Skip to content

C
coderai
Commit e28f11e9 authored by Stefy Lanza (nextime / spora )'s avatar Stefy Lanza (nextime / spora )
Browse files
Options

township: pre-fight intro clips, per-clip prompt regen, whole-match 2x enhance 

- Every match now opens with 3 intro clips before the fight: a bold solo
  entrance for each fighter, then a referee-officiated face-off stare-down
  with the start signal. The real fight begins at clip 4. New intro prompt
  templates + LLM system prompt + PromptGenerator.intro_shot().
- Factor out _build_match_clip_specs() and _fill_clip_prompt() so all four
  clip-building paths (stage_videos Phase A, new-match, replan, full regen)
  build intros consistently; intro clips attach only their own participants
  (solo entrance = one fighter; face-off = both + match referee).
- New "clip-prompt" job scope + per-clip "prompt↻" link: rewrite ONLY one
  clip's prompt in place (role-aware), steering fight clips away from the
  match's other shots; renders nothing.
- "Create whole match" and "Regenerate whole match" (scope "full") now finish
  with a 2x AI upscale + 2x frame interpolation pass over the final short/long
  assemblies and outcome videos, reusing the existing enhance machinery.
Co-Authored-By: 's avatarClaude Opus 4.8 <noreply@anthropic.com>
parent 3bfefed0
Show whitespace changes
Inline Side-by-side
Showing with 343 additions and 103 deletions
tools/gen_township_fighters.py
View file @ e28f11e9
......@@ -378,6 +378,28 @@ FIGHT_PROMPT_SUFFIX = ("African township free fight, fast-paced, rapid explosive
"reverses, rewinds or loops back, cinematic, consistent "
"characters, wardrobe and setting")
# Pre-fight intro clips: the OPENING of a match is NOT a fight — two solo fighter
# entrances followed by a stare-down where the referee signals the start. These
# carry a non-fight suffix so the model doesn't start throwing strikes early.
INTRO_PROMPT_SUFFIX = ("African township fight night, cinematic, dramatic, scenic, "
"atmospheric, consistent characters, wardrobe and setting, "
"tension before the fight, NO punches or kicks thrown yet")
ENTRANCE_SHOT_TEMPLATES = [
"striding into the arena through a parting roaring crowd, arms raised and chin high, slow heroic low-angle push-in through dust and spotlight haze",
"bursting onto the scene pounding their own chest and roaring at the crowd, sweeping crane shot rising over the packed yard",
"stalking in slowly, cracking their knuckles and rolling their shoulders with a menacing glare into the lens, tight dramatic push-in",
"leaping down off a low wall into the fighting ground and throwing their arms wide to the screaming crowd, dynamic dropping crane shot",
"marching through smoke and fire-barrel glow shadow-boxing a few sharp strikes, kinetic tracking shot circling the entrance",
"emerging from the crowd with a slow swagger and pointing to the centre of the ground, golden dusk backlight, sweeping orbit",
]
FACEOFF_SHOT_TEMPLATES = [
"standing face to face foreheads almost touching in a tense menacing stare-down, the referee stepping between them and chopping a hand down to START the fight, crowd electric, slow orbit snapping to a push-in",
"squaring off chest to chest trading furious words, the referee splitting them apart then sweeping an arm to signal FIGHT, dramatic low-angle shot, crowd roaring",
"locked in a furious eye-to-eye stare-down with fists clenched, the referee raising a hand between them and dropping it to launch the bout, tight handheld push-in",
]
def _continuity_clause(env_name: str) -> str:
"""Deterministic wardrobe + environment continuity phrase appended to EVERY
......@@ -1061,6 +1083,15 @@ location as the fight, with consistent background, surfaces and lighting. Never
to a different place.
Return ONLY the prompt, no quotes or explanation."""
_LLM_INTRO_SYSTEM = """\
You are a creative director writing vivid 18-32 word video-generation prompts for the DRAMATIC OPENING of an African township fight night — BEFORE any fighting starts.
You write ONE of these shot types at a time (the user says which):
• an ENTRANCE — ONE fighter ALONE making a bold, scenic, dramatic entrance into the arena: striding or bursting in, playing to the roaring crowd, flexing or shadow-boxing, glaring at the camera. NO opponent in frame and NO actual fighting. Make it a hero shot with a kinetic camera move (low-angle push-in, sweeping crane, orbit) and rich atmosphere (dust, smoke, fire-barrel glow, spotlights).
• a FACE-OFF — BOTH fighters squaring off face to face in a tense menacing stare-down, with the REFEREE between them giving the gesture to START the fight (chopping or dropping a hand, sweeping an arm). Show the tension and the referee's start signal explicitly. Still NO punches or kicks thrown yet.
Always refer to each fighter (and the referee) by their NAME given in the user message.
WARDROBE + LOCATION CONTINUITY (critical): keep each fighter in the IDENTICAL outfit (same garments, exact colours, hair, accessories) described, in the SAME township location with consistent background, surfaces, lighting and crowd.
Do NOT use generic phrases like "high quality" or "realistic". Return ONLY the prompt, no quotes."""
# Snapshot the built-in defaults now that every template/system prompt is
# defined (before any saved override is applied), so the web "Prompts" page can
# offer a reset-to-defaults.
......@@ -1077,6 +1108,7 @@ class PromptGenerator:
self.char_descriptions: dict = char_descriptions or {}
self._used_fight: list[str] = []
self._used_outcome: dict[str, list[str]] = {}
self._used_intro: dict[str, list[str]] = {}
def fight_shot(self, f1: str, f2: str, env_desc: str, match_context: str = "",
avoid: list = None, action_focus: str = "") -> str:
......@@ -1210,6 +1242,65 @@ class PromptGenerator:
used.append(choice)
return choice
def intro_shot(self, role: str, f1: str, f2: str = None, env_desc: str = "",
referee: str = None) -> str:
"""Generate a pre-fight INTRO shot prompt for one opening clip of a match.
`role`:
• "entrance" → ONE fighter (f1) alone making a bold, scenic solo entrance.
• "faceoff" → f1 vs f2 stare-down with the referee signalling START.
No fighting happens in these clips — they set the scene before the bout.
"""
is_face = role == "faceoff"
src = FACEOFF_SHOT_TEMPLATES if is_face else ENTRANCE_SHOT_TEMPLATES
used = self._used_intro.setdefault(role, [])
if self.model:
if is_face:
_d1 = self.char_descriptions.get(f1, "")
_d2 = self.char_descriptions.get(f2, "")
f1d = f"{f1} ({_d1})" if _d1 else f1
f2d = f"{f2} ({_d2})" if _d2 else f2
ref_line = f" Referee: {referee}." if referee else " A referee."
label = (f"a FACE-OFF stare-down between {f1d} and {f2d} with the "
"referee giving the signal to START the fight")
user_extra = f"Fighter 1: {f1d}. Fighter 2: {f2d}.{ref_line} "
else:
_d1 = self.char_descriptions.get(f1, "")
f1d = f"{f1} ({_d1})" if _d1 else f1
label = f"a bold, dramatic, scenic solo ENTRANCE of {f1d}"
user_extra = f"Fighter: {f1d}. "
_avoid = used[-2:]
for attempt in range(2):
try:
used_hint = f" Avoid: {'; '.join(_avoid)}." if _avoid else ""
prompt = self.client.chat_complete(
model=self.model,
system=_LLM_INTRO_SYSTEM,
user=(
f"{user_extra}Shot to write: {label}. Location: {env_desc}."
f"{used_hint} Write this ONE intro shot prompt. Refer to the "
"fighters (and referee) by their NAME."
),
max_tokens=110,
).strip()
if len(prompt) < 8:
raise ValueError(f"LLM returned too-short response: {prompt!r}")
used.append(prompt[:60])
return prompt
except Exception as e:
if attempt == 0:
_avoid = [] # drop the hint and retry once
continue
_log(f" (LLM intro prompt failed after retry: {e} — using template)")
available = [t for t in src if t not in used]
if not available:
available = src
choice = random.choice(available)
used.append(choice)
return choice
# ─────────────────────────────────────────────────────────────────────────────
# Form auto-generation — invent a fresh character / referee / environment (or a
......@@ -2117,6 +2208,71 @@ def _fighter_desc_hint(name: str, char_descriptions: dict) -> str:
return f"{name} ({visual})"
def _build_match_clip_specs(fps: int, cf_lo: int, cf_hi: int,
long_target: float, f1: str, f2: str) -> list:
"""Build a match's ordered clip specs: the pre-fight INTRO (a solo entrance for
each fighter, then a referee-officiated face-off), followed by fight clips until
`long_target` playback seconds is reached. The intro clips are extra and do NOT
count toward `long_target`. Prompts are filled later by `_fill_clip_prompt`."""
specs, ci = [], 0
for _role, _who in (("entrance", [f1]), ("entrance", [f2]), ("faceoff", [f1, f2])):
nf = random.randint(cf_lo, cf_hi)
specs.append({"idx": ci, "clip_seconds": round(nf / max(1, fps), 2),
"nf": nf, "intensity": "introduction", "role": _role,
"fighters": list(_who), "shot": None, "prompt": None})
ci += 1
planned = 0.0
while planned < long_target:
round_num = (ci - 3) // 3 + 1
intensity = ("early exchanges" if round_num == 1
else "midpoint battle" if round_num == 2
else "climactic final exchange")
nf = random.randint(cf_lo, cf_hi)
specs.append({"idx": ci, "clip_seconds": round(nf / max(1, fps), 2),
"nf": nf, "role": "fight", "intensity": intensity,
"shot": None, "prompt": None})
planned += nf / max(1, fps)
ci += 1
return specs
def _fill_clip_prompt(prompter, c: dict, f1: str, f2: str, env, env_desc: str,
char_descriptions: dict, referee: str = None,
match_avoid: list = None, focus: str = "") -> str:
"""Populate `c['shot']` and `c['prompt']` for one clip according to its role.
role "entrance" → a solo intro for the clip's single fighter; "faceoff" → a
stare-down between both fighters with the referee signalling START; anything
else → a normal fight shot (appended to `match_avoid` so the match stays varied).
"""
role = c.get("role", "fight")
cont = _continuity_clause(env or None)
if role == "entrance":
who = (c.get("fighters") or [f1])[0]
shot = prompter.intro_shot("entrance", who, env_desc=env_desc)
hint = _fighter_desc_hint(who, char_descriptions)
c["shot"] = shot
c["prompt"] = f"{hint} — {shot} — {cont} — {INTRO_PROMPT_SUFFIX}"
elif role == "faceoff":
shot = prompter.intro_shot("faceoff", f1, f2, env_desc=env_desc, referee=referee)
h1 = _fighter_desc_hint(f1, char_descriptions)
h2 = _fighter_desc_hint(f2, char_descriptions)
c["shot"] = shot
c["prompt"] = f"{h1} vs {h2} — {shot} — {cont} — {INTRO_PROMPT_SUFFIX}"
else:
shot = prompter.fight_shot(
f1, f2, env_desc,
match_context=f"Match stage: {c.get('intensity', '')}. ",
avoid=match_avoid, action_focus=focus)
h1 = _fighter_desc_hint(f1, char_descriptions)
h2 = _fighter_desc_hint(f2, char_descriptions)
c["shot"] = shot
c["prompt"] = f"{h1} vs {h2} — {shot} — {cont} — {FIGHT_PROMPT_SUFFIX}"
if match_avoid is not None:
match_avoid.append(shot[:60])
return c["shot"]
def _write_concat(clips: list, out_path: str, label: str):
paths = [p for p, _ in clips]
if not paths:
......@@ -2831,8 +2987,21 @@ def _generate_keyframes(client: CoderAIClient, image_model: str, keyframe_dir: P
jobs = []
for m in fight_plan:
for c in m["clips"]:
# Intro clips appear with only their own participants: a solo entrance
# shows ONE fighter; the face-off shows both fighters + the referee.
# Fight clips (and legacy clips with no role) show both fighters.
_role = c.get("role", "fight")
if _role in ("entrance", "faceoff"):
_cf = list(c.get("fighters") or [m["f1"], m["f2"]])
if _role == "faceoff":
_cref = (m.get("referee")
or _referee_for(_out_dir, m.get("match_name") or m.get("f1", "")))
if _cref and _cref not in _cf:
_cf.append(_cref)
else:
_cf = [m["f1"], m["f2"]]
jobs.append((_clip_stem_fight(m["match_name"], c["idx"]),
c["prompt"], [m["f1"], m["f2"]], m.get("env"),
c["prompt"], _cf, m.get("env"),
m.get("env_desc"), c.get("kf_prompt")))
for o in outcome_plan:
_of = _mf_map.get(o.get("match_name")) or [o["fighter"]]
......@@ -3092,8 +3261,25 @@ def stage_videos(client: CoderAIClient, video_model: str, out_dir: Path,
env = random.choice(env_names) if env_names else None
env_desc = _env_description(env) if env else "African township"
clips_spec, planned, ci = [], 0.0, 0
# Pre-fight INTRO clips first: a bold solo entrance for EACH fighter, then a
# stare-down face-off where the referee signals the start. The real fight
# begins only after these. `fighters` records who appears in the clip so the
# keyframe + render attach the right profiles (the face-off also gets the
# referee, resolved at prompt/render time). These clips are extra — they do
# NOT count against the long_target so the fight still fills the cut.
for _intro_role, _intro_who in (("entrance", [f1]),
("entrance", [f2]),
("faceoff", [f1, f2])):
_nf = random.randint(_cf_lo, _cf_hi)
clips_spec.append({
"idx": ci, "clip_seconds": round(_nf / max(1, fps), 2),
"nf": _nf, "intensity": "introduction",
"role": _intro_role, "fighters": list(_intro_who),
"shot": None, "prompt": None,
})
ci += 1
while planned < long_target:
round_num = ci // 3 + 1
round_num = (ci - 3) // 3 + 1
intensity = ("early exchanges" if round_num == 1
else "midpoint battle" if round_num == 2
else "climactic final exchange")
......@@ -3103,7 +3289,7 @@ def stage_videos(client: CoderAIClient, video_model: str, out_dir: Path,
clip_seconds = round(_nf / max(1, fps), 2)
clips_spec.append({
"idx": ci, "clip_seconds": clip_seconds,
"nf": _nf,
"nf": _nf, "role": "fight",
"intensity": intensity, "shot": None, "prompt": None,
})
# Accumulate playback duration so the match reaches long_target.
......@@ -3166,22 +3352,22 @@ def stage_videos(client: CoderAIClient, video_model: str, out_dir: Path,
# disciplines (kicks, clinch, ground, submissions…) instead of all punches.
_focus_cycle = list(FIGHT_ACTION_FOCUS)
random.shuffle(_focus_cycle)
for _ci, c in enumerate(m["clips"]):
_ref = m.get("referee") or _referee_for(out_dir, m.get("match_name") or m["f1"])
_ffi = 0 # fight-clip index (drives the technique-focus cycle, skips intros)
for c in m["clips"]:
_pidx += 1
shot = prompter.fight_shot(
m["f1"], m["f2"], m["env_desc"],
match_context=f"Match stage: {c['intensity']}. ",
avoid=match_avoid,
action_focus=_focus_cycle[_ci % len(_focus_cycle)])
c["shot"] = shot
f1_hint = _fighter_desc_hint(m['f1'], char_descriptions)
f2_hint = _fighter_desc_hint(m['f2'], char_descriptions)
c["prompt"] = (
f"{f1_hint} vs {f2_hint} — {shot} — {_continuity_clause(m.get('env'))} "
f"— {FIGHT_PROMPT_SUFFIX}"
)
match_avoid.append(shot[:60])
_log(f" │ [{_pidx}/{_ptot}] {m['f1']} vs {m['f2']} clip{c['idx']:02d}: {shot}")
_role = c.get("role", "fight")
_focus = ("" if _role != "fight"
else _focus_cycle[_ffi % len(_focus_cycle)])
if _role == "fight":
_ffi += 1
shot = _fill_clip_prompt(prompter, c, m["f1"], m["f2"], m.get("env"),
m["env_desc"], char_descriptions, referee=_ref,
match_avoid=match_avoid, focus=_focus)
_who = (f"{(c.get('fighters') or [m['f1']])[0]} entrance" if _role == "entrance"
else f"{m['f1']} vs {m['f2']} face-off" if _role == "faceoff"
else f"{m['f1']} vs {m['f2']}")
_log(f" │ [{_pidx}/{_ptot}] {_who} clip{c['idx']:02d}: {shot}")
for o in outcome_plan:
_pidx += 1
_plan_outcome_shots(prompter, o, char_descriptions, o.get("opponent"))
......@@ -3284,13 +3470,17 @@ def _stage_videos_render(client, video_model, video_dir, fight_plan, outcome_pla
# fighter the scene is named after. Read from the saved plan too, so this
# still resolves when re-rendering outcomes alone (fight_plan == []).
_mf_map = {}
_mref_map = {}
_out_dir = Path(video_dir).parent.parent
for _m in fight_plan:
_mf_map[_m.get("match_name")] = [x for x in (_m.get("f1"), _m.get("f2")) if x]
_mref_map[_m.get("match_name")] = _m.get("referee")
try:
_saved = json.loads((Path(video_dir) / "prompts.json").read_text())
for _m in _saved.get("fight_plan", []):
_mf_map.setdefault(_m.get("match_name"),
[x for x in (_m.get("f1"), _m.get("f2")) if x])
_mref_map.setdefault(_m.get("match_name"), _m.get("referee"))
except Exception:
pass
......@@ -3560,13 +3750,28 @@ def _stage_videos_render(client, video_model, video_dir, fight_plan, outcome_pla
# into ≤single-render-cap parts when it exceeds one model call.
_nf = int(c.get("nf") or frames_for_seconds(c["clip_seconds"], 8))
_nf = min(_nf, MAX_PLANNED_FRAMES)
# Intro clips appear with only their own participants (solo entrance =
# one fighter; face-off = both fighters + referee) so the right profiles
# are attached; fight/legacy clips use both fighters.
_role = c.get("role", "fight")
if _role in ("entrance", "faceoff"):
_cf = list(c.get("fighters") or [m["f1"], m["f2"]])
if _role == "faceoff":
_cref = (_mref_map.get(m.get("match_name"))
or _referee_for(_out_dir, m.get("match_name") or m.get("f1", "")))
if _cref and _cref not in _cf:
_cf.append(_cref)
_clabel = f"{_role} clip {c['idx']:02d} — {'/'.join(_cf)}"
else:
_cf = [m["f1"], m["f2"]]
_clabel = f"clip {c['idx']:02d} — {m['f1']} vs {m['f2']}"
_log(f" │ clip {c['idx']:02d} {c['clip_seconds']:.1f}s → {_nf}f @ {fps}fps "
f"= {_nf/max(1,fps):.1f}s")
_clip("start")
ok, dur, is_fatal = _render(
f"clip {c['idx']:02d} — {m['f1']} vs {m['f2']}",
c["prompt"], [m["f1"], m["f2"]], m["env"], _nf, str(clip_path),
stem=clip_stem, fighters=[m["f1"], m["f2"]], step_cb=_step)
_clabel,
c["prompt"], _cf, m["env"], _nf, str(clip_path),
stem=clip_stem, fighters=_cf, step_cb=_step)
if is_fatal:
fatal = True
_clip("end", False)
......@@ -4297,37 +4502,23 @@ def launch_web_ui(default_args):
getattr(default_args, "clip_min_frames", CLIP_MIN_FRAMES),
getattr(default_args, "clip_max_frames", CLIP_MAX_FRAMES))
long_target = float(m["long_target"])
new_clips, planned, ci = [], 0.0, 0
while planned < long_target:
round_num = ci // 3 + 1
intensity = ("early exchanges" if round_num == 1
else "midpoint battle" if round_num == 2
else "climactic final exchange")
_nf = random.randint(_cf_lo, _cf_hi)
new_clips.append({"idx": ci, "clip_seconds": round(_nf / max(1, fps), 2),
"nf": _nf, "intensity": intensity,
"shot": None, "prompt": None})
planned += _nf / max(1, fps)
ci += 1
f1_hint = _fighter_desc_hint(f1, char_descriptions)
f2_hint = _fighter_desc_hint(f2, char_descriptions)
new_clips = _build_match_clip_specs(fps, _cf_lo, _cf_hi, long_target, f1, f2)
_ref = m.get("referee") or _referee_for(out_dir, name)
match_avoid = []
_focus_cycle = list(FIGHT_ACTION_FOCUS)
random.shuffle(_focus_cycle)
_ffi = 0
for i, c in enumerate(new_clips):
if _cancelled():
_fail("cancelled before prompts were complete")
return
shot = prompter.fight_shot(
f1, f2, env_desc,
match_context=f"Match stage: {c['intensity']}. ",
avoid=match_avoid,
action_focus=_focus_cycle[i % len(_focus_cycle)])
c["shot"] = shot
c["prompt"] = (f"{f1_hint} vs {f2_hint} — {shot} "
f"— {_continuity_clause(env or None)} "
f"— {FIGHT_PROMPT_SUFFIX}")
match_avoid.append(shot[:60])
_focus = ("" if c.get("role", "fight") != "fight"
else _focus_cycle[_ffi % len(_focus_cycle)])
if c.get("role", "fight") == "fight":
_ffi += 1
_fill_clip_prompt(prompter, c, f1, f2, env, env_desc,
char_descriptions, referee=_ref,
match_avoid=match_avoid, focus=_focus)
_prog(8 + int(60 * (i + 1) / max(1, len(new_clips))),
f"clip {i+1}/{len(new_clips)} prompt written")
m["clips"] = new_clips
......@@ -4731,41 +4922,28 @@ def launch_web_ui(default_args):
_prog(12, f"re-planning clips for {match_name} @ {fps}fps "
f"({_cf_lo}-{_cf_hi}f/clip)…")
# Rebuild the clip list with the same planner the full run uses:
# frame budget within the configured range, match length counted in
# PLAYBACK seconds (nf/fps).
new_clips, planned, ci = [], 0.0, 0
while planned < long_target:
round_num = ci // 3 + 1
intensity = ("early exchanges" if round_num == 1
else "midpoint battle" if round_num == 2
else "climactic final exchange")
_nf = random.randint(_cf_lo, _cf_hi)
clip_seconds = round(_nf / max(1, fps), 2)
new_clips.append({"idx": ci, "clip_seconds": clip_seconds,
"nf": _nf, "intensity": intensity,
"shot": None, "prompt": None})
planned += _nf / max(1, fps)
ci += 1
# intro clips (entrances + face-off) then fight clips, frame budget
# within the configured range, match length counted in PLAYBACK
# seconds (nf/fps).
new_clips = _build_match_clip_specs(
fps, _cf_lo, _cf_hi, long_target, m["f1"], m["f2"])
_ref = m.get("referee") or _referee_for(out_dir, match_name)
# Write a fresh, varied prompt for each new clip.
f1_hint = _fighter_desc_hint(m["f1"], char_descriptions)
f2_hint = _fighter_desc_hint(m["f2"], char_descriptions)
match_avoid = []
_focus_cycle = list(FIGHT_ACTION_FOCUS)
random.shuffle(_focus_cycle)
_ffi = 0
for i, c in enumerate(new_clips):
if _cancelled():
_cancel_done("⏹ cancelled — re-plan stopped (prompts unchanged)")
return
shot = prompter.fight_shot(
m["f1"], m["f2"], m["env_desc"],
match_context=f"Match stage: {c['intensity']}. ",
avoid=match_avoid,
action_focus=_focus_cycle[i % len(_focus_cycle)])
c["shot"] = shot
c["prompt"] = (f"{f1_hint} vs {f2_hint} — {shot} "
f"— {_continuity_clause(m.get('env'))} "
f"— {FIGHT_PROMPT_SUFFIX}")
match_avoid.append(shot[:60])
_focus = ("" if c.get("role", "fight") != "fight"
else _focus_cycle[_ffi % len(_focus_cycle)])
if c.get("role", "fight") == "fight":
_ffi += 1
_fill_clip_prompt(prompter, c, m["f1"], m["f2"], m.get("env"),
m["env_desc"], char_descriptions, referee=_ref,
match_avoid=match_avoid, focus=_focus)
_prog(12 + int(84 * (i + 1) / max(1, len(new_clips))),
f"clip {i+1}/{len(new_clips)} prompt written")
m["clips"] = new_clips
......@@ -4783,6 +4961,64 @@ def launch_web_ui(default_args):
f"previous prompts), THEN Re-render")
return
# ── Re-write ONLY one clip's prompt (text model) ───────────────────
# Rewrites a single clip's prompt in place (role-aware: entrance /
# face-off / fight), preserving the clip's frame budget and position.
# Nothing is rendered — afterwards regenerate that clip's keyframe and
# re-render it.
if scope == "clip-prompt":
m = next((x for x in fight_plan if x.get("match_name") == match_name), None)
if not m:
_fail("match not found in prompts.json")
return
try:
idx = int(params.get("idx"))
except (TypeError, ValueError):
_fail("clip index required")
return
c = next((x for x in m.get("clips", []) if int(x["idx"]) == idx), None)
if not c:
_fail("clip not found in prompts.json")
return
_prog(8, "preparing text model…")
client = CoderAIClient(default_args.base_url,
getattr(default_args, "api_key", None))
text_model = None
if not getattr(default_args, "no_llm", False):
try:
text_model = pick_model(client, "text",
getattr(default_args, "text_model", None))
except Exception as e:
_log(f" [clip-prompt] no text model ({e}); using template prompts")
text_model = getattr(default_args, "text_model", None)
char_descriptions = _build_char_descriptions(out_dir)
prompter = PromptGenerator(client, text_model,
char_descriptions=char_descriptions)
_ref = m.get("referee") or _referee_for(out_dir, match_name)
# Steer a fight-clip rewrite away from this match's OTHER clip shots
# so the new prompt stays distinct from its neighbours.
_avoid = [str(x.get("shot"))[:60] for x in m.get("clips", [])
if x is not c and x.get("shot")]
_focus = (random.choice(FIGHT_ACTION_FOCUS)
if c.get("role", "fight") == "fight" else "")
_prog(40, f"rewriting clip {idx:02d} prompt…")
if _cancelled():
_cancel_done("⏹ cancelled — clip prompt unchanged")
return
_fill_clip_prompt(prompter, c, m["f1"], m["f2"], m.get("env"),
m["env_desc"], char_descriptions, referee=_ref,
match_avoid=_avoid, focus=_focus)
try:
pf.write_text(json.dumps(
{"fight_plan": fight_plan, "outcome_plan": outcome_plan,
"fps": data.get("fps") or fps}, indent=2))
except Exception as e:
_fail(f"could not save prompts.json: {e}")
return
_done(f"rewrote clip {idx:02d} prompt for {match_name} — now "
f"regenerate its keyframe (kf↻), then re-render it")
return
# ── Re-plan ONLY this match's outcome prompts (text model) ──────────
# Rewrites the finish + victory shots for every outcome of the match,
# leaving the fight-clip prompts untouched.
......@@ -4947,37 +5183,24 @@ def launch_web_ui(default_args):
_cf_lo, _cf_hi = _clip_frame_range(
getattr(default_args, "clip_min_frames", CLIP_MIN_FRAMES),
getattr(default_args, "clip_max_frames", CLIP_MAX_FRAMES))
new_clips, planned, ci = [], 0.0, 0
while planned < long_target:
round_num = ci // 3 + 1
intensity = ("early exchanges" if round_num == 1
else "midpoint battle" if round_num == 2
else "climactic final exchange")
_nf = random.randint(_cf_lo, _cf_hi)
new_clips.append({"idx": ci, "clip_seconds": round(_nf / max(1, fps), 2),
"nf": _nf, "intensity": intensity,
"shot": None, "prompt": None})
planned += _nf / max(1, fps)
ci += 1
f1_hint = _fighter_desc_hint(m["f1"], char_descriptions)
f2_hint = _fighter_desc_hint(m["f2"], char_descriptions)
new_clips = _build_match_clip_specs(
fps, _cf_lo, _cf_hi, long_target, m["f1"], m["f2"])
_ref = m.get("referee") or _referee_for(out_dir, match_name)
match_avoid = []
_focus_cycle = list(FIGHT_ACTION_FOCUS)
random.shuffle(_focus_cycle)
_ffi = 0
for i, c in enumerate(new_clips):
if _cancelled():
_cancel_done("⏹ cancelled during prompt planning — nothing rendered")
return
shot = prompter.fight_shot(
m["f1"], m["f2"], m["env_desc"],
match_context=f"Match stage: {c['intensity']}. ",
avoid=match_avoid,
action_focus=_focus_cycle[i % len(_focus_cycle)])
c["shot"] = shot
c["prompt"] = (f"{f1_hint} vs {f2_hint} — {shot} "
f"— {_continuity_clause(m.get('env'))} "
f"— {FIGHT_PROMPT_SUFFIX}")
match_avoid.append(shot[:60])
_focus = ("" if c.get("role", "fight") != "fight"
else _focus_cycle[_ffi % len(_focus_cycle)])
if c.get("role", "fight") == "fight":
_ffi += 1
_fill_clip_prompt(prompter, c, m["f1"], m["f2"], m.get("env"),
m["env_desc"], char_descriptions, referee=_ref,
match_avoid=match_avoid, focus=_focus)
_prog(4 + int(14 * (i + 1) / max(1, len(new_clips))),
f"clip {i+1}/{len(new_clips)} prompt written")
m["clips"] = new_clips
......@@ -5072,13 +5295,27 @@ def launch_web_ui(default_args):
return
# ── Phase 4/4 — assemble the final short/long videos (last) ──────
_prog(94, "phase 4/4 — assembling final videos…")
_prog(90, "phase 4/4 — assembling final videos…")
st = float(m.get("short_target", 45))
lt = float(m.get("long_target", 70))
n_assembled = _reassemble_finals(vdir, match_name, st, lt)
# A whole-match regen finishes with a quality pass: 2× AI upscale and
# 2× frame interpolation of the final short/long assemblies and the
# outcome videos (the per-clip fight videos are left at base res).
_enh = 0
try:
_prog(94, "phase 4/4 — upscaling ×2 + interpolating ×2 finals + outcomes…")
_um = _upscale_model_for(default_args, 2) or None
_im = (getattr(default_args, "interpolation_model", None) or None)
_enh = _stage_enhance_videos(
client, _um, vdir, [mm], match_outcomes,
upscale=2, fps_mult=2, interpolation_model=_im)
except Exception as e:
_log(f" ⚠ enhance pass failed (finals left at base res): {e}")
_done(f"regenerated match {match_name} end to end — "
f"{len(mm['clips'])} clip(s), {len(match_outcomes)} outcome(s), "
f"finals assembled from {n_assembled} clip(s)")
f"finals assembled from {n_assembled} clip(s), "
f"{_enh} video(s) upscaled ×2 + interpolated ×2")
return
# ── Regenerate keyframes (image model) ─────────────────────────────
......@@ -6802,14 +7039,15 @@ async function reMatch(ev, scope, params){
'keyframes-missing':'Generate only the MISSING keyframe images for this match (uses the image model)? Existing keyframes are kept; nothing is re-rendered.',
'keyframe':'Regenerate this keyframe image (uses the image model)? The clip video is NOT re-rendered — click Re-render afterwards.',
'replan':'Re-plan this match: rebuild its clip list and rewrite all clip prompts (frame budget from the Clip min/max frames settings). Only this match changes — other matches and outcomes are untouched. AFTERWARDS, in order: 1) Regenerate keyframes (the old keyframes match the PREVIOUS prompts and would anchor the video to the wrong image — causing static/low-motion clips), 2) Re-render all clips, 3) Reassemble finals.',
'replan-outcomes':'Rewrite ONLY this match\\'s outcome prompts (the finish + victory shots for win / ko / retire / draw). The fight-clip prompts are left untouched. AFTERWARDS regenerate the outcome keyframes, then re-render the outcomes.'};
'replan-outcomes':'Rewrite ONLY this match\\'s outcome prompts (the finish + victory shots for win / ko / retire / draw). The fight-clip prompts are left untouched. AFTERWARDS regenerate the outcome keyframes, then re-render the outcomes.',
'clip-prompt':'Rewrite ONLY this single clip\\'s prompt (text model). The other clips are untouched and nothing is rendered. AFTERWARDS regenerate this clip\\'s keyframe (kf↻), then Re-render it.'};
const kf=(scope==='keyframes'||scope==='keyframe'||scope==='keyframes-missing');
const kfMiss=(scope==='keyframes-missing');
const isReplan=(scope==='replan'||scope==='replan-outcomes');
const isReplan=(scope==='replan'||scope==='replan-outcomes'||scope==='clip-prompt');
const isFull=(scope==='full');
if(!(await uiConfirm(labels[scope]||'Proceed?',
{title:(isFull?'Regenerate whole match':(isReplan?'Re-plan match prompts':(kfMiss?'Generate missing keyframes':(kf?'Regenerate keyframes':'Regenerate')))),
okText:(isFull?'Regenerate match':(scope==='reassemble'?'Reassemble':(isReplan?'Re-plan':(kfMiss?'Generate missing':(kf?'Regenerate':'Re-render'))))),
{title:(isFull?'Regenerate whole match':(scope==='clip-prompt'?'Rewrite clip prompt':(isReplan?'Re-plan match prompts':(kfMiss?'Generate missing keyframes':(kf?'Regenerate keyframes':'Regenerate'))))),
okText:(isFull?'Regenerate match':(scope==='reassemble'?'Reassemble':(scope==='clip-prompt'?'Rewrite prompt':(isReplan?'Re-plan':(kfMiss?'Generate missing':(kf?'Regenerate':'Re-render')))))),
danger:(scope!=='reassemble'&&!kf&&!isReplan)})))return;
const stEl=_findStatus(ev);
const setSt=(c,t)=>{ if(stEl){ stEl.style.color=c; stEl.textContent=t; } };
......@@ -7242,6 +7480,8 @@ function pollNewMatch(jobId, st){
f' <div class=hint style="display:flex;justify-content:space-between;align-items:center">'
f'<span>clip {idx:02d}</span>'
f'<span>'
f'<a href="#" style="color:#9fdca0" title="Rewrite ONLY this clip\'s prompt (text model). Afterwards regenerate its keyframe and re-render." '
f'onclick="reMatch(event,\'clip-prompt\',{{match:\'{_esc(name)}\',idx:\'{idx}\'}})">prompt↻</a> '
f'<a href="#" style="color:#c79bf0" title="Regenerate this keyframe (image model)" '
f'onclick="reMatch(event,\'keyframe\',{{match:\'{_esc(name)}\',idx:\'{idx}\'}})">kf↻</a> '
f'<a href="#" style="color:#7eb8f7" '
......@@ -8283,7 +8523,7 @@ async function resetPrompts(ev){
if scope not in ("full", "match-clips", "clip", "reassemble", "outcomes",
"outcome", "enhance", "keyframes",
"keyframes-missing", "keyframe", "replan",
"replan-outcomes"):
"replan-outcomes", "clip-prompt"):
self._send(400, "application/json",
_j.dumps({"error": "invalid scope"}))
return
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment