The general format is picture of [pose] woman. [๐ง๐ท๐ช๐ฆ๐แด], [V] [C], [background].
Some keywords were translated into l33t versions, the same as my other Flux LoRAs. I have not been able to prove if this is necessary or not, but it has been working well for me.
- picture of reclining sulty woman. medium br34sts in black corset, pretty pink v4g1n4 and 4nus thick black cum
- picture of reclining beautiful princess. large br34sts, ????????? her spread v4g1n4 with open v4g1n4l c4n4l and tight 4nus
- picture of standing s3xy witch. small perky br34sts, tight pink v4g1n4 dripping cum
- picture of standing redhead barbarian queen. small perky br34sts, pretty pink v4g1n4 dripping cum
- picture of brunette barmaidโs 4ss, ???????????, 4ss shot. medium br34sts, thin ????? panties
- picture of raven-haired beautyโs 4ss, ???????????, 4ss shot. large br34sts with perky n1ppl3s, barely visible v4g1n4
- a close-up picture of pink v4g1n4 in crotchless panties soaked in white ???. faded background
- a close-up picture of gaping pink v4g1n4 slightly moist. faded background
Some keywords were translated into l33t versions, the same as my other Flux LoRAs. Those words include:
- penis: p3n1s
- vagina: v4g1n4
- vaginal ???: v4g1n4l s3x
- nipple: n1ppl3
- sex toy: d1ld0
- sex: s3x
- oral: 0r4l
- oral ???:0r4l s3x
- breast: br34st
- clitoral: cl1t0r4l
- clitoris: cl1t0r1s
- anus: 4nus
- anal: 4n4l
- anal ???: 4n4l s3x
- ass: 4ss
- buttocks: 4ss
- butt: 4ss
- dildo: d1ld0
- naked: n4k3d
- nude: nud3
## Dataset
Trained on about 250 images that I generated in the desired poses.
Images with bad hands, multiple belly buttons, or watermarks were removed.
v1 captions were written by a human, which does not seem to be as effective as VLM captioning (I recommend Joy or Florence 2).
v2 captions used Joy captioning.
Reclining After S[Qwen/Flux]
v2 beta 1 out now. Please try it out and post your images in the gallery, thoughts in the comments.
v2 changes:
- better hands
- better prompt adherence
- support for a greater variety of styles
- better compatibility with other LoRAs
Example Prompts
The general format is picture of [pose] woman. [br34@sts], [\/ag1n@], [background].
Some keywords were translated into l33t versions, the same as my other Flux LoRAs. I have not been able to prove if this is necessary or not, but it has been working well for me.
Some keywords were translated into l33t versions, the same as my other Flux LoRAs. Those words include:
Dataset
Trained on about 250 images that I generated in the desired poses.
Images with bad hands, multiple belly buttons, or watermarks were removed.
v1 captions were written by a human, which does not seem to be as effective as VLM captioning (I recommend Joy or Florence 2).
v2 captions used Joy captioning.