|
The dress frames are made. What the tries taught is in frames.txt beside
each line: klein turns the blindfold into an eye mask under strong
expressions, so those take it off (BARE) and the rest name it as one solid
band (EYES); standing "shoved" read as walking in four seeds, so it is
airborne, edited from the seed-7 attempt kept in candidates/; a back view
fits "watch", her watching the screen while the user types; klein would
neither draw the game's Virtuous Treaty nor move a sword, nor turn her
three-quarters, so muse takes only the arms of her official art.
gen-frames now takes several references per frame (a + b): the first is
edited, the rest are images the prompt points at. The leotard set loses
the blindfold too. shoved gets a frame of its own instead of mad: in the
spec's table and the plan's Mood.js and its test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
|
The frames are a list now, not a session in the web UI: frames.txt holds
one line per frame (output, the image klein edits, the prompt, with
shared sentences as variables), and gen-frames makes the missing ones,
or the ones named, through the running sd-server's job API
(/sdcpp/v1/img_gen, polled at /sdcpp/v1/jobs/<id>). sd-cli would reload
the model for every frame. Adding or changing a frame is an edit and a
rerun; --seed rerolls one. Then prep-frames cuts each out of its white
background at one common scale.
Checked on dress/smile: about 2 minutes a frame on klein, pose and
outfit kept.
The art stays outside the repo, in ~/.local/share/keystone/companion.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
