# Krea2 POV Pose Atlas

Local reference root:

`/media/unraid/davinci/Qwen_edit_lora/POV/dataset_v2`

Use this dataset as the pose-geometry reference for POV prompt tuning. The pose
folders contain rendered POV examples; matching `_control` folders contain the
solo/control image for the same pose family. Ignore `bg` and `*_bg` folders for
pose analysis; they are background plates without people.

Machine-readable pose variants live in
`categories/krea2_pov_pose_variants.json`. That catalog is intentionally smaller
than the full atlas: it only contains variants that are proven or useful
candidates for fixed-seed Krea2 tuning. Add a variant there when it has a compact
geometry summary, cue phrases, avoid phrases, references, and a known generator
hook. Code should read it through `krea2_pose_variant_catalog.py` instead of
parsing the JSON directly.

In ComfyUI, use the `SxCP Krea2 Pose Variant` node when you want a workflow to
select one catalog variant and emit a compatible `hardcore_position_config` for
the existing Position Pool / Action Filter / Insta-OF chain. Pair it with
`SxCP Krea2 Variant Evidence` to display the fixed-seed eval entry, image paths,
and generator decision behind that variant.

For command-line planning, `python tools/krea2_tuning_report.py` shows which
catalog variants are proven or pending and which atlas pose folders are still
unmapped by the catalog. Unmapped folders include sample pose/control image
paths and a suggested candidate key to start the next catalog entry.

## Inventory

| Family | Pose images | Control images | First sample |
| --- | ---: | ---: | --- |
| cowgirl | 63 | 63 | `5.cowgirl/100_cowgirl.png` |
| cowgirl alt | 62 | 62 | `5.cowgirl_alt/101_cowgirl_alt.png` |
| reverse cowgirl | 58 | 58 | `cowgirl_reverse/101_cowgirl_reverse.png` |
| reverse cowgirl alt | 50 | 50 | `cowgirl_reversere_alt/100_cowgirl_reversere_alt.png` |
| doggy | 57 | 57 | `doggy/101_doggy.png` |
| doggy alt | 45 | 45 | `doggy_alt/100_doggy_alt.png` |
| missionary | 74 | 74 | `missionary/101_missionary.png` |
| missionary folded | 12 | 12 | `missionary_folded/16_missionary_folded.png` |
| sixty-nine | 29 | 29 | `69/105_sixtynine.png` |
| ballsucking | 25 | 25 | `ballsucking/101_ballsucking.png` |
| blowjob laying | 42 | 42 | `blowjob_laying/101_blowjob_laying.png` |
| blowjob side | 17 | 17 | `blowjob_side/103_blowjob_side.png` |
| blowjob sitting | 27 | 27 | `blowjob_sitting/100_blowjob_sitting.png` |
| blowjob top view | 17 | 17 | `blowjob_top_view/102_blowjob_top_view.png` |
| boobjob | 11 | 11 | `boobjob/100_boobjob.png` |
| handjob | 24 | 24 | `handjob/18_handjob.png` |
| footjob | 2 | 2 | `footjob/59_footjob.png` |
| fingering | 10 | 10 | `fingering/103_fingering.png` |
| spread | 55 | 55 | `spread/100_spread_.png` |
| ready | 19 | 19 | `ready/105_ready_.png` |
| wand | 7 | 7 | `wand/106_wand_.png` |

## Tuning Method

For each pose family:

1. Sample 5-10 pose images and 2-3 control images.
2. Write a compact geometry summary using only repeated visual facts.
3. Test one prompt variant with a fixed seed.
4. Test the same wording on a second seed or character.
5. Patch generator defaults only when the wording improvement repeats or the
   generated prompt is structurally wrong before rendering.
6. Record the evidence in `docs/krea2-prompt-guide.md`.

## Confirmed Notes

### Doggy / Rear-Entry

Dataset references show that visible POV thighs, lower torso, or pelvis can be
correct. They should be treated as natural foreground cues, not automatic
failures.

Better Krea2 wording:

- `top-down POV doggy position from behind`
- `camera looks down over the viewer's hands onto the woman's raised hips`
- `woman is on all fours with chest low, forearms folded, cheek turned sideways`
- `back arched, hips raised high toward the camera`
- `viewer hands hold her hips with natural lower-body POV cues in the foreground`

Avoid using visible shoes or lower legs as the standing cue. In seed `65`, that
wording pulled Krea2 toward oral contact and weakened rear-entry geometry.

### Boobjob / Titjob

The boobjob folder shows a repeated upright, frontal geometry rather than a
forward-bent one: the woman faces the viewer between his thighs, breasts pressed
together around a vertical shaft, with the glans above the cleavage near her
mouth. For Krea2, name hand ownership when hands matter. In POV prompts, generic
`hands` can become the viewer's hands.

### Handjob

The handjob folder repeats a centered first-person layout: the viewer's thighs
frame the lower edges, the woman faces the viewer between his legs, and her hand
is the contact anchor on the shaft. Prompt the woman's hand ownership directly;
viewer hands should not cover the action unless that is the intended variant.

## Candidate Notes

### Footjob

The footjob folder is small but visually consistent: the viewer reclines with
thighs framing the lower foreground, the penis is upright near the center, and
the woman's soles/toes are the contact anchor while her body and face remain
behind the feet. Treat `pov_footjob_frontal_sole_stroke` as a candidate until it
has fixed-seed Krea2 evidence.

### Fingering

The fingering folder repeats a first-person manual-contact layout: the woman is
reclined or sitting back with thighs spread wide toward camera, her face and
torso visible behind the open-leg frame, and the viewer hand entering from the
foreground as the contact anchor. Treat `pov_fingering_reclined_open_thighs` as
a candidate until it has fixed-seed Krea2 evidence.

### Spread / Open-Thigh Presentation

The spread folder is a setup/presentation family rather than a required contact
action: the woman faces the camera with legs raised or knees held wide, thighs
forming a wide V-frame, and her face and torso visible behind the open-leg pose.
Treat `pov_spread_open_thigh_presentation` as a candidate until it has
fixed-seed Krea2 evidence.