img2img or inpaint sketch → send to img2img. That workflow gives near-perfect results 95 % of the time.

V1:

Due to very small dataset, Version 1 will generate random results.

This is my first attempt to make lora about concept with multiples objects and if you have advice of good tagging, please make a post and do so. Since I might have made insufficient tagging, trying to control numbers of worms.

Base trigger: fate_crest_worms,

Ordinary prompt if you want crest worms <10: fate_crest_worms, several, on_skin

Ordinary prompt if you want a massive amount of worms: fate_crest_worms, swarm, covering, large mass

Or you can mix all in one and see what you will get. Alter the parameter since the result might be not what you wanted.

Additional tags: infiltration, x-ray, inside_body, bloated belly

Also you can try add: tentacles, tentacle pit, huge worms, gigantic worms and additional booru tags that might be useful within concept.

If you generating locally, do not hesitate to use img2img or inpaint sketch, since LoRa is generating well in such method. Txt2img can be random, while these two can give you more control and flexibility.

I also downloaded training data to that post, so if you are a professional concept maker, you can make your own version of LoRa. Just make sure to PM me with how you did it better than mine.