POV of a character covering the viewer's mouth. Z-Image Base LoRA is the best one yet!
The success rate is lower in portrait aspect ratio (except Z-Image)!
Z-Image Base:
Trigger: A woman/man extends her/his arm directly toward the viewer, placing her/his flexed palm in sharp foreground focus while her face remains softly blurred in the background.
Then describe what you want.
ILLUSTRIOUS:
Recommended weight: 1.0
Sampling steps: 20-30 (too high = risks of extra fingers!)
Checkpoints: Trained on v0.1 should work across other checkpoints.
SDXL (V4) (outdated):
This is a complex model with many keywords and possibilities. The dataset comprises 480 pictures, most captioned to 75 tokens.
Recommended weight: 1.0, but up to 1.2 - 1.3 on heavily trained checkpoints like NatViS.
Sampling steps: Follow the checkpoint's recommendations
Checkpoints: RealVisXL, Juggernaut, Animagine XL
Keywords:
Camera position: wide-shot, close-up, or low-angle
Camera focus: focused_hand or blurry_hand.
When using 'focused_hand', I recommend adding the keywords 'hand palm, fingerprints' to increase details.
Arm position: flexed arm or extended arm, left_hand or right_hand
Skin texture: thin skin creases, deep skin creases, soft skin creases, many skin creases, dark skin creases, rosy skin
Nails: long nails / pointed nails / short nails / no nails, nails + color
Example: long nails, red nails
Gloves: gloved, glove + color, glove + material.
Example: gloved, leather glove, black glove.
For better success, I recommend increasing the weights of the keywords, like '(gloved, latex glove, blue glove:1.15)'
Rings: single ring / two rings / multiple rings/wedding ring, ring + material
Example: single ring, wedding ring, gold ring, large ring
Bonus: spit handgag: spit_handgag, wet hand, spittle, drool
Full prompt example: hand palm, left_hand, (cupping, focused_hand, fingerprints), black hair, goth, bdsm room, heavy makeup, low-angle, shushing mouth, piercing blue eyes, dark makeup, punk aesthetic, red backlight, hard shadows, black robe, single ring, large ring, soft skinfolds, hair fringe, red nails, short nails, [wide-shot]
FLUX (outdated):
Recommended weight: 1.0
Sampling steps: 20 - 30
Checkpoints: Flux Dev
Use natural language. Just trick the model with that sentence first: "A [low-angle] photograph of a woman with her hand palm raised at the camera." Then describe the hand, the woman, and the environment with natural language.
PONY (outdated):
Recommended weight: 1.0
Sampling steps: 20 - 30
Checkpoints: AutismMix, Pony 🦄FaeTality, Real Dream
Key words: gloves + the type of glove, glove + the color you want, wearing ring(s), wearing bracelet(s), nails + the color you want, outstretched hand, cupping hand, few/deep hand creases, subtle/deep hand creases, low-angle
Description
FAQ
Comments (2)
While I do appreciate the effort, this doesn't have the appearance of a true POV hand gag. Alas, it only looks like a woman is casually standing with her arm outstretched. POV would be what you would see if you try putting your hand over your own mouth as an example:
The very bottom of the screen would show just part of the top finger--maybe at a slight angle--and the thumb sticking up from the bottom-left or bottom-right edge. The thumb might even block 30-40% of the picture depending on if the hand-gagger is holding their thumb firmly in a 90-degree angle to the hand, or is covering one of the eyes. The very top edge of the finger and outer edge of the thumb would have very thin lighting, while the rest if the finger and thumb would be dark.
In the above example, you wouldn't see the person, just the parts of the hand. Some kind of motion blur effect could help make the scene look a little more dramatic, as if the subject had just turned their head abruptly in response to the unexpected mouth grab.
Hopefully this doesn't come off as hostile. it's just a subject near and dear to my heart.
I appreciate your honest feedback.
Try to add the "close-up" keywords and set a higher weight for it like <close-up:1.5> (it's not perfect but it improves your desired outcome)
I will improve the future version of the model with better close-ups, bokeh, and motion blurs.
Regards