Kevin Gilper
f7061c7057
Traced the real provenance (Kevin's Codex dev sessions) and verified the
mechanism against the Krea 2 model on the render box. Two earlier claims
(07d2f0b) were wrong and are now fixed across the docs:
1. Mechanism: the 12 "deepstack chunks" are the 12 TEXT-encoder layer taps
Krea 2 conditions on (comfy/text_encoders/krea2.py: hidden_states
[2,5,8,...,35], (B,12,seq,2560) flattened to 30720), NOT the 3 vision
deepstack taps (qwen3vl.py deepstack_visual_indexes=[5,11,17]) the name
implies. 30720 % 12 == 0, so the node's split lands on the taps and the
per-layer gains do apply (no fallback) - confirmed from the model weights
(txtfusion.projector [1,12], layerwise blocks at 2560).
2. Provenance: the 12 values were NOT derived by a per-chunk sweep. They
were adopted from the third-party ComfyUI-ConditioningKrea2Rebalance node
(DEFAULT_WEIGHTS spike 8>10>7, matching STYLE_LAYER_PULL) plus the
shallow=structure/deep=appearance principle; only the scalar shape (0.45
->0.35) and cap (->0.9) were empirically tuned. The four role tables are
scalings of one template, so their agreement is design self-consistency,
not independent measurement - the "convergent evidence" framing is
withdrawn.
docs/deepstack-layers/ rewritten to separate verified mechanism, traced
provenance, and the still-unrun single-tap sweep. Corrections propagated to
the V9 technical paper (5.2 + architecture note + glossary), the V10 guide
and companion, custom_recipes/README, the V10 node doc, and CHANGELOG. Full
evidence chain preserved in local_records (gitignored).
Documentation
Start here if you want to understand, install, or extend Krea Reference.
Choose A Path
| I want to... | Read this |
|---|---|
| See what the nodes can do | V10 visual HTML guide or V10 Markdown guide |
| Learn the core mental model | V9 visual HTML guide or V9 Markdown guide |
| Load a workflow quickly | Example workflows |
| Understand every recipe | Recipe section in the V9 Markdown guide plus the new V10 recipes |
| Configure a guide card | Guide Card V10 or Guide Card V9 |
| Configure the stack encoder | Stack Encoder V10 or Stack Encoder V9 |
| Understand how and why it works, or extend it | V9 technical paper (shared architecture) and V10 technical companion |
| Run maintainer checks | Testing guide |
Core Guides
- Krea 2 V10 documentation index
- Krea 2 V10 visual user guide
- Krea 2 V10 Markdown user guide
- Krea 2 V10 technical companion
- Krea 2 V9 documentation index
- Krea 2 V9 visual user guide
- Krea 2 V9 Markdown user guide
- Krea 2 V9 technical paper
- Node documentation
- Example workflows
Included Node Pages
- KG Krea 2 Image Guide Card V10
- KG Krea 2 Reference Stack Encoder V10
- KG Krea 2 Image Guide Card V9
- KG Krea 2 Reference Stack Encoder V9
Demo Assets
- V10 demo journeys with embedded workflows - input, settings, prompt, and result per demo
- V9 recipe demo PNGs with embedded workflows
- Synthetic source images