diff --git a/h3-plan-gate/SKILL.md b/h3-plan-gate/SKILL.md index a04279d..fe3ccd8 100755 --- a/h3-plan-gate/SKILL.md +++ b/h3-plan-gate/SKILL.md @@ -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, eyes, build; never clothing, never a nude/bare state); a character WITH a reference image gets " binds to : what the image actually - shows" (the writing LLM is multimodal — look at it); and a user refmod - (`sam:&identity: ...; voice: ...&` in the story's characters block) is carried - through VERBATIM as " identity: ...; voice: ..." — never rewritten, - never trimmed, exempt from the length and clothing rules. Weave the head-to-toe + shows" (the writing LLM is multimodal — look at it, no length cap); and a user + refmod (`sam:&identity: ...; voice: ...&` in the story's characters block) is + carried through VERBATIM as " identity: ...; voice: ..." — never + 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 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 diff --git a/h3-plan-gate/h3_plan_json_system_prompt.md b/h3-plan-gate/h3_plan_json_system_prompt.md index 1b20ec5..83e72ee 100755 --- a/h3-plan-gate/h3_plan_json_system_prompt.md +++ b/h3-plan-gate/h3_plan_json_system_prompt.md @@ -254,8 +254,8 @@ Shots are flat. A shot never contains a nested object. marks and carry the text through VERBATIM: " identity: ; voice: ". Never rewrite, trim, reorder, or "improve" a 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. + other rule about definitions (only the no-image form has a length cap, and no + form's limits ever apply to a refmod line). Define the same way when the story casts a second recurring person. Never define a label twice; never define a label you do not use. 2. "summary:" then "[reference generation]" — exactly this directive on every scene, diff --git a/h3-plan-gate/h3_scene_agent_system_prompt.md b/h3-plan-gate/h3_scene_agent_system_prompt.md index 8f83493..657b67c 100755 --- a/h3-plan-gate/h3_scene_agent_system_prompt.md +++ b/h3-plan-gate/h3_scene_agent_system_prompt.md @@ -368,7 +368,8 @@ The sound picture is its own element and lives in the overall_soundscape section carry the text through VERBATIM: " identity: ; voice: ". Never rewrite, trim, reorder, or "improve" a 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 the same way when the outline casts a second person. Never define a label twice; never define a label you do not use. 2. "summary:" then "[reference generation]" — exactly this directive on every scene diff --git a/h3-plan-gate/verify_plan_rules.py b/h3-plan-gate/verify_plan_rules.py index ba01191..6111b2a 100755 --- a/h3-plan-gate/verify_plan_rules.py +++ b/h3-plan-gate/verify_plan_rules.py @@ -32,12 +32,14 @@ REF2VA_RETENTION_MARKERS = {"fully_preserved", "partially_preserved", "attribute_transfer", "weak_reference"} # Rule 7 - a subject_definitions line comes in exactly one of three forms: # 1. no reference image -> " is a ... with , and -# " - physical traits only: never clothing, never a nude/bare skin state +# " - lean physical traits only (25-word cap): never clothing, never a +# nude/bare skin state # 2. reference image -> " binds to : " - compact and faithful; the writer is multimodal +# actually shows>" - faithful to the image, NO length cap; the writer is +# multimodal # 3. user refmod -> " identity: ; voice: " -# - the user's own tested wording, carried through untouched (exempt from the -# length cap and the clothing/skin-state bans) +# - the user's own tested wording, carried through untouched (exempt from +# every no-image trait limit: length, clothing, skin state) # In every form the full look is still woven into the [Shot 1] body. REF2VA_DEF_MAX_WORDS = 25 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 " f"after the colon - look at and write what " 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 " - 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: if not re.match(r"^is\b", anchor, re.I): raise ValueError(f"{where}: {m.group(1)} definition is none of the three " @@ -762,7 +757,16 @@ def run_fragment_suite(): noimage["prompt"][1] = (" is a woman in her thirties with dark brown " "hair in a low bun, calm grey eyes and a slim build.") 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] = (" binds to : 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:") no_ret = _ref2va_short()