Sign In

Shrunk (PDXL/SDXL)

402
4.1k
67.6k
117
Verified:
SafeTensor
Type
LyCORIS
Stats
1,446
45,245
Reviews
Published
Jun 11, 2024
Base Model
Pony
Training
Steps: 25,060
Epochs: 70
Usage Tips
Clip Skip: 2
Strength: 0.8
Trigger Words
tiny
size difference
shrunk
microxl
Hash
AutoV2
D70EA575FA

A Lycoris concept that attempts to skew perspective and prompt tiny people, show size difference and sense of scale. For PDXL best around ~0.75 strength, for realistic up to ~1.5 may be required for complex prompts. Best while using ADetailer to restore tiny faces.

Pony XL / SDXL V2:

  • See version details.

  • Better faces, re-trained with new trigger words.

  • More pose/couple focused, more flexible.

  • 'tiny' -> top trigger word. WD14 also tagged things like 'minigirl' 'miniguy' so those may help assist.

  • 'size difference' -> more likely to prompt a 2nd subject for comparison.

  • standing/sitting/laying on <object/surface> is more likely to prompt a shrunk perspective/scale, e.g.: 'standing on counter'

V1:

  • See version details

  • Still fairly unpredictable but output needs less img2img fixing

A Lora Locon that attempts to skew perspective and create shrunken people. Trained on 155 images, it works okay SD is pretty bad at faraway(small) people as subjects. Still, it worked out okay so here it is. Recommend generating then dropping the strength to ~0.3 for img2img cleanup and using ADetailer to fix faces. Also try putting the subject on a non-floor surface, like a table/counter/bed/chair.