Keep the Anima illustration skill to judgment only

A skill now says how to judge and names the parts; models and values
live in the way and in the user's own combos and workflows. The Anima
skill drops the tested models, LoRA and sampler numbers (the
upscale_anima_tile way keeps its second-pass values), says to use a
workflow or combo the user saved for this route as it is, and keeps the
findings as judgment: show the base first, what Tile does and does not
keep, and when changing models between passes is untried.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

CHANGELOG.md

@@ -2,9 +2,9 @@
 
 ## v1.63.0 — unreleased
 
-- 新增按需 skill **Anima 插画·底图与 Tile 细化**，以及自动接入的 **Anima Tile 放大细化** 做法：保留底图，以 1.22 倍潜空间放大和 Tile 引导高降噪细化，检查模型架构与必需权重，结果另存供对比。参数来自三组本机对照，不保证修手或复刻原图。
-- New on-demand **Anima illustration: base and Tile detail** skill and **Anima Tile detail upscale** way: retain the base and save a guided 1.22× latent redraw separately, checking the model family and required weight. Settings come from three local comparisons and do not guarantee hand repair or reproduction.
-
+- 新增 **Anima Tile 放大细化** 做法：保留底图，以 1.22 倍潜空间放大和 Tile 引导高降噪细化，检查模型架构与必需权重，结果另存供对比；参数来自三组本机对照，不保证修手或复刻原图。配套的按需 skill **Anima 插画·底图与 Tile 细化** 只写怎么判断（先出底图给你看、再用这块做法、你存过这条路线的工作流或搭配就直接用、你没选的画风和 LoRA 不加），具体模型和数值留在做法和你自己的搭配里。
+- New **Anima Tile detail upscale** way: retain the base and save a guided 1.22× latent redraw separately, checking the model family and required weight; settings come from three local comparisons and do not guarantee hand repair or reproduction. Its on-demand skill **Anima illustration: base and Tile detail** holds judgment only (show the base first, then this way; a workflow or combo you saved for this route is used as it is; no style or LoRA you did not pick); the models and values stay in the way and in your own combos.
+
 - 跑完的图下面新增 **效果好 / 不太行** 两个按钮，点了直接告诉 AI（点“不太行”它就换做法、设置或提示词再跑），再点一下收回。**关卡** 里新增 **第三方节点**：选“只用做法里的”，AI 自己加节点时只能用做法里搭好的组合和你给任务挑的，ComfyUI 自带的节点照常能用。等不到回答时设成按推荐的继续，挑图面板里 AI 推荐的那张也会替你选上。
 - Under a finished run's pictures, **Turned out well / Not quite** tells the AI at once (with "Not quite" it changes the way, a setting or the prompt before running again); a click again takes it back. The **Gates** window has a new **Custom nodes** row: with "Only the ways'" the AI may add only the nodes of the ready-made ways and what you picked for the task; ComfyUI's built-in nodes stay allowed. With "go on with the recommended" for unanswered questions, a picks panel takes the picture the AI recommended too.
 - 新增 **隐私模式**：任务标题栏的 **🔒** 一点，这个任务的**图和提示词原文**都不会发给 AI。图一张都不给（放大看图、图片信息、跑完的预览、你递给它的图、AI 软件里任务卡片上的图都一样），它只知道出了几张、成没成功；提示词和图片文件名换成 **[P1]**、**[img1]** 这样的代号：画布上、图片信息里、提示词库、搭配、模型的触发词和介绍、你自己改画布的记录里都一样（只有节点的“文字输入”被换掉，模型名、数字、选项照常给，第三方节点也一样）。AI 拿代号去跑图、写提示词、放挑图面板，AMB 再换回原文，所以它照样能干活，你那边看到的也都是原样。要新的提示词时，AI 会放一个 **提示词框**：你在里面写几条提示词、反向提示词，可以从图库选或拖进参考图，原文只存在 AMB 里；框上方还会按画布主模型的大小提醒你（小模型手容易画坏，可以选简单动作或藏手，只是可选）。新增 skill“隐私模式·怎么干活”，隐私任务里的 AI 开工时会被指过去；隐私任务里 AI 只拿得到这一个 skill，**做法** 面板的 skill 一栏也只显示它，关掉隐私后才换回普通的 skill。**关卡** 窗口里设新任务默认开不开；打开、关掉都记在时间线上并告诉 AI。所有工具的结果出门前都过同一道关，测试会在每个工具的结果里埋“暗号”提示词和图片名，确认一个都漏不出去。

api/skills/anima_illustration.json

@@ -1,16 +1,18 @@
 {
-  "short": "Anima illustration with a high-detail Tile-guided second pass",
-  "title": {"zh": "Anima 插画·底图与 Tile 细化", "en": "Anima illustration: base and Tile detail"},
+  "short": "Anima illustration: base first, then the Tile way",
+  "title": {
+    "zh": "Anima 插画·底图与 Tile 细化",
+    "en": "Anima illustration: base and Tile detail"
+  },
   "about": {
-    "zh": "按已测试的 Anima 插画路线先出底图，再用 Tile 控制网放大细化；先确认模型、LoRA 和结果，不保证原图复刻。",
-    "en": "Make an Anima illustration and refine a selected base with Tile guidance, checking models, LoRA and results; not a promise to reproduce a reference."
+    "zh": "Anima 插画分两步时怎么判断：先出底图给你看，再用“Anima Tile 放大细化”这块做法细化；你存过这条路线的工作流或搭配就直接用；你没选的画风、LoRA、画师词不加。具体数值在做法和你自己的搭配里，这里不写。",
+    "en": "How to judge an Anima illustration in two passes: show the base first, then refine it with the \"Anima Tile detail upscale\" way; a workflow or combo you saved for this route is used as it is; no style, LoRA or artist you did not pick. The values live in the way and in your own combos, not here."
   },
   "text": [
-    "Use for an explicitly requested Anima illustration or this Tile-guided two-pass recipe, not as a default for unrelated models or styles. Preserve the user's text, requested character and pose; do not add an artist, style, LoRA or hidden hands unless the user asks or accepts it.",
-    "Starting point tested on RTX 4050 Laptop 6 GB: MiaoMiao Harem Anima 1.5, compatible Qwen 3 0.6B text encoder and Qwen image VAE, optional Chen bin style LoRA at model/CLIP 1.0. Verify models and LoRA compatibility; filename similarity is not file identity. The separately named MiaoMiao text encoder was not compared. Use the user's chosen model or style when specified.",
-    "Base pass: 832x1216 batch 1, 35 steps, CFG 4.4, dpmpp_2m_sde_gpu/beta, denoise 1.0, with tiled VAE if needed. Preserve and show the base; let the user select it before a further pass when selection was requested. Do not force a long second pass on an unwanted base.",
-    "For an Anima canvas with the tested Anima Tile Repair v2.0 weight and ComfyUI-Anima-LLLite available, add_nodes(block=\"upscale_anima_tile\") builds the second pass: tiled VAE encode, bislerp 1.22, 30 steps, CFG 4.4, dpmpp_2m_sde_gpu/sgm_uniform, denoise 0.85, Tile 0.7 from 0 to 0.8 using the base image, tiled decode and optional ColorMatch mkl. It saves a separate result. Check the canvas and missing nodes/models first; do not silently substitute another Tile model or attach to a non-Anima model.",
-    "Use a modest single-image size and avoid simultaneous large jobs on 6 GB. 0.85 redraws substantially even with Tile: compare base and result, zoom in on important details and hands, keep the original, and let the user choose; lower denoise if preservation matters more. Three locally tested SFW base/Tiled comparisons suggest stronger composition retention, not guaranteed hand repair or a general success rate. The original reference graph also contains a Tile preprocessor that was not used in this test; its actual runtime and exact reference reproduction are unproven.",
-    "If the user asks for a controlled comparison, use the same saved base, prompt, size, seed, sampler and color matching with and without Tile. Report what actually ran, missing models, choices, and visual changes. Do not claim the LoRA alone caused the style: no weight ablation has been done."
+    "Use for an Anima illustration the user wants refined in two passes, not as a default for other models or styles. Preserve the user's text, requested character and pose; add a style, artist or LoRA only when the user picked it (a combo or saved prompt of theirs).",
+    "If the user saved this route as a workflow (list_workflows) or a combo (list_combos), use it: run_workflow with their subject and seed is the daily run. Its models and values are theirs: do not change them.",
+    "Run the base first and show it; when the user wants to choose, let them pick it before the second pass. Then add_nodes(block=\"upscale_anima_tile\") on the Anima canvas: the way carries the tested second-pass values and checks the model family and the Tile weight. Report a missing node or model; never swap in a look-alike.",
+    "What to expect: a high-denoise second pass without Tile tends to change the pose and ornaments; with Tile the composition holds better, but faces, hands and small details are still redrawn. Keep the base, compare, and lower the denoise when keeping more matters to the user.",
+    "Changing models between the passes: only within one family sharing the text encoder and VAE; otherwise say it is untried. These findings come from a few single-seed comparisons: say what ran and what changed, and do not promise a reproduction or repaired hands."
   ]
 }

docs/guides/mcp.md

@@ -150,7 +150,7 @@ claude mcp add --transport http anomalous http://127.0.0.1:8188/anomalous/mcp
 
 **参考做法**：有几件常见的活，AMB 带着现成的做法，是从擅长这些活的工作流里提炼出来的：**局部重绘·裁切缝合**（圈出的地方裁出来放大到 1024 重画再缝回去，需要节点包 ComfyUI-Inpaint-CropAndStitch）、**局部重绘（官方节点）**（没装那个包时用）、**局部重绘·LanPaint**（融合最好、任何模型都行，但慢好几倍，需要节点包 LanPaint）、**放大**（1.25 倍再轻轻重画一遍，只用自带节点，另存一张，文件名以 upscaled 开头）、**分块放大**（2 倍，需要节点包 ComfyUI_UltimateSDUpscale）、**照着姿势画**和**照着构图画**（你递给 AI 一张图，AMB 读出姿势或深度，挑一个和底模对得上的控制网，需要节点包 comfyui_controlnet_aux）、**迭代放大**（分 3 小步放大到 1.3 倍，用 Impact Pack）、**双模型放大**（重画换一个更稳的模型）、**背景移除**（需要节点包 comfyui-inspyrenet-rembg）、**读图出标签**（需要节点包 ComfyUI-WD14-Tagger，AI 跑完会收到读出的标签）、**部位细化（脸和手）**（需要 Impact Pack、Impact Subpack 和 models/ultralytics/bbox 里的检测模型，细化后的图另存一张，文件名以 detailed 开头）、**修手·先放大再修**（你递给 AI 一张图：整张放大到约 400 万像素轻轻重画，原图越小放得越大，最多 2 倍，让手变大，再逐只修手，另存一张，文件名以 hands_fixed 开头；Anima 模型上有 Anima Turbo LoRA 时只跑 10 步，1280×832 的图在 6GB 显卡上约 100 秒。AI 一次出 4 张，自己先放大看手，把手指数不对、粘连、大小不对的去掉，都不行就再出 4 张，最多两轮，再把好的放进只放大手那块的面板让你挑）。AI 工作台右边的**做法**面板列着全部做法：缺节点包或模型的变灰并写明缺什么，每块有个开关，关掉的 AI 就看不到。AI 开工时只收到开着又能用的那些；它说用哪块，AMB 就把那几个节点搭好、接进画布上那个出图的工作流里（采样器、VAE 解码、保存图片都由 AMB 按插口认出来），一次就好，也是一步 Ctrl+Z。画布上有好几个采样器时，AI 得指明用哪一个；缺节点包时，它会知道缺哪个、该用哪块代替。
 
-**skill**：和做法并列，是 AI 遇到某类活时按需读取的建议；做法负责自动搭节点。AI 开工时只看到每个 skill 的名字和一句话，真要做那类活时才用 `get_toolbox(skill=名字)` 取全文，不增加工具数量。**写提示词·动作适合模型**按画布主模型的大小给动作建议，用户指定的姿势和提示词照原样；**Anima 插画·底图与 Tile 细化**记录经过三组本机对照的底图及高降噪细化路线，可调用 `add_nodes(block="upscale_anima_tile")` 自动接入放大做法。它需要确认画布是 Anima、装有 ComfyUI-Anima-LLLite 和 `animaTileRepair_v20.safetensors`；保留底图，比较脸、手及细节，不能保证复刻原图。做法面板最下面列着 skill，每个有开关，关掉的 AI 就拿不到。AI 每取一个 skill，时间线上多一行 **Skill**；之后那次跑图下面也写着参照了哪个。隐私任务只提供隐私 skill，不提供普通 skill。
+**skill**：和做法并列，是 AI 遇到某类活时按需读取的建议；做法负责自动搭节点。AI 开工时只看到每个 skill 的名字和一句话，真要做那类活时才用 `get_toolbox(skill=名字)` 取全文，不增加工具数量。**写提示词·动作适合模型**按画布主模型的大小给动作建议，用户指定的姿势和提示词照原样；**Anima 插画·底图与 Tile 细化**只写怎么判断：先出底图给你看，再用 `add_nodes(block="upscale_anima_tile")` 接入放大做法；你把这条路线存成过工作流或搭配，AI 就直接用你的；具体模型和数值在做法和你的搭配里，不写进 skill。它需要确认画布是 Anima、装有 ComfyUI-Anima-LLLite 和 `animaTileRepair_v20.safetensors`；保留底图，比较脸、手及细节，不能保证复刻原图。做法面板最下面列着 skill，每个有开关，关掉的 AI 就拿不到。AI 每取一个 skill，时间线上多一行 **Skill**；之后那次跑图下面也写着参照了哪个。隐私任务只提供隐私 skill，不提供普通 skill。
 
 **先对一下重画的范围**：圈得不准也没关系。AI 准备只重画标出的地方时，会先在展台把 **你圈的** 和 **AI 调整后** 的范围并排给你看（红色是要重画的地方；AI 可以挪位置、改成方框或椭圆、往外扩、把边缘做软，图下写着外扩和羽化了多少像素）。点 **用 AI 调整的**、**用我圈的** 或 **我重新圈**，也可以在下面写一句（比如“再往手腕多圈一点”），AI 会照着再调一次；点 **让 AI 定** 就用 AI 调整的。选好以后 AMB 才把带这个范围的 `anomalous_inpaint_….png` 放进 input 文件夹，节点怎么搭由 AI 自己来。