File size: 32,108 Bytes
7e61d08 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 | #!/usr/bin/env python3
"""PPE · MiniMax-H3 ref2va 换装工作流:节点插件安装 + 模型补齐 + 校验 + 工作流整理(RunPod 用)
只做准备和静态检查,不启动生成或训练。
用法(在装了 ComfyUI 的 RunPod 机器终端里):
python3 /workspace/ppe_models.py --workflow /workspace/1111_api.json
可选:
--with-full 另外下载 B200 满血版(ref2va bf16 66 GB、官方文本模型 bf16 52 GB、视频 VAE fp16 5 GB)
--models-dir D 手动指定模型目录(默认 ComfyUI/models,必须在持久化存储上)
--check-only 只检查,不安装、不下载
--force 模型目录不在持久化存储上也继续(不建议)
大小和 SHA256 已于 2026-11-03 在 Hugging Face 接口上逐个核对。
"""
from __future__ import annotations
import argparse
import copy
import datetime as dt
import hashlib
import json
import os
import re
import shutil
import struct
import subprocess
import sys
import urllib.request
HF = "https://huggingface.co"
def u(repo, path, rev="main"):
return f"{HF}/{repo}/resolve/{rev}/{path}"
# ---------------------------------------------------------------------------
# 模型:name, 子目录, 地址, 大小, SHA256, 分组, 说明
# exact = 与工作流引用同名同文件
# renamed = 作者改了名,按作者记录是同一文件(工作流引用会改成新名)
# chosen = 你指定的替代文件(工作流引用会改成它)
# full = 满血版,加 --with-full 才下载
# ---------------------------------------------------------------------------
MODELS = [
("sam3.1_multiplex_fp16.safetensors", "checkpoints", u("Comfy-Org/sam3.1", "checkpoints/sam3.1_multiplex_fp16.safetensors"),
1745546848, "9ba99c92703c2e8b4f47de2d34a539bb8e18923049e238b780d70dbe6368eb03", "exact", "人物识别 SAM3.1"),
("minimax_h3_ref2va_int8_convrot.safetensors", "diffusion_models", u("Comfy-Org/MiniMax-H3", "diffusion_models/minimax_h3_ref2va_int8_convrot.safetensors"),
34038894550, "9eef934046a0671bc8a5daf87100705e1478419c574cfde70c50fbe6885f76a9", "exact", "前半段主模型 int8"),
("minimax_h3_ref2va_pruned_w4a8_mixed.safetensors", "diffusion_models", u("Kijai/MiniMax-H3-experimental", "minimax_h3_ref2va_pruned_w4a8_mixed.safetensors"),
11770657048, "de2c6c29c4ee702b45e48e40daae3834aeee58ab681c732d9152589a87c89910", "exact", "后半段主模型 w4a8"),
("qwen3vl_32b_h3_ultra_uncensored_heretic_int8_convrot.safetensors", "text_encoders",
u("ethanfel/Qwen3-VL-32B-Ultra-Heretic-H3-ComfyUI-INT8-ConvRot", "qwen3vl_32b_h3_ultra_uncensored_heretic_int8_convrot.safetensors"),
26363476151, "d84547412144b7c50a6ec77437a889b869d3ace88da77ef1775d3d2a4901c192", "exact", "文本模型(社区改版)"),
("minimax_h3_video_vae_int8_convrot.safetensors", "vae", u("Comfy-Org/MiniMax-H3", "vae/minimax_h3_video_vae_int8_convrot.safetensors"),
2811065184, "52a2c8c73583c86e4f41cdcce3a6ad0ea562987bc0bf3d60a0cef5f5c8e60c0e", "exact", "视频 VAE int8"),
("minimax_h3_audio_vae_fp32.safetensors", "vae", u("Comfy-Org/MiniMax-H3", "vae/minimax_h3_audio_vae_fp32.safetensors"),
605254808, "8e505d95dd1561d47abd43d4238fd40d9bb1ae9e147ed0a4cba778d76ae4db48", "exact", "音频 VAE"),
("taeh3.safetensors", "vae_approx", u("OzzyGT/taeh3", "taeh3.safetensors"),
22709752, "4fd022bfcab08772fe0536b17ea1a3bbb5625be11e397868d1c5d891863d4c13", "exact", "预览小 VAE"),
("dmad_minimax_h3_4step_full_critic_comfyui.safetensors", "loras", u("ZhengmingYu/DMAD", "minimax_h3/dmad_minimax_h3_4step_full_critic_comfyui.safetensors"),
1956195976, "d3ee9f599ef1ac2d76850e90372ce9064fdc2c16e7f3de6a7ba9bc90cb91eb1b", "exact", "4 步加速 LoRA"),
("h3_character_swap_pro4500_1000.safetensors", "loras", u("akatz-ai/MiniMax-H3-Character-Swap-LoRA", "h3_character_swap_pro4500_1000.safetensors"),
155110320, "4b2a3f420ae804c0aa3422761ff84dbd1bf52eef6900ffab6d2e66df63cb4e79", "exact", "换人 LoRA"),
("MMH3-ExposureSlider-V1.safetensors", "loras", u("Kutches/minmax", "MMH3-ExposureSlider-V1.safetensors"),
217896, "530334bb668fc222a179d55097632ef398288e497728518b096fb69e38c8d9f7", "exact", "曝光 LoRA(镜像仓库,原作者 alcaitiff)"),
("h3_upscaler_lms_v0.1_fp32.safetensors", "latent_upscale_models", u("Alissonerdx/Minimax-H3-ComfyUI", "latent_upscaler/h3_upscaler_lms_v0.1_fp32.safetensors"),
1381153616, "7b70711ebf13051bac4b637247f2e6f0db5e1c6da88af3f28ad0165c6483c8c4", "renamed",
"放大模型;工作流旧名 h3_upscaler_sharpness_2000steps_v0.1_fp32(按作者记录为同一文件)"),
("minimax_h3_lms_v1.0_r64.safetensors", "loras", u("Alissonerdx/Minimax-H3-ComfyUI", "loras/minimax_h3_lms_v1.0_r64.safetensors"),
1239664536, "16f3195bc6bffa431c0598dc2031e520ba058b4fcbc9f104db9e3fdfdeacf60c", "chosen",
"LMS 作者原版,你指定替代节点 610 的 _r2(不是同一文件)"),
("MysticXXX_MMH3-V4-ref2va.safetensors", "loras",
u("jinksa77/Minimax_H3", "MysticXXX_MMH3-V4-ref2va.safetensors", "21a3da2332b977f756a001192b106fe95f020277"),
137654280, "106243744bb11cf962d66a92fd837f9744431ecdf2d4ebd63ff5385afec11b59", "chosen",
"你指定替代节点 636/660 的 MysticXXX-ref2va-V4(成人向 LoRA;与旧名文件是否同一文件未确认)"),
("minimax_h3_ref2va_bf16.safetensors", "diffusion_models", u("Comfy-Org/MiniMax-H3", "diffusion_models/minimax_h3_ref2va_bf16.safetensors"),
66280487368, "e32c54c1a7b4f5f397f195cea267ccb18806303bb665678c4bee60953bdf3026", "full", "满血版主模型 bf16(B200 测试用)"),
("qwen3vl_32b_minimax_h3_bf16.safetensors", "text_encoders", u("Comfy-Org/MiniMax-H3", "text_encoders/qwen3vl_32b_minimax_h3_bf16.safetensors"),
51506295256, "600d567f6a9629c8574e8e7041b199bdd9c59a986afa7906910a81919610607d", "full", "官方文本模型 bf16(B200 测试用)"),
("minimax_h3_video_vae_fp16.safetensors", "vae", u("Comfy-Org/MiniMax-H3", "vae/minimax_h3_video_vae_fp16.safetensors"),
5207808496, "7c1f131492e7eddacaac9069a61b81bdd39de5cc96561e677c5eab1cdce5e522", "full", "视频 VAE fp16(B200 测试用)"),
]
GROUP = {"exact": "完全匹配", "renamed": "改名同文件", "chosen": "指定替代", "full": "满血版"}
# 工作流里要改的模型引用(旧名 → 新名)
RENAME = {
"h3_upscaler_sharpness_2000steps_v0.1_fp32.safetensors": "h3_upscaler_lms_v0.1_fp32.safetensors", # 节点 645
"minimax_h3_lms_v1.0_r64_r2.safetensors": "minimax_h3_lms_v1.0_r64.safetensors", # 节点 610
"MysticXXX-ref2va-V4.safetensors": "MysticXXX_MMH3-V4-ref2va.safetensors", # 节点 636 / 660
}
REMOVE_BRIDGE = [632] # jamie LoRA:删除节点,上游模型直接接到下游
REMOVE_NODES = [650, 651] # 已静音的打包节点(VideoCombineNode 找不到来源,CompressImages 用不上)
# ---------------------------------------------------------------------------
# 节点插件:目录名, 地址, 提供的节点
# ---------------------------------------------------------------------------
NODE_PACKS = [
("ComfyUI-KJNodes", "https://github.com/kijai/ComfyUI-KJNodes",
["GetImageSizeAndCount", "ImageConcatMulti", "ModelPreviewOverrideKJ", "MiniMaxChunkFeedForward",
"MiniMaxLowVRAMAttention", "MiniMaxH3MemoryEfficientSageAttentionPatch"]),
("ComfyUI-Easy-Use", "https://github.com/yolain/ComfyUI-Easy-Use", ["easy ifElse", "easy int", "easy positive", "easy showAnything"]),
("ComfyUI-VideoHelperSuite", "https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite", ["VHS_LoadVideo", "VHS_VideoCombine"]),
("ComfyUI_LayerStyle", "https://github.com/chflame163/ComfyUI_LayerStyle", ["LayerUtility: ImageScaleByAspectRatio V2"]),
("ComfyUI_Comfyroll_CustomNodes", "https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes", ["CR Prompt Text"]),
("ComfyUI-utils-nodes", "https://github.com/zhangp365/ComfyUI-utils-nodes", ["ConcatTextOfUtils"]),
("ComfyUI_SwiftCut", "https://github.com/yichengup/ComfyUI_SwiftCut", ["YCImageBatchBlend"]),
("Comfyui_Minimax_h3_latent_Upscaler", "https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler", ["MinimaxH3LatentUpscaler3D"]),
]
# 单文件节点:SolAttnMiniMax(Sol 稀疏注意力),放在 custom_nodes 根目录
SINGLE_FILE_NODES = [
("sol_attn_minimax_v2.py",
"https://raw.githubusercontent.com/xm6018924/BSAI-ComfyUI-Sol-H3/bd05744bd2c98a9cfd62daac1ccfece4f8ee5521/sol_attn_minimax_v2.py",
["SolAttnMiniMax"]),
]
# ComfyUI 本体需要有的新节点(太旧要更新 ComfyUI)
CORE_NEEDED = ["SAM3_VideoTrack", "SCAIL2ColoredMask", "MiniMaxH3ReferenceToVideo", "MiniMaxH3SigmaShift", "EmptyAudio",
"ComfyMathExpression", "ModelAttentionBackend", "LoraLoaderBypassModelOnly", "LTXVConcatAVLatent", "ManualSigmas"]
LICENSE = [
("Comfy-Org/MiniMax-H3 官方模型、VAE、文本模型", "以 MiniMax-H3 官方许可为准,未核实商用条款"),
("SAM3.1", "以 Meta / Comfy-Org 页面为准,未核实"),
("MMH3-ExposureSlider-V1", "未核实独立商用授权(镜像仓库 Kutches/minmax)"),
("h3_upscaler_lms_v0.1_fp32", "上游 LBH 为 Apache-2.0;Alissonerdx 微调文件为混合许可,商用条款未确认"),
("minimax_h3_lms_v1.0_r64", "混合许可,商用条款未确认"),
("MysticXXX_MMH3-V4-ref2va", "成人向;原作者商用许可未核实(jinksa77 为镜像)"),
("dmad / character_swap / taeh3 / heretic 文本模型 / w4a8", "未核实"),
("各节点插件", "以各 GitHub 仓库的 LICENSE 为准,未逐个核实"),
]
LINES: list[str] = []
def out(s=""):
print(s, flush=True)
LINES.append(s)
def sh(cmd, timeout=600):
try:
r = subprocess.run(cmd, shell=True, capture_output=True, text=True, timeout=timeout)
return (r.stdout + ("\n" + r.stderr if r.returncode else "")).strip(), r.returncode
except Exception as e: # noqa: BLE001
return f"(失败: {e})", 1
def sho(cmd, timeout=120):
return sh(cmd, timeout)[0]
def human(n):
for unit in ("B", "KB", "MB", "GB"):
if n < 1024 or unit == "GB":
return f"{n:.1f} {unit}" if unit != "B" else f"{n} B"
n /= 1024
return str(n)
# ---------------- 环境 ----------------
def find_comfy():
ok = lambda c: c and os.path.isfile(os.path.join(c, "folder_paths.py")) and os.path.isfile(os.path.join(c, "main.py"))
for c in [os.environ.get("COMFY_DIR", ""), "/workspace/ComfyUI", "/workspace/comfyui", "/ComfyUI", "/opt/ComfyUI", "/root/ComfyUI"]:
if ok(c):
return os.path.realpath(c)
for p in sho("find /workspace /opt /root / -maxdepth 4 -name folder_paths.py -not -path '/proc/*' "
"-not -path '*/site-packages/*' 2>/dev/null", timeout=120).splitlines():
if ok(os.path.dirname(p)):
return os.path.realpath(os.path.dirname(p))
return None
def comfy_python(comfy):
for p in ("venv/bin/python", ".venv/bin/python", "../venv/bin/python", "../.venv/bin/python"):
f = os.path.join(comfy, p)
if os.path.isfile(f):
return os.path.realpath(f)
return shutil.which("python3") or sys.executable
def mount_of(path):
return sho(f"df -P '{path}' | tail -1 | awk '{{print $6}}'") or "?"
def is_persistent(path):
m = mount_of(path)
return m not in ("/", "?", ""), m
def extra_paths(comfy):
res = []
y = os.path.join(comfy, "extra_model_paths.yaml")
if not os.path.isfile(y):
return y, None, res
txt = open(y, encoding="utf-8", errors="replace").read()
try:
import yaml
data = yaml.safe_load(txt) or {}
for _, sec in data.items():
if not isinstance(sec, dict):
continue
base = sec.get("base_path", "")
for k, v in sec.items():
if k in ("base_path", "is_default"):
continue
for line in str(v).splitlines():
if line.strip():
res.append((k, os.path.normpath(os.path.join(base, line.strip()))))
except Exception as e: # noqa: BLE001
res.append(("解析失败", str(e)))
return y, txt, res
# ---------------- 校验 / 下载 ----------------
def safetensors_ok(path):
try:
with open(path, "rb") as f:
n = struct.unpack("<Q", f.read(8))[0]
if not (2 <= n <= 200 * 1024 * 1024):
return False, f"文件头长度不合理 ({n})"
head = f.read(n)
if not head.lstrip().startswith(b"{"):
return False, "文件头不是 JSON(可能下载到了网页)"
json.loads(head)
return True, "文件头正常"
except Exception as e: # noqa: BLE001
return False, f"读不了文件头: {e}"
def sha256(path):
h = hashlib.sha256()
with open(path, "rb") as f:
for b in iter(lambda: f.read(16 * 1024 * 1024), b""):
h.update(b)
return h.hexdigest()
def verify(path, size, sha):
if not os.path.isfile(path):
return False, "不存在"
real = os.path.getsize(path)
if real != size:
return False, f"大小不对:{real} ≠ {size}"
ok, msg = safetensors_ok(path)
if not ok:
return False, msg
out(f" 计算 SHA256 … ({human(size)})")
got = sha256(path)
if got != sha:
return False, f"SHA256 不对:{got}"
return True, "大小、文件头、SHA256 全部通过"
def download(url, dst, size, sha):
os.makedirs(os.path.dirname(dst), exist_ok=True)
part = dst + ".part"
hdr = f"-H 'Authorization: Bearer {os.environ['HF_TOKEN']}'" if os.environ.get("HF_TOKEN") else ""
for attempt in range(1, 4):
if os.path.exists(part) and os.path.getsize(part) > size:
os.replace(part, part + f".oversize.{int(dt.datetime.now().timestamp())}")
res, _ = sh(f"curl -L --fail -C - --retry 5 --retry-delay 5 --retry-all-errors --connect-timeout 30 "
f"{hdr} -o '{part}' -w '%{{http_code}}' '{url}' 2>/dev/null; echo \" rc=$?\"", timeout=6 * 3600)
have = os.path.getsize(part) if os.path.exists(part) else 0
out(f" 第 {attempt} 次:HTTP {res.splitlines()[0] if res else '?'},已下载 {human(have)} / {human(size)}")
if have == size:
break
ok, msg = verify(part, size, sha)
if ok:
os.replace(part, dst)
return ok, msg
# ---------------- 节点 ----------------
def scan_types(roots, types):
blob = []
for root in roots:
for r, dirs, files in os.walk(root):
dirs[:] = [d for d in dirs if d not in (".git", "node_modules", "models", "output", "input", "temp", "__pycache__", "tests")]
for f in files:
if f.endswith(".py"):
try:
blob.append(open(os.path.join(r, f), encoding="utf-8", errors="ignore").read())
except OSError:
pass
text = "\n".join(blob)
return {t: (f'"{t}"' in text or f"'{t}'" in text) for t in types}
def running_types():
try:
with urllib.request.urlopen("http://127.0.0.1:8188/object_info", timeout=10) as r:
return set(json.load(r))
except Exception: # noqa: BLE001
return None
def install_nodes(comfy, py, check_only):
cn = os.path.join(comfy, "custom_nodes")
os.makedirs(cn, exist_ok=True)
existing = {d.lower(): d for d in os.listdir(cn)}
rows = []
for name, url, types in NODE_PACKS:
have = existing.get(name.lower())
if have:
rows.append((name, "已安装", os.path.join(cn, have)))
continue
if check_only:
rows.append((name, "缺失(只检查)", url))
continue
out(f" 安装 {name} ← {url}")
res, rc = sh(f"git clone --depth 1 '{url}' '{os.path.join(cn, name)}'", timeout=900)
if rc:
rows.append((name, "安装失败:" + res[-200:], url))
continue
req = os.path.join(cn, name, "requirements.txt")
if os.path.isfile(req):
res, rc = sh(f"'{py}' -m pip install -r '{req}'", timeout=1800)
out(f" 依赖:{'装好了' if rc == 0 else '有报错 → ' + res[-300:]}")
rows.append((name, "新装", os.path.join(cn, name)))
for fname, url, types in SINGLE_FILE_NODES:
dst = os.path.join(cn, fname)
if os.path.isfile(dst):
rows.append((fname, "已安装", dst))
elif check_only:
rows.append((fname, "缺失(只检查)", url))
else:
res, rc = sh(f"curl -L --fail --retry 5 -o '{dst}.part' '{url}' && mv '{dst}.part' '{dst}'", timeout=300)
rows.append((fname, "新装" if rc == 0 else "安装失败:" + res[-200:], dst))
return rows
# ---------------- 工作流 ----------------
def wf_refs(wf):
refs = []
for n in wf.get("nodes", []):
vals = n.get("widgets_values")
vals = vals if isinstance(vals, list) else list((vals or {}).values())
for v in vals:
if isinstance(v, str) and re.search(r"\.(safetensors|pt|pth|ckpt|bin|gguf)$", v):
st = vals[1] if "Lora" in n["type"] and len(vals) > 1 and isinstance(vals[1], (int, float)) else None
refs.append(dict(id=n["id"], type=n["type"], mode=n.get("mode", 0), name=v, strength=st))
return refs
MODE = {0: "启用", 2: "静音", 4: "旁路"}
def rename_refs(wf, mapping):
changed = []
for n in wf.get("nodes", []):
for key in ("widgets_values", "widgets_values_named"):
vals = n.get(key)
items = list(enumerate(vals)) if isinstance(vals, list) else list((vals or {}).items())
for k, v in items:
if isinstance(v, str) and v in mapping:
vals[k] = mapping[v]
if key == "widgets_values":
changed.append(f"节点 {n['id']} {n['type']}:{v} → {mapping[v]}")
return changed
def _drop_link(wf, lid):
L = {l[0]: l for l in wf["links"]}
if lid not in L:
return
_, o, os_, t, ts, _ = L[lid]
N = {n["id"]: n for n in wf["nodes"]}
if o in N:
outs = N[o]["outputs"][os_].get("links") or []
N[o]["outputs"][os_]["links"] = [x for x in outs if x != lid]
if t in N and N[t]["inputs"][ts].get("link") == lid:
N[t]["inputs"][ts]["link"] = None
wf["links"] = [l for l in wf["links"] if l[0] != lid]
def remove_bridge(wf, nid):
"""删掉一个“模型进、模型出”的节点(如 LoRA),把上游直接接到所有下游。"""
N = {n["id"]: n for n in wf["nodes"]}
if nid not in N:
return f"节点 {nid} 不在工作流里,跳过"
n = N[nid]
L = {l[0]: l for l in wf["links"]}
inp = next((i for i in n["inputs"] if i.get("type") == "MODEL" and i.get("link") is not None), None)
outp = next((o for o in n["outputs"] if o.get("type") == "MODEL"), None)
if not inp or not outp:
return f"节点 {nid} 不是模型进出的节点,没有删"
src = L[inp["link"]]
downstream = [L[x] for x in (outp.get("links") or [])]
for lid in [i.get("link") for i in n["inputs"] if i.get("link") is not None] + list(outp.get("links") or []):
_drop_link(wf, lid)
N = {x["id"]: x for x in wf["nodes"]}
for d in downstream:
wf["last_link_id"] = wf.get("last_link_id", max(l[0] for l in wf["links"])) + 1
lid = wf["last_link_id"]
wf["links"].append([lid, src[1], src[2], d[3], d[4], "MODEL"])
N[src[1]]["outputs"][src[2]].setdefault("links", []).append(lid)
N[d[3]]["inputs"][d[4]]["link"] = lid
wf["nodes"] = [x for x in wf["nodes"] if x["id"] != nid]
tgt = ", ".join(f"{d[3]} {N[d[3]]['type']}" for d in downstream)
return f"删除节点 {nid} {n['type']}({(n.get('widgets_values') or [''])[0]}),{src[1]} {N[src[1]]['type']} 直接接到 {tgt}"
def remove_nodes(wf, ids):
msgs = []
for nid in ids:
N = {n["id"]: n for n in wf["nodes"]}
if nid not in N:
continue
n = N[nid]
lids = [i.get("link") for i in n.get("inputs", []) if i.get("link") is not None]
lids += [x for o in n.get("outputs", []) for x in (o.get("links") or [])]
for lid in lids:
_drop_link(wf, lid)
wf["nodes"] = [x for x in wf["nodes"] if x["id"] != nid]
msgs.append(f"删除节点 {nid} {n['type']}(原本已静音)")
return msgs
def check_links(wf):
N = {n["id"]: n for n in wf["nodes"]}
bad = 0
for i, o, os_, t, ts, _ in wf["links"]:
if o not in N or t not in N or i not in (N[o]["outputs"][os_].get("links") or []) or N[t]["inputs"][ts].get("link") != i:
bad += 1
for n in wf["nodes"]:
for inp in n.get("inputs", []):
if inp.get("link") is not None and not any(l[0] == inp["link"] for l in wf["links"]):
bad += 1
return bad
# ---------------- 主流程 ----------------
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--workflow", default="")
ap.add_argument("--models-dir", default="")
ap.add_argument("--with-full", action="store_true")
ap.add_argument("--check-only", action="store_true")
ap.add_argument("--force", action="store_true")
a = ap.parse_args()
ts = dt.datetime.now().strftime("%Y%m%d_%H%M%S")
out("=" * 72)
out(f"PPE 模型与节点准备 {ts} (只做准备和静态检查,没有运行生成)")
out("=" * 72)
# 1. 环境
out("\n[1] 环境")
comfy = find_comfy()
out(f" ComfyUI 目录:{comfy or '没找到'}")
if not comfy:
out(" !! 这台机器上没有 ComfyUI,停止。请换用装了 ComfyUI 的机器(或先装 ComfyUI)。")
return finish(ts, 2)
py = comfy_python(comfy)
out(f" ComfyUI 用的 Python:{py}")
out(" ComfyUI 版本:" + sho(f"cd '{comfy}' && git log -1 --format='%h %cd' --date=short 2>/dev/null"))
out(" 显卡:" + (sho("nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader") or "无"))
out(" torch / CUDA:" + sho(f"'{py}' -c 'import torch;print(torch.__version__, torch.version.cuda)' 2>&1 | tail -1"))
out(" comfy-kitchen:" + sho(f"'{py}' -c 'import comfy_kitchen as c;print(getattr(c,\"__version__\",\"?\"), \"sol_attn=\", hasattr(c,\"sol_attn\"))' 2>&1 | tail -1"))
sage, sage_rc = sh(f"'{py}' -c 'import sageattention as s;print(getattr(s,\"__version__\",\"已安装\"))'")
sage = sage.splitlines()[-1] if sage else ""
out(f" SageAttention:{sage if sage_rc == 0 else '没装'}")
out(" " + (sho("df -h /workspace 2>/dev/null | tail -1") or "/workspace 不存在"))
models_dir = os.path.realpath(a.models_dir or os.path.join(comfy, "models"))
pers, mnt = is_persistent(models_dir)
out(f" 模型目录:{models_dir}(挂载点 {mnt})→ {'持久化存储 ✓' if pers else '容器临时盘 ✗,删机器就没了'}")
y, txt, extra = extra_paths(comfy)
out(f" extra_model_paths.yaml:{'有' if txt else '没有'}({y})")
for k, p in extra:
out(f" {k}: {p} [{'持久化' if is_persistent(p)[0] else '临时盘'}]")
if not pers and not a.force and not a.check_only:
out(" !! 模型目录不在持久化存储上,停止下载(可用 --models-dir /workspace/... 指定)。")
a.check_only = True
comfy_roots = [models_dir] + [p for _, p in extra if os.path.isdir(p)]
search_roots = list(dict.fromkeys(comfy_roots + ["/workspace"]))
def locate(name):
hits = []
for r in search_roots:
hits += [x for x in sho(f"find '{r}' \\( -name .venv -o -name venv -o -name .uv -o -name node_modules -o -name .git "
f"-o -name custom_nodes -o -name site-packages \\) -prune -o -name '{name}' -type f -print "
f"2>/dev/null | head -5", timeout=600).splitlines() if x]
return list(dict.fromkeys(hits))
# 2. 节点插件
out("\n[2] 节点插件")
node_rows = install_nodes(comfy, py, a.check_only)
for name, st, where in node_rows:
out(f" {name:<40}{st} {where}")
# 3. 模型
out("\n[3] 模型")
want = [m for m in MODELS if m[5] != "full" or a.with_full]
results = []
for name, sub, url, size, sha, group, note in want:
dst = os.path.join(models_dir, sub, name)
out(f"\n · {name} [{GROUP[group]}]")
status = None
for c in [dst] + [h for h in locate(name) if h != dst]:
if not os.path.isfile(c):
continue
ok, msg = verify(c, size, sha)
out(f" 已有 {c}:{msg}")
if ok:
status, path = "已有,校验通过", c
inside = any(os.path.realpath(c).startswith(os.path.realpath(r) + os.sep) for r in comfy_roots)
if not inside and not os.path.exists(dst) and not a.check_only:
os.makedirs(os.path.dirname(dst), exist_ok=True)
os.symlink(c, dst)
out(f" 在 ComfyUI 读取范围外,已建链接 {dst} → {c}(没有复制)")
status, path = "已有(链接),校验通过", dst
elif not inside:
out(" 注意:这个位置 ComfyUI 可能读不到")
break
if not status:
if a.check_only:
status, path = "缺失(只检查,没下载)", dst
else:
if os.path.isfile(dst):
os.replace(dst, dst + f".bad_{ts}")
out(f" 校验不过的旧文件已改名为 {dst}.bad_{ts}")
out(f" 下载 → {dst}")
ok, msg = download(url, dst, size, sha)
out(f" {msg}")
status, path = ("下载完成,校验通过" if ok else f"下载失败:{msg}"), dst
results.append((name, group, status, path, size, sha, note))
out("\n 本机旧名文件比对(只有哈希一致才算同一文件):")
for old, h in {"MysticXXX-ref2va-V4.safetensors": "106243744bb11cf962d66a92fd837f9744431ecdf2d4ebd63ff5385afec11b59",
"h3_upscaler_sharpness_2000steps_v0.1_fp32.safetensors": "7b70711ebf13051bac4b637247f2e6f0db5e1c6da88af3f28ad0165c6483c8c4"}.items():
hits = locate(old)
if not hits:
out(f" {old}:本机没有,无法比对")
for p in hits:
got = sha256(p)
out(f" {p}:{'哈希一致,确认同一文件' if got == h else '哈希不同,不是同一文件 (' + got + ')'}")
out(f" minimax_h3_lms_v1.0_r64_r2.safetensors:{'找到 ' + ', '.join(locate('minimax_h3_lms_v1.0_r64_r2.safetensors')) if locate('minimax_h3_lms_v1.0_r64_r2.safetensors') else '本机没有'}")
# 4. 工作流
out("\n[4] 工作流")
wf_path = a.workflow
if not wf_path:
c = sho("grep -rl --include=*.json MinimaxH3LatentUpscaler3D /workspace 2>/dev/null | grep -v -e '\\.bak_' -e '_runpod' | head -1")
wf_path = c.splitlines()[0] if c else ""
wf_types = []
if not wf_path or not os.path.isfile(wf_path):
out(" 没找到工作流 JSON(用 --workflow 指定),跳过工作流检查。")
else:
wf = json.load(open(wf_path, encoding="utf-8"))
base, _ = os.path.splitext(wf_path)
bak = f"{base}.bak_{ts}.json"
shutil.copyfile(wf_path, bak)
out(f" 原工作流(不改动):{wf_path}\n 备份:{bak}")
new = copy.deepcopy(wf)
changes = rename_refs(new, RENAME)
for nid in REMOVE_BRIDGE:
changes.append(remove_bridge(new, nid))
changes += remove_nodes(new, REMOVE_NODES)
bad = check_links(new)
p_new = f"{base}_runpod.json"
json.dump(new, open(p_new, "w", encoding="utf-8"), ensure_ascii=False, indent=2)
out(f" 新工作流:{p_new}(连线检查:{'全部正常' if bad == 0 else f'有 {bad} 处异常'})")
for c in changes:
out(" " + c)
have = {r[0] for r in results if "通过" in r[2]}
out("\n 新工作流里引用的模型:")
blocking = []
for r in wf_refs(new):
ok = r["name"] in have or bool(locate(r["name"]))
w = f",权重 {r['strength']}" if r["strength"] is not None else ""
out(f" 节点 {r['id']:>4} {r['type']:<28} {MODE.get(r['mode'], r['mode'])}{w}:{r['name']} {'✓' if ok else '✗ 缺失'}")
if not ok and r["mode"] == 0:
blocking.append(r)
out(" 会阻塞运行的缺失模型:" + ("、".join(f"节点 {r['id']} {r['name']}" for r in blocking) if blocking else "无"))
wf_types = sorted({n["type"] for n in new["nodes"] if n.get("mode", 0) == 0})
if sage_rc != 0:
nos = copy.deepcopy(new)
for n in nos["nodes"]:
if n["type"] == "MiniMaxH3MemoryEfficientSageAttentionPatch":
n["mode"] = 4
p2 = f"{base}_runpod_免Sage.json"
json.dump(nos, open(p2, "w", encoding="utf-8"), ensure_ascii=False, indent=2)
out(f" SageAttention 没装:节点 633(KJ Sage 省显存补丁)运行时会报错。"
f"\n 另存了一份把 633 设为旁路的版本:{p2}(B200 显存够,这个补丁本来就可以不要)")
# 5. 节点检查
out("\n[5] 节点检查")
rt = running_types()
if rt is not None:
how = "ComfyUI 正在运行;但新装的插件要重启 ComfyUI 才会加载,下面结果以重启前为准"
avail = {t: t in rt for t in set(wf_types) | set(CORE_NEEDED)}
else:
how = "ComfyUI 没在运行,按代码里能否搜到节点名判断(启动后以界面为准)"
avail = scan_types([comfy], sorted(set(wf_types) | set(CORE_NEEDED)))
out(f" ({how})")
core_miss = [t for t in CORE_NEEDED if not avail.get(t)]
out(" ComfyUI 本体缺的节点:" + (", ".join(core_miss) + " → ComfyUI 版本太旧,需要更新(cd ComfyUI && git pull && pip install -r requirements.txt)" if core_miss else "无"))
miss = [t for t in wf_types if not avail.get(t)]
out(" 工作流用到、仍然缺失的节点:" + (", ".join(miss) if miss else "无"))
# 6. 汇总
out("\n[6] 汇总")
for name, group, status, path, size, sha, note in results:
out(f" {name}\n {GROUP[group]} | {status}\n {path} | {size} 字节 | sha256 {sha}\n {note}")
if not a.with_full:
out(" 满血版没下载(需要时加 --with-full)")
out("\n 替代说明:")
out(" 节点 610 LMS:_r2 找不到,按你的指定换成作者原版 minimax_h3_lms_v1.0_r64,效果没验证")
out(" 节点 636/660 MysticXXX:按你的指定换成 MysticXXX_MMH3-V4-ref2va(成人向),和原工作流文件是否相同未确认")
out(" 节点 632 jamie:按你的要求已删除,上游模型直接接到下游")
out(" 节点 650/651:原本静音的打包节点,已删除(VideoCombineNode 找不到来源)")
out("\n 许可(都没有确认可以商用):")
for k, v in LICENSE:
out(f" {k}:{v}")
out("\n 注意:以上只是安装、下载和静态检查,工作流没有实际运行验证。新装了插件的话要重启 ComfyUI。")
return finish(ts, 0)
def finish(ts, code):
d = "/workspace" if os.path.isdir("/workspace") else "."
p = os.path.join(d, f"ppe_model_report_{ts}.txt")
with open(p, "w", encoding="utf-8") as f:
f.write("\n".join(LINES) + "\n")
print(f"\n报告已保存:{p}")
return code
if __name__ == "__main__":
sys.exit(main())
|