v2.0
better than v1.0?
This commit is contained in:
+76
-37
@@ -1,52 +1,91 @@
|
||||
# adding custom library
|
||||
# scaling guide
|
||||
|
||||
## example: hair color library
|
||||
## current system
|
||||
|
||||
your reference: `HairColors.txt`
|
||||
300+ options across 17 categories.
|
||||
|
||||
```
|
||||
Platinum blonde hair: Blindingly bright, metallic white-blonde...
|
||||
Icy white hair: Pure, snow-white hair strands with cool blue...
|
||||
```
|
||||
all alphabetically sorted.
|
||||
|
||||
convert to:
|
||||
"None" option skips blocks.
|
||||
|
||||
## adding more options
|
||||
|
||||
### from reference docs
|
||||
|
||||
you have:
|
||||
- AngleShot.txt (100+ angles)
|
||||
- HairColors.txt (40 colors)
|
||||
- HairTexturesStyle.txt (200+ styles)
|
||||
- Outfits.txt (50+ outfits)
|
||||
- RevealingOutfit.txt (100+ variations)
|
||||
- SexyPoses.txt (200+ poses)
|
||||
- LocationsIndoor.txt (300+ locations)
|
||||
- Atmospheres.txt (100+ atmospheres)
|
||||
- Filters.txt (40 filters)
|
||||
- etc.
|
||||
|
||||
### parse and add
|
||||
|
||||
edit `libraries/comprehensive_portrait.json`:
|
||||
|
||||
```json
|
||||
{
|
||||
"meta": {
|
||||
"name": "hair_colors",
|
||||
"version": "1.0"
|
||||
},
|
||||
|
||||
"blocks": {
|
||||
"color": {
|
||||
"Platinum Blonde": "Blindingly bright, metallic white-blonde hair that reflects light",
|
||||
"Icy White": "Pure, snow-white hair strands with cool blue undertones",
|
||||
"Silver Grey": "Metallic grey hair that looks like spun sterling silver"
|
||||
}
|
||||
}
|
||||
"angle": {
|
||||
"Ceiling Shot": "ceiling-mounted camera angle looking down",
|
||||
"Extreme Low Angle": "extreme low angle from ground",
|
||||
"Helicopter Shot": "aerial helicopter perspective"
|
||||
}
|
||||
```
|
||||
|
||||
save as `libraries/hair_colors.json`
|
||||
add as many as needed.
|
||||
|
||||
## notes
|
||||
alphabetical sorting happens automatically.
|
||||
|
||||
- dropdown name = key ("Platinum Blonde")
|
||||
- description = value (the actual text)
|
||||
- block name ("color") must match schema order
|
||||
### multi-library approach
|
||||
|
||||
## schema update
|
||||
if one lib gets too big (1000+ options), split by context:
|
||||
|
||||
add "color" to assembly order in `schema/connector.json`:
|
||||
|
||||
```json
|
||||
"order": [
|
||||
"subject",
|
||||
"color",
|
||||
"hairstyle",
|
||||
...
|
||||
]
|
||||
```
|
||||
portrait_casual.json → everyday looks
|
||||
portrait_formal.json → elegant/professional
|
||||
portrait_fantasy.json → magical/ethereal
|
||||
portrait_cyberpunk.json → neon/tech
|
||||
```
|
||||
|
||||
restart comfy.
|
||||
switch library dropdown to change context.
|
||||
|
||||
## "None" handling
|
||||
|
||||
selecting "None" = blank = skip that block.
|
||||
|
||||
example:
|
||||
- hair_ornament: "None" → no ornament text in prompt
|
||||
- outfit_detail: "None" → just outfit, no detail
|
||||
|
||||
clean composition.
|
||||
|
||||
## current scale
|
||||
|
||||
17 categories × ~20 avg options = 340 selections
|
||||
|
||||
can scale to:
|
||||
17 categories × 100 options = 1700 selections
|
||||
|
||||
or add more categories:
|
||||
25 categories × 50 options = 1250 selections
|
||||
|
||||
system handles any scale.
|
||||
|
||||
## alphabetical advantage
|
||||
|
||||
with 100 options, alphabetical makes sense:
|
||||
|
||||
```
|
||||
A-D: 25 options
|
||||
E-H: 25 options
|
||||
I-L: 25 options
|
||||
M-Z: 25 options
|
||||
```
|
||||
|
||||
easy to scan.
|
||||
|
||||
no search needed if well-organized.
|
||||
|
||||
Reference in New Issue
Block a user