better than v1.0?
This commit is contained in:
transcendedhacker
2026-01-17 21:02:22 +05:30
parent 42fcd9c7a3
commit 232cb8b8bb
10 changed files with 780 additions and 197 deletions
+40 -21
View File
@@ -1,37 +1,56 @@
# notes
# dev notes
## v2.0 - massive scale
comprehensive library with 300+ atomic options across 17 categories.
## structure
atomic parts in libs → schema defines assembly → engine composes
atomic selections → alphabetically sorted dropdowns → "None" skips blocks → composed output
no bloat in dropdowns. each option = one thing.
## categories
## adding libs
subject → age → hair_color → hair_style → hair_ornament → outfit → outfit_detail → pose → mood → location → location_detail → atmosphere → lighting → angle → frame → filter → technical
1. drop json in libraries/
2. match block names in schema order
3. restart comfy
## "None" behavior
## schema tweaks
selecting "None" in any dropdown = blank/skip that block entirely.
weights in connector.json apply to all libs.
break positions = where to insert BREAK.
prompt only includes selected options.
## model profiles
## alphabetical sorting
sdxl: weight_multiplier 1.0
flux: weight_multiplier 0.85 (lighter)
all dropdowns auto-sorted alphabetically (except "None" which stays on top).
makes options easy to find without search.
## scaling
current: 300+ options
can scale to 1000+ by adding more libs
each lib is independent, composer handles all.
## perf
clean atomic composition should keep gen <100s.
if slow: check for malformed prompts or excessive BREAK tokens.
## todo
- [ ] add more hair options from reference
- [ ] parse outfit list properly
- [ ] atmosphere lib needs work
- [ ] maybe add location block
- add more angles from AngleShot.txt (100+ options)
- expand outfits from Outfits.txt + RevealingOutfit.txt
- add all poses from SexyPoses.txt
- parse LocationsIndoor.txt for location block
- add all atmosphere variations
- all filters from Filters.txt
- lighting effects from full frame light effect.txt
## perf notes
## hierarchical dependency (future)
gen time should be <100s with clean prompts.
if slow, check for malformed output or clip issues.
comfyui limitation: dropdowns can't dynamically change based on other selections.
old system was hitting 300s - too much crap in prompt.
workaround: use comprehensive libs that cover all combos.
or: create specialized libs (portrait_casual, portrait_formal, etc) and switch library.