No length cap on the binds-to-image definition form
This commit is contained in:
parent
c2559f0572
commit
962c667418
@ -177,10 +177,10 @@ and sampler latent or the video can come out square.
|
|||||||
20 year old woman with blonde hair, blue eyes and a medium build" — age, hair,
|
20 year old woman with blonde hair, blue eyes and a medium build" — age, hair,
|
||||||
eyes, build; never clothing, never a nude/bare state); a character WITH a
|
eyes, build; never clothing, never a nude/bare state); a character WITH a
|
||||||
reference image gets "<Subject 1> binds to <Image 1> : what the image actually
|
reference image gets "<Subject 1> binds to <Image 1> : what the image actually
|
||||||
shows" (the writing LLM is multimodal — look at it); and a user refmod
|
shows" (the writing LLM is multimodal — look at it, no length cap); and a user
|
||||||
(`sam:&identity: ...; voice: ...&` in the story's characters block) is carried
|
refmod (`sam:&identity: ...; voice: ...&` in the story's characters block) is
|
||||||
through VERBATIM as "<Subject 1> identity: ...; voice: ..." — never rewritten,
|
carried through VERBATIM as "<Subject 1> identity: ...; voice: ..." — never
|
||||||
never trimmed, exempt from the length and clothing rules. Weave the head-to-toe
|
rewritten, never trimmed, exempt from every defs limit. Weave the head-to-toe
|
||||||
look into the action as the scene exposes it — hair, build, and face as they
|
look into the action as the scene exposes it — hair, build, and face as they
|
||||||
move, each garment where the action touches it, footwear where steps land —
|
move, each garment where the action touches it, footwear where steps land —
|
||||||
nothing is shared between scenes or from a prefix. Name every body part the
|
nothing is shared between scenes or from a prefix. Name every body part the
|
||||||
|
|||||||
@ -254,8 +254,8 @@ Shots are flat. A shot never contains a nested object.
|
|||||||
marks and carry the text through VERBATIM: "<Subject 1> identity: <their
|
marks and carry the text through VERBATIM: "<Subject 1> identity: <their
|
||||||
identity text>; voice: <their voice text>". Never rewrite, trim, reorder, or
|
identity text>; voice: <their voice text>". Never rewrite, trim, reorder, or
|
||||||
"improve" a refmod line — it is the user's tested wording and overrides every
|
"improve" a refmod line — it is the user's tested wording and overrides every
|
||||||
other rule about definitions, including the length cap and the
|
other rule about definitions (only the no-image form has a length cap, and no
|
||||||
clothing/skin-state ban.
|
form's limits ever apply to a refmod line).
|
||||||
Define <Subject 2> the same way when the story casts a second recurring person.
|
Define <Subject 2> the same way when the story casts a second recurring person.
|
||||||
Never define a label twice; never define a label you do not use.
|
Never define a label twice; never define a label you do not use.
|
||||||
2. "summary:" then "[reference generation]" — exactly this directive on every scene,
|
2. "summary:" then "[reference generation]" — exactly this directive on every scene,
|
||||||
|
|||||||
@ -368,7 +368,8 @@ The sound picture is its own element and lives in the overall_soundscape section
|
|||||||
carry the text through VERBATIM: "<Subject 1> identity: <their identity text>;
|
carry the text through VERBATIM: "<Subject 1> identity: <their identity text>;
|
||||||
voice: <their voice text>". Never rewrite, trim, reorder, or "improve" a
|
voice: <their voice text>". Never rewrite, trim, reorder, or "improve" a
|
||||||
refmod line — it is the user's tested wording and overrides every other rule
|
refmod line — it is the user's tested wording and overrides every other rule
|
||||||
about definitions, including the length cap and the clothing/skin-state ban.
|
about definitions (only the no-image form has a length cap, and no form's
|
||||||
|
limits ever apply to a refmod line).
|
||||||
Define <Subject 2> the same way when the outline casts a second person. Never
|
Define <Subject 2> the same way when the outline casts a second person. Never
|
||||||
define a label twice; never define a label you do not use.
|
define a label twice; never define a label you do not use.
|
||||||
2. "summary:" then "[reference generation]" — exactly this directive on every scene
|
2. "summary:" then "[reference generation]" — exactly this directive on every scene
|
||||||
|
|||||||
@ -32,12 +32,14 @@ REF2VA_RETENTION_MARKERS = {"fully_preserved", "partially_preserved",
|
|||||||
"attribute_transfer", "weak_reference"}
|
"attribute_transfer", "weak_reference"}
|
||||||
# Rule 7 - a subject_definitions line comes in exactly one of three forms:
|
# Rule 7 - a subject_definitions line comes in exactly one of three forms:
|
||||||
# 1. no reference image -> "<Subject N> is a <age> ... with <hair>, <eyes> and
|
# 1. no reference image -> "<Subject N> is a <age> ... with <hair>, <eyes> and
|
||||||
# <build>" - physical traits only: never clothing, never a nude/bare skin state
|
# <build>" - lean physical traits only (25-word cap): never clothing, never a
|
||||||
|
# nude/bare skin state
|
||||||
# 2. reference image -> "<Subject N> binds to <Image N> : <what the image
|
# 2. reference image -> "<Subject N> binds to <Image N> : <what the image
|
||||||
# actually shows>" - compact and faithful; the writer is multimodal
|
# actually shows>" - faithful to the image, NO length cap; the writer is
|
||||||
|
# multimodal
|
||||||
# 3. user refmod -> "<Subject N> identity: <verbatim>; voice: <verbatim>"
|
# 3. user refmod -> "<Subject N> identity: <verbatim>; voice: <verbatim>"
|
||||||
# - the user's own tested wording, carried through untouched (exempt from the
|
# - the user's own tested wording, carried through untouched (exempt from
|
||||||
# length cap and the clothing/skin-state bans)
|
# every no-image trait limit: length, clothing, skin state)
|
||||||
# In every form the full look is still woven into the [Shot 1] body.
|
# In every form the full look is still woven into the [Shot 1] body.
|
||||||
REF2VA_DEF_MAX_WORDS = 25
|
REF2VA_DEF_MAX_WORDS = 25
|
||||||
REF2VA_REFMOD_RE = re.compile(r"^identity\s*:", re.I)
|
REF2VA_REFMOD_RE = re.compile(r"^identity\s*:", re.I)
|
||||||
@ -186,13 +188,6 @@ def check_ref2va_prompt(lines, dur, where, prev_handoff=None):
|
|||||||
raise ValueError(f"{where}: {m.group(1)} binds line needs a description "
|
raise ValueError(f"{where}: {m.group(1)} binds line needs a description "
|
||||||
f"after the colon - look at <Image N> and write what "
|
f"after the colon - look at <Image N> and write what "
|
||||||
f"you see: {line[:60]!r}")
|
f"you see: {line[:60]!r}")
|
||||||
words = len(desc.split())
|
|
||||||
if words > REF2VA_DEF_MAX_WORDS:
|
|
||||||
raise ValueError(f"{where}: {m.group(1)} binds description is {words} "
|
|
||||||
f"words - describe what you actually see in <Image N> "
|
|
||||||
f"compactly (age, hair, eyes, build, what they wear in "
|
|
||||||
f"the image); the look in motion is still woven into "
|
|
||||||
f"the [Shot 1] body: {line[:60]!r}")
|
|
||||||
else:
|
else:
|
||||||
if not re.match(r"^is\b", anchor, re.I):
|
if not re.match(r"^is\b", anchor, re.I):
|
||||||
raise ValueError(f"{where}: {m.group(1)} definition is none of the three "
|
raise ValueError(f"{where}: {m.group(1)} definition is none of the three "
|
||||||
@ -762,7 +757,16 @@ def run_fragment_suite():
|
|||||||
noimage["prompt"][1] = ("<Subject 1> is a woman in her thirties with dark brown "
|
noimage["prompt"][1] = ("<Subject 1> is a woman in her thirties with dark brown "
|
||||||
"hair in a low bun, calm grey eyes and a slim build.")
|
"hair in a low bun, calm grey eyes and a slim build.")
|
||||||
validate_fragment(noimage, index=1, total=1)
|
validate_fragment(noimage, index=1, total=1)
|
||||||
print(" all ref2va PASS cases ok (binds, verbatim refmod, no-image traits)")
|
long_binds = _ref2va_short()
|
||||||
|
long_binds["prompt"][1] = ("<Subject 1> binds to <Image 1> : a woman in her "
|
||||||
|
"thirties with dark brown hair pulled back in a low "
|
||||||
|
"bun, calm grey eyes, a slim build, wearing a pale "
|
||||||
|
"green scrub top and matching scrub trousers, white "
|
||||||
|
"sneakers, and a clipped ID badge on a lanyard at her "
|
||||||
|
"chest.")
|
||||||
|
validate_fragment(long_binds, index=1, total=1)
|
||||||
|
print(" all ref2va PASS cases ok (binds uncapped, verbatim refmod, no-image "
|
||||||
|
"traits)")
|
||||||
|
|
||||||
print("\nRef2VA edge cases that must FAIL:")
|
print("\nRef2VA edge cases that must FAIL:")
|
||||||
no_ret = _ref2va_short()
|
no_ret = _ref2va_short()
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user