dsh-image-gen 0.9.2 → 0.9.3

package/README.en.md

@@ -339,6 +339,8 @@ Bring private image generation on your local GPU directly into Agent conversatio
 > Official OpenAI size presets can be customized with `openaiSizes`, shaped as ratio → tier → pixels, e.g. `{ "16:9": { "2K": "2048x1152", "4K": "3840x2160" } }`. For the gpt-image-2 family, dimensions must be multiples of 16, the longest edge at most 3840, aspect ratio within 1:3–3:1, and total pixels within 655360–8294400. Outputs above 3686400 pixels have experimental upstream support. See the [official API reference](https://developers.openai.com/api/reference/resources/images/methods/generate). Older GPT Image models retain their three standard sizes.
 > Compatible providers use `openaiCompatSizes` in the same format; an empty table retains the legacy sizes. `openaiCompatQuality` sets the default generation quality (empty means omitted), and `openaiCompatQualityOptions` sets Studio choices. Choosing “Provider default (omit)” omits quality for that request. Conversation tools also accept independent `size` and `quality` parameters.
 > “Generation extra fields” maps to `openaiCompatExtraBody`, e.g. `{ "watermark": false }`. These fields apply only to generation, merge after standard fields, and can override matching keys; `false` and nested objects are preserved. Editing continues to use `openaiCompatEditExtra`. For JSON image URL array edits, `openaiCompatEditSizeMode` defaults to `auto`; set it to `selected` to send the selected dimensions. EditExtra can still override size.
+> Compatible channels can set a separate editing model with `openaiCompatEditModel`; empty reuses the generation model, and an explicit conversation-tool `model` still overrides one edit. Official SiliconFlow `/v1` endpoints automatically use `siliconflowJson`; relays implementing the same contract can select it manually. Edits post an `image` string to `/images/generations`. `Qwen/Qwen-Image-Edit` supports one reference; `Qwen/Qwen-Image-Edit-2509` supports up to three using `image`, `image2`, and `image3` in order. Generic size and quality fields are omitted; `size`, `image_size`, or `quality` in editing extras produce an actionable error, and the plugin fills reference fields. See the [SiliconFlow API reference](https://docs.siliconflow.cn/docs/api/images-generations-post).
+> Ordinary conversation requests should omit provider and model to use the saved default. The model receives live default-provider and parameter instructions. Omit unused optional fields; blank dimensions are treated as absent, while two nonempty `size` / `aspect_ratio` values still conflict. ChatGPT subscription does not accept `image_size`; use `size` or `aspect_ratio` rather than a tier such as `1K`.
 > The workbench shows “Visible watermark: Channel default / Off / On” beside ratio and resolution, checking generation and editing separately. Seedream, native DashScope Qwen Image, and recognized official SenseNova and Qwen compatible endpoints enable it automatically; recognized compatible endpoints also select their JSON edit contracts automatically. SenseNova matches `https://token.sensenova.cn/v1` and `sensenova-u1.5-lite`; Qwen matches official workspace hosts under `/compatible-mode/v1` with Qwen Image 3.0/3.0 Pro or 2.1 Pro/Turbo. Other compatible channels can bind a boolean `watermark`, `watermark_enabled`, or `parameters.watermark` through the matching operation's extra fields; editing requires a JSON format. Unknown channels and Zhipu display a disabled “Support unconfirmed” control; interfaces without a confirmed switch also show an explanation. A model name alone does not identify an unknown relay. Channel default preserves configured fields or omits the parameter; an explicit choice overrides only the watermark field and is retained for batches, comparison models supporting the operation, and regeneration.
 
 </details>

package/README.md

@@ -339,6 +339,8 @@ pnpm dsh plugin --profile web add ./dsh-image-gen
 > OpenAI 官方可通过设置中的 `openaiSizes` 自定义尺寸表，格式为「比例 → 档位 → 像素」，例如 `{ "16:9": { "2K": "2048x1152", "4K": "3840x2160" } }`。gpt-image-2 系列按官方限制校验：边长为 16 的倍数、最大 3840、比例在 1:3 到 3:1、总像素 655360–8294400；超过 3686400 像素的输出属于上游实验支持。[官方说明](https://developers.openai.com/api/reference/resources/images/methods/generate)。旧版 GPT Image 模型仍只使用三种标准尺寸。
 > OpenAI Compatible 的尺寸表为 `openaiCompatSizes`，格式相同；留空时保持原来的标准尺寸。`openaiCompatQuality` 设置默认生成质量（留空不发送），`openaiCompatQualityOptions` 设置 Studio 可选值。选择「渠道默认（不发送）」可在单次请求中省略 quality；对话工具也支持独立的 `size` 和 `quality` 参数。
 > 兼容渠道的「文生图额外字段」对应 `openaiCompatExtraBody`，例如 `{ "watermark": false }`。它仅用于文生图，在常规字段之后合并，可覆盖同名参数；`false` 和嵌套对象都会保留。图生图继续使用 `openaiCompatEditExtra`；JSON 图片 URL 数组格式的 `openaiCompatEditSizeMode` 默认 `auto`，改为 `selected` 才发送所选尺寸，EditExtra 仍可覆盖该尺寸。
+> 兼容渠道可设置独立的「图生图模型」`openaiCompatEditModel`，留空沿用文生图模型；对话中的显式 `model` 仍可覆盖单次编辑。硅基流动官方 `/v1` 地址自动使用 `siliconflowJson` 编辑格式，其他采用同一协议的中转可手动选择：通过 `/images/generations` 发送 `image` 字符串。`Qwen/Qwen-Image-Edit` 支持一张参考图，`Qwen/Qwen-Image-Edit-2509` 最多三张，依次使用 `image`、`image2`、`image3`。编辑不发送通用尺寸和质量字段；图生图额外字段中的 `size`、`image_size`、`quality` 会给出明确错误，图片字段由插件填入。协议以[硅基流动官方文档](https://docs.siliconflow.cn/docs/api/images-generations-post)为准。
+> 普通对话生图无需指定 provider 或 model，会使用已保存的默认渠道；模型可读取随配置更新的默认渠道及参数提示。无关可选字段应省略，空白尺寸会按未提供处理；两个有效的 `size` / `aspect_ratio` 仍互斥。ChatGPT 订阅不支持 `image_size`，请使用 `size` 或 `aspect_ratio`，不要填入 `1K` 等档位。
 > 工作台在比例、分辨率旁提供「可见水印：渠道默认 / 关闭 / 开启」，分别判断文生图和图生图支持情况。Seedream、通义原生 Qwen Image，以及已确认的商汤和通义官方兼容接口自动启用；官方兼容接口同时自动使用对应的 JSON 编辑格式。商汤匹配 `https://token.sensenova.cn/v1` 与 `sensenova-u1.5-lite`；通义匹配官方工作空间域名的 `/compatible-mode/v1` 与 Qwen Image 3.0/3.0 Pro、2.1 Pro/Turbo。其他兼容渠道可通过对应操作的额外字段绑定布尔值 `watermark`、`watermark_enabled` 或 `parameters.watermark`；图生图绑定仅用于 JSON 编辑格式。未知渠道和智谱显示禁用的「支持未确认」，无已确认开关的接口也显示说明。不会仅凭模型名识别未知中转。渠道默认保留原配置或省略参数，单次选择只覆盖水印字段；批量生成、支持该操作的对比模型及再次生成会保留选择。
 
 </details>

package/lib/client.js

@@ -75,7 +75,7 @@ window.__ModuleLoader__.load({
 		//#region lib/types/watermark.js
 		/** Model names alone do not establish the protocol of an unknown relay. */
 		function officialContract(config, mode) {
-			const model = (mode === "generate" ? config.openaiCompatExtraBody : config.openaiCompatEditExtra)?.model ?? config.openaiCompatModel;
+			const model = (mode === "generate" ? config.openaiCompatExtraBody : config.openaiCompatEditExtra)?.model ?? (mode === "edit" ? config.openaiCompatEditModel?.trim() || config.openaiCompatModel : config.openaiCompatModel);
 			if (typeof model !== "string") return void 0;
 			try {
 				const url = new URL(config.openaiCompatBaseURL ?? "");
@@ -87,6 +87,10 @@ window.__ModuleLoader__.load({
 		}
 		/** Known official edit contracts take precedence over relay-specific format settings. */
 		function compatEditFormat(config) {
+			try {
+				const url = new URL(config.openaiCompatBaseURL ?? "");
+				if (url.protocol === "https:" && !url.port && !url.search && !url.hash && (url.hostname === "api.siliconflow.cn" || url.hostname === "api.siliconflow.com") && url.pathname.replace(/\/+$/, "") === "/v1") return "siliconflowJson";
+			} catch {}
 			const contract = officialContract(config, "edit");
 			if (contract === "sensenova") return "jsonImageUrlArray";
 			if (contract === "qwen") return "qwenJson";
@@ -181,6 +185,10 @@ window.__ModuleLoader__.load({
 			"off",
 			"on"
 		];
+		/** A model picker/request must use the same operation-specific default. */
+		function studioModelForMode(profile, mode) {
+			return mode === "edit" ? profile.editModel ?? profile.model : profile.model;
+		}
 		function studioWatermarkCapability(profile, mode) {
 			return profile?.watermarkCapabilities?.[mode] ?? (mode === "generate" && profile?.watermarkOptions?.length ? { status: "supported" } : {
 				status: "unknown",
@@ -117762,6 +117770,7 @@ img.ProseMirror-separator {
 				};
 			}, []);
 			const activeProfile = (0, react.useMemo)(() => config?.providers.find((item) => item.provider === provider), [config, provider]);
+			const requestModel = mode === "edit" && activeProfile?.editModel ? studioModelForMode(activeProfile, mode) : model;
 			const configuredCount = config?.providers.filter((item) => item.configured).length ?? 0;
 			const apiConfiguredCount = config?.providers.filter((item) => item.configured && !isSubscriptionProvider(item.provider)).length ?? 0;
 			const subscriptionSignedInCount = config?.providers.filter((item) => item.configured && isSubscriptionProvider(item.provider)).length ?? 0;
@@ -118105,7 +118114,7 @@ img.ProseMirror-separator {
 								body: JSON.stringify({
 									mode,
 									provider: target.profile.provider,
-									model: target.profile.model,
+									model: studioModelForMode(target.profile, mode),
 									prompt: prompt.trim(),
 									ratio: target.ratio,
 									quality: target.quality,
@@ -118174,7 +118183,7 @@ img.ProseMirror-separator {
 						body: JSON.stringify({
 							mode,
 							provider,
-							model,
+							model: requestModel,
 							prompt: prompt.trim(),
 							ratio,
 							quality: selectedQuality,
@@ -118759,7 +118768,7 @@ img.ProseMirror-separator {
 																className: p.configured ? "is-configured" : "is-missing",
 																children: [(0, react_jsx_runtime.jsxs)("div", {
 																	className: "dsh-ig-status-item-name",
-																	children: [(0, react_jsx_runtime.jsx)("strong", { children: p.label }), (0, react_jsx_runtime.jsx)("small", { children: p.model })]
+																	children: [(0, react_jsx_runtime.jsx)("strong", { children: p.label }), (0, react_jsx_runtime.jsx)("small", { children: studioModelForMode(p, mode) })]
 																}), (0, react_jsx_runtime.jsx)("span", {
 																	className: "dsh-ig-status-item-tag",
 																	children: isSubscriptionProvider(p.provider) ? p.configured ? t("subSignedIn") : t("subSignedOut") : p.configured ? t("configured") : t("unconfigured")
@@ -118883,7 +118892,7 @@ img.ProseMirror-separator {
 													children: (0, react_jsx_runtime.jsx)(Sparkles, { size: 28 })
 												}),
 												(0, react_jsx_runtime.jsx)("strong", { children: comparisonEnabled ? t("comparing", { count: String(comparisonTargets.length) }) : count > 1 ? t("generatingCount", { count: String(count) }) : t("generating") }),
-												(0, react_jsx_runtime.jsx)("span", { children: comparisonEnabled ? comparisonTargets.map((target) => target.profile.label).join(" · ") : `${activeProfile?.label ?? ""} · ${model}` })
+												(0, react_jsx_runtime.jsx)("span", { children: comparisonEnabled ? comparisonTargets.map((target) => target.profile.label).join(" · ") : `${activeProfile?.label ?? ""} · ${requestModel}` })
 											]
 										}) : currentBatch && currentBatch.length > 1 ? (0, react_jsx_runtime.jsx)("div", {
 											className: "dsh-ig-canvas-cluster",
@@ -119197,8 +119206,8 @@ img.ProseMirror-separator {
 																	(0, react_jsx_runtime.jsxs)("span", {
 																		className: "dsh-ig-model-copy",
 																		children: [(0, react_jsx_runtime.jsx)("strong", { children: item.label }), (0, react_jsx_runtime.jsx)("small", {
-																			title: item.model,
-																			children: item.model
+																			title: studioModelForMode(item, mode),
+																			children: studioModelForMode(item, mode)
 																		})]
 																	}),
 																	target && (0, react_jsx_runtime.jsxs)("span", {
@@ -119267,11 +119276,11 @@ img.ProseMirror-separator {
 													}))
 												}), (0, react_jsx_runtime.jsx)(FieldSelect, {
 													label: t("model"),
-													value: model,
+													value: requestModel,
 													onChange: setModel,
 													options: activeProfile === void 0 ? [] : [{
-														value: activeProfile.model,
-														label: activeProfile.model
+														value: studioModelForMode(activeProfile, mode),
+														label: studioModelForMode(activeProfile, mode)
 													}]
 												})]
 											}),
@@ -125436,7 +125445,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 `;
 		//#endregion
 		//#region lib/types/client/index.js
-		console.info(`[dsh-image-gen] canvas bundle 2026-10-19T16:02:35.438Z`);
+		console.info(`[dsh-image-gen] canvas bundle 2026-10-26T00:24:12.240Z`);
 		/** Right-sidebar tab-type registration id (DSH 0.1.5 `sidebarRightTabs`). */
 		const SIDEBAR_STUDIO_TAB_ID = "dsh-image-gen/studio";
 		/** The tab kind: a page type opened by kind, recognizing no resource address. */
@@ -125519,11 +125528,15 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 				endpointHintOpenAI: "官方 api.openai.com 的 /v1 地址；中转站请使用下方「OpenAI 兼容」行。",
 				endpointHintOpenAICompat: "中转站/自建服务的 OpenAI 兼容 /v1 地址（必填），例如 https://your-relay.example.com/v1。",
 				editFormat: "图生图请求形态",
+				editModel: "图生图模型（可选）",
+				editModelHint: "留空沿用上方模型。硅基流动可分别填写 Qwen/Qwen-Image（文生图）和 Qwen/Qwen-Image-Edit（图生图）。",
+				editFormatSiliconFlow: "硅基流动 JSON（image 字符串）",
+				editFormatSiliconFlowHint: "编辑尺寸由参考图决定，不发送尺寸和质量字段。Qwen-Image-Edit 支持一张图，Edit-2509 最多三张；参考图字段由插件填入。",
 				editFormatMultipart: "标准 multipart（OpenAI 官方）",
 				editFormatAutomatic: "自动：官方接口 JSON 格式",
 				editFormatJsonImageUrlArray: "JSON images 数组（商汤等中转）",
 				editFormatFormReferenceImages: "multipart + reference_images JSON 数组（base64）",
-				editFormatHint: "商汤、通义已确认的官方接口自动使用对应 JSON 格式；其他中转按接口文档选择。文生图不受影响。",
+				editFormatHint: "商汤、通义、硅基流动已确认的官方接口自动使用对应 JSON 格式；其他中转按接口文档选择。文生图不受影响。",
 				editExtra: "附加 JSON 字段",
 				editExtraPlaceholder: "{\"watermark\": false, \"prompt_extend\": true}",
 				editExtraHint: "仅 JSON 形态生效；会合并到请求体末尾，可覆盖默认字段。留空表示不附加。",
@@ -125689,11 +125702,15 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 				endpointHintOpenAI: "Official api.openai.com /v1 base URL; for relays use the \"OpenAI-compatible\" row below.",
 				endpointHintOpenAICompat: "OpenAI-compatible /v1 base URL of your relay or self-hosted service (required), e.g. https://your-relay.example.com/v1.",
 				editFormat: "Edit request format",
+				editModel: "Editing model (optional)",
+				editModelHint: "Leave empty to reuse the model above. SiliconFlow can use Qwen/Qwen-Image for generation and Qwen/Qwen-Image-Edit for editing.",
+				editFormatSiliconFlow: "SiliconFlow JSON (image string)",
+				editFormatSiliconFlowHint: "Edit dimensions follow the reference; size and quality fields are omitted. Qwen-Image-Edit supports one reference, Edit-2509 up to three. The plugin fills the image fields.",
 				editFormatMultipart: "Standard multipart (official OpenAI)",
 				editFormatAutomatic: "Automatic: official JSON contract",
 				editFormatJsonImageUrlArray: "JSON images array (SenseNova etc.)",
 				editFormatFormReferenceImages: "Multipart with a reference_images JSON array (base64)",
-				editFormatHint: "Recognized official SenseNova and Qwen endpoints automatically use their JSON edit contracts. Choose other relay formats according to their documentation. Text-to-image is unaffected.",
+				editFormatHint: "Recognized official SenseNova, Qwen and SiliconFlow endpoints automatically use their JSON edit contracts. Choose other relay formats according to their documentation. Text-to-image is unaffected.",
 				editExtra: "Extra JSON fields",
 				editExtraPlaceholder: "{\"watermark\": false, \"prompt_extend\": true}",
 				editExtraHint: "Only used with the JSON format; merged into the request body last and may override defaults. Leave empty for none.",
@@ -126328,6 +126345,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 			return {
 				expanded: false,
 				model: "",
+				editModel: "",
 				baseURL: "",
 				googleNative: false,
 				keyInput: "",
@@ -126410,7 +126428,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 		}
 		/** Persisted or selected edit-format value → row state; unknown values fall back to standard multipart. */
 		function parseEditFormat(value) {
-			return value === "jsonImageUrlArray" || value === "formReferenceImages" ? value : "multipart";
+			return value === "jsonImageUrlArray" || value === "formReferenceImages" || value === "siliconflowJson" ? value : "multipart";
 		}
 		/** Build one row per provider from persisted settings, including ComfyUI extras. */
 		function rowsFromSettings(value) {
@@ -126427,6 +126445,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 					timeoutSeconds: Math.max(1, Math.round((value?.comfyuiTimeoutMs ?? 3e5) / 1e3))
 				} : {},
 				...provider === "openai-compat" ? {
+					editModel: value?.openaiCompatEditModel ?? "",
 					editFormat: parseEditFormat(value?.openaiCompatEditFormat),
 					editExtraText: editExtraTextOf(value?.openaiCompatEditExtra),
 					sizeTableText: editExtraTextOf(value?.openaiCompatSizes),
@@ -126679,6 +126698,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 					for (const provider of IMAGE_PROVIDERS) next[provider] = {
 						...current[provider],
 						model: modelOf(provider, value),
+						...provider === "openai-compat" ? { editModel: value?.openaiCompatEditModel ?? "" } : {},
 						baseURL: baseURLOf(provider, value),
 						...provider === "comfyui" ? {
 							workflows: resolveComfyUIWorkflows(value ?? {}),
@@ -126809,6 +126829,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 						if (provider === "google") await saveSetting("googleNative", row.googleNative);
 						if (provider === "openai") await saveSetting("openaiSizes", sizeTable);
 						if (provider === "openai-compat") {
+							await saveSetting("openaiCompatEditModel", row.editModel.trim());
 							await saveSetting("openaiCompatEditFormat", row.editFormat);
 							await saveSetting("openaiCompatEditExtra", editExtra);
 							await saveSetting("openaiCompatSizes", sizeTable);
@@ -127160,7 +127181,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 				const row = rows[provider];
 				const officialEditFormat = provider === "openai-compat" ? compatEditFormat({
 					openaiCompatBaseURL: row.baseURL,
-					openaiCompatModel: row.model
+					openaiCompatModel: row.editModel.trim() || row.model
 				}) : "multipart";
 				const automaticEditFormat = officialEditFormat !== "multipart";
 				const editFormat = automaticEditFormat ? officialEditFormat : row.editFormat;
@@ -127308,6 +127329,28 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 									}) : null
 								]
 							}),
+							provider === "openai-compat" ? (0, react_jsx_runtime.jsxs)("label", {
+								className: "dsh-ig-field",
+								children: [
+									(0, react_jsx_runtime.jsx)("span", {
+										className: "dsh-ig-label",
+										children: t("editModel")
+									}),
+									(0, react_jsx_runtime.jsx)("input", {
+										className: "dsh-ig-input",
+										value: row.editModel,
+										onChange: (event) => {
+											updateRow(provider, { editModel: event.target.value });
+										},
+										list: "dsh-ig-openai-compat-model-options",
+										disabled: !settingsWritable
+									}),
+									(0, react_jsx_runtime.jsx)("span", {
+										className: "dsh-ig-hint",
+										children: t("editModelHint")
+									})
+								]
+							}) : null,
 							provider === "openai-compat" ? (0, react_jsx_runtime.jsxs)("label", {
 								className: "dsh-ig-field",
 								children: [
@@ -127338,12 +127381,16 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 											(0, react_jsx_runtime.jsx)("option", {
 												value: "formReferenceImages",
 												children: t("editFormatFormReferenceImages")
+											}),
+											(0, react_jsx_runtime.jsx)("option", {
+												value: "siliconflowJson",
+												children: t("editFormatSiliconFlow")
 											})
 										]
 									}),
 									(0, react_jsx_runtime.jsx)("span", {
 										className: "dsh-ig-hint",
-										children: t("editFormatHint")
+										children: t(editFormat === "siliconflowJson" ? "editFormatSiliconFlowHint" : "editFormatHint")
 									})
 								]
 							}) : null,
@@ -127516,7 +127563,7 @@ body[data-ds-dark-theme] .dsh-ig-workbench{--ig-danger:#f08c8c;--ig-warn:#e5b76a
 									]
 								})
 							] }) : null,
-							provider === "openai-compat" && (editFormat === "jsonImageUrlArray" || editFormat === "qwenJson") ? (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [editFormat === "jsonImageUrlArray" ? (0, react_jsx_runtime.jsxs)("label", {
+							provider === "openai-compat" && (editFormat === "jsonImageUrlArray" || editFormat === "qwenJson" || editFormat === "siliconflowJson") ? (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [editFormat === "jsonImageUrlArray" ? (0, react_jsx_runtime.jsxs)("label", {
 								className: "dsh-ig-field",
 								children: [(0, react_jsx_runtime.jsx)("span", {
 									className: "dsh-ig-label",

package/lib/index.js

@@ -5604,7 +5604,7 @@ function configuredField(extra) {
 }
 /** Model names alone do not establish the protocol of an unknown relay. */
 function officialContract(config, mode) {
-	const model = (mode === "generate" ? config.openaiCompatExtraBody : config.openaiCompatEditExtra)?.model ?? config.openaiCompatModel;
+	const model = (mode === "generate" ? config.openaiCompatExtraBody : config.openaiCompatEditExtra)?.model ?? (mode === "edit" ? config.openaiCompatEditModel?.trim() || config.openaiCompatModel : config.openaiCompatModel);
 	if (typeof model !== "string") return void 0;
 	try {
 		const url = new URL(config.openaiCompatBaseURL ?? "");
@@ -5616,6 +5616,10 @@ function officialContract(config, mode) {
 }
 /** Known official edit contracts take precedence over relay-specific format settings. */
 function compatEditFormat(config) {
+	try {
+		const url = new URL(config.openaiCompatBaseURL ?? "");
+		if (url.protocol === "https:" && !url.port && !url.search && !url.hash && (url.hostname === "api.siliconflow.cn" || url.hostname === "api.siliconflow.com") && url.pathname.replace(/\/+$/, "") === "/v1") return "siliconflowJson";
+	} catch {}
 	const contract = officialContract(config, "edit");
 	if (contract === "sensenova") return "jsonImageUrlArray";
 	if (contract === "qwen") return "qwenJson";
@@ -5827,6 +5831,10 @@ const STUDIO_WATERMARK_VALUES = [
 	"off",
 	"on"
 ];
+/** A model picker/request must use the same operation-specific default. */
+function studioModelForMode(profile, mode) {
+	return mode === "edit" ? profile.editModel ?? profile.model : profile.model;
+}
 function studioWatermarkCapability(profile, mode) {
 	return profile?.watermarkCapabilities?.[mode] ?? (mode === "generate" && profile?.watermarkOptions?.length ? { status: "supported" } : {
 		status: "unknown",
@@ -5961,6 +5969,7 @@ const Config = z.object({
 	openaiSizes: volatile(z.dict(z.dict(z.string())).default({})),
 	openaiCompatBaseURL: volatile(z.string().default("")),
 	openaiCompatModel: volatile(z.string().default("")),
+	openaiCompatEditModel: volatile(z.string().default("")),
 	openaiCompatQuality: volatile(z.string().default("")),
 	openaiCompatQualityOptions: volatile(z.array(z.string()).default([
 		"auto",
@@ -5973,7 +5982,8 @@ const Config = z.object({
 	openaiCompatEditFormat: volatile(z.union([
 		z.const("multipart"),
 		z.const("jsonImageUrlArray"),
-		z.const("formReferenceImages")
+		z.const("formReferenceImages"),
+		z.const("siliconflowJson")
 	]).default("multipart")),
 	openaiCompatEditExtra: volatile(z.dict(z.any()).default({})),
 	openaiCompatSizes: volatile(z.dict(z.dict(z.string())).default({})),
@@ -6101,13 +6111,12 @@ function resolveProvider(config) {
 * Apply a per-call provider and/or model override on top of the saved config.
 * `model` is ignored for ComfyUI, whose per-call equivalent is `workflow`.
 */
-function withProviderOverrides(config, provider, model) {
+function withProviderOverrides(config, provider, model, mode = "generate") {
 	const base = provider === void 0 ? { ...config } : {
 		...config,
 		provider
 	};
-	if (model === void 0) return base;
-	const trimmed = model.trim();
+	const trimmed = model?.trim() || (mode === "edit" && base.provider === "openai-compat" ? base.openaiCompatEditModel?.trim() : "") || "";
 	if (trimmed.length === 0) return base;
 	switch (base.provider ?? "google") {
 		case "google": return {
@@ -6120,7 +6129,8 @@ function withProviderOverrides(config, provider, model) {
 		};
 		case "openai-compat": return {
 			...base,
-			openaiCompatModel: trimmed
+			openaiCompatModel: trimmed,
+			...mode === "edit" ? { openaiCompatEditModel: trimmed } : {}
 		};
 		case "seedream": return {
 			...base,
@@ -8263,6 +8273,39 @@ async function generateOpenAICompatibleImage(input) {
 	}), input.apiKey), input.provider, input);
 }
 async function editOpenAICompatibleImage(input) {
+	if (input.editFormat === "siliconflowJson") {
+		const body = {
+			model: input.model,
+			prompt: input.prompt,
+			...input.editExtra ?? {},
+			...input.extraBody ?? {}
+		};
+		const model = typeof body.model === "string" ? body.model.trim() : "";
+		if (!model) throw new Error("硅基流动图生图需要编辑模型");
+		body.model = model;
+		if (model === "Qwen/Qwen-Image") throw new Error("Qwen/Qwen-Image 是文生图模型；请配置图生图模型，例如 Qwen/Qwen-Image-Edit");
+		const maxReferences = model === "Qwen/Qwen-Image-Edit-2509" ? 3 : 1;
+		if (input.sourceImages.length < 1 || input.sourceImages.length > maxReferences) throw new Error(`${model} 图生图支持 1–${String(maxReferences)} 张参考图，当前 ${String(input.sourceImages.length)} 张；请精简参考图或选择支持多图的编辑模型`);
+		if ([
+			"size",
+			"image_size",
+			"quality"
+		].some((key) => key in body)) throw new Error("此硅基流动图生图格式不发送 size、image_size 或 quality；请从图生图额外字段中移除这些字段");
+		delete body.image;
+		delete body.image2;
+		delete body.image3;
+		for (const [index, source] of input.sourceImages.entries()) body[index === 0 ? "image" : `image${String(index + 1)}`] = toDataUrl$2(source);
+		return parseImageResponse(await guardedFetch("siliconflow image edit request", input.signal, () => fetch(imageEndpoint$1(input.baseURL, "generations"), {
+			method: "POST",
+			redirect: "error",
+			signal: input.signal,
+			headers: {
+				authorization: `Bearer ${input.apiKey}`,
+				"content-type": "application/json"
+			},
+			body: JSON.stringify(body)
+		}), input.apiKey), "siliconflow", input);
+	}
 	if (input.editFormat === "qwenJson") return parseImageResponse(await guardedFetch("openai-compat image edit request", input.signal, () => fetch(imageEndpoint$1(input.baseURL, "generations"), {
 		method: "POST",
 		redirect: "error",
@@ -10158,16 +10201,18 @@ function jsonError$3(res, status, code) {
 //#region lib/types/subscription.js
 /** Translate image tool options into each subscription channel's wire inputs. */
 function subscriptionToolParameters(provider, options) {
-	if (options.size !== void 0 && options.aspectRatio !== void 0) throw new Error("size 与 aspect_ratio 请只选一个");
-	const ratio = options.aspectRatio;
+	const requestedSize = options.size?.trim() || void 0;
+	const ratio = options.aspectRatio?.trim() || void 0;
+	const imageSize = options.imageSize?.trim() || void 0;
+	if (requestedSize !== void 0 && ratio !== void 0) throw new Error("size 与 aspect_ratio 请只选一个；无关可选字段请省略，不要填写空字符串");
 	if (ratio !== void 0 && !SUBSCRIPTION_CAPABILITIES[provider].ratioOptions.some((option) => option.value === ratio)) throw new Error(`${provider} 不支持比例 ${ratio}`);
-	const size = options.size ?? (ratio === void 0 ? void 0 : provider === "chatgpt-sub" ? SUBSCRIPTION_CAPABILITIES[provider].sizeFor?.(ratio, "auto") : ratio);
+	const size = requestedSize ?? (ratio === void 0 ? void 0 : provider === "chatgpt-sub" ? SUBSCRIPTION_CAPABILITIES[provider].sizeFor?.(ratio, "auto") : ratio);
 	let quality;
-	if (options.imageSize !== void 0) {
-		if (provider === "grok-sub" && (options.imageSize === "1K" || options.imageSize === "2K")) quality = options.imageSize.toLowerCase();
-		else if (provider === "google-sub" && options.imageSize === "4K") quality = "hd";
-		else if (provider === "google-sub" && options.imageSize === "1K") quality = void 0;
-		else throw new Error(`${provider} 不支持清晰度 ${options.imageSize}`);
+	if (imageSize !== void 0) {
+		if (provider === "grok-sub" && (imageSize === "1K" || imageSize === "2K")) quality = imageSize.toLowerCase();
+		else if (provider === "google-sub" && imageSize === "4K") quality = "hd";
+		else if (provider === "google-sub" && imageSize === "1K") quality = void 0;
+		else throw new Error(`${provider} 不支持清晰度 ${imageSize}；${provider === "chatgpt-sub" ? "请省略 image_size，使用 size 或 aspect_ratio 指定尺寸" : `image_size 仅支持 ${provider === "grok-sub" ? "1K、2K" : "1K、4K"}`}`);
 	}
 	return {
 		...size === void 0 ? {} : { size },
@@ -28122,8 +28167,10 @@ function subscriptionResponse(input, startedAt, requestedCount, failedCount, ite
 }
 function studioProfile(config, provider, configured) {
 	const result = cloudStudioProfile(config, provider, configured);
+	const editModel = provider === "openai-compat" ? config.openaiCompatEditModel?.trim() : void 0;
 	return {
 		...result,
+		...editModel ? { editModel } : {},
 		...watermarkProfile(config, provider, result.model)
 	};
 }
@@ -28221,7 +28268,7 @@ function option(value) {
 	};
 }
 function assertAllowed(profile, input) {
-	if (input.model !== profile.model) throw new Error("模型配置已变化，请刷新工作台后重试");
+	if (input.model !== studioModelForMode(profile, input.mode)) throw new Error("模型配置已变化，请刷新工作台后重试");
 	if (!profile.ratioOptions.some((option) => option.value === input.ratio)) throw new Error("该 Provider 不支持所选比例");
 	const selectedQuality = input.provider === "openai" ? input.generationQuality ?? input.quality : input.resolution ?? input.quality;
 	if (!profile.qualityOptions.some((option) => option.value === selectedQuality)) throw new Error("该 Provider 不支持所选清晰度");
@@ -28791,14 +28838,53 @@ function serveSettingsHealth(req, res, health) {
 	res.end(JSON.stringify(health));
 }
 //#endregion
+//#region lib/types/image-tool-context.js
+/** Live, credential-free instructions for the conversation model's image tools. */
+function imageToolContext(config) {
+	const provider = config.provider ?? "google";
+	let model = "not configured";
+	let editModel = model;
+	try {
+		const active = resolveProvider(config);
+		model = active.provider === "comfyui" ? active.workflow?.name ?? "workflow selected in settings" : active.model;
+		const edit = resolveProvider(withProviderOverrides(config, void 0, void 0, "edit"));
+		editModel = edit.provider === "comfyui" ? edit.workflow?.name ?? model : edit.model;
+	} catch {}
+	return [
+		`Image tools current default provider: ${provider}; generation model: ${JSON.stringify(model)}; editing model: ${JSON.stringify(editModel)}.`,
+		"Unless the user explicitly requests a different image provider or model, OMIT provider and model from generate_image, generate_images and edit_image. Do not guess a channel or switch channels after an error.",
+		"Omit unused optional arguments entirely; never fill them with empty strings. size and aspect_ratio are alternatives: provide at most one.",
+		"Subscription parameters: chatgpt-sub accepts size OR aspect_ratio, but NOT image_size. grok-sub image_size: 1K or 2K. google-sub image_size: 1K or 4K. Do not copy parameters from another provider.",
+		"Minimal calls: generate_image({\"prompt\":\"a sailboat\"}); edit_image({\"prompt\":\"make the sail blue\"}). A valid ChatGPT subscription ratio example is {\"prompt\":\"a sailboat\",\"aspect_ratio\":\"1:1\"} with image_size and size omitted."
+	].join("\n");
+}
+//#endregion
 //#region lib/types/index.js
 const name = "dsh-image-gen";
 const inject = [
 	"tools",
 	"attachments",
-	"credentials",
-	"webServer"
+	"credentials"
 ];
+/** Optional strings from tool callers may contain empty placeholders (#86). */
+function cleanImageToolArgs(args) {
+	const clean = { ...args };
+	const fields = clean;
+	for (const key of [
+		"model",
+		"size",
+		"aspect_ratio",
+		"image_size",
+		"quality",
+		"workflow"
+	]) {
+		const value = fields[key];
+		if (typeof value !== "string") continue;
+		if (value.trim().length === 0) delete fields[key];
+		else fields[key] = value.trim();
+	}
+	return clean;
+}
 /** Validate the untrusted per-call provider override from tool arguments. */
 function providerOverrideOf(value) {
 	if (value === void 0 || value === null || value === "") return void 0;
@@ -28815,142 +28901,152 @@ function apply(ctx, config = {}) {
 	const knownWorkspaceRoots = /* @__PURE__ */ new Set();
 	const canvasMirror = new CanvasMirror();
 	const subscriptionManager = new SubscriptionManager(ctx);
-	registerSubscriptionRoutes(ctx, subscriptionManager);
 	const settingsHealth = installImageSettings(ctx, current(), {
 		setSource: (source) => {
 			current = () => migrateOpenAICompatConfig(plainConfig(source()));
 		},
 		onChange: () => {}
 	});
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: SETTINGS_HEALTH_ROUTE,
-		handler: (req, res) => serveSettingsHealth(req, res, settingsHealth())
-	}), "dsh-image-gen: settings health");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: IMAGE_ROUTE,
-		handler: (req, res) => serveImage(req, res, { readImage: (ref) => ctx.attachments.readImage(ref) })
-	}), "dsh-image-gen: image route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: IMPORT_ROUTE,
-		handler: (req, res) => serveImport(req, res, {
-			saveImage: (image) => ctx.attachments.saveImage(image),
-			maxImageBytes: ctx.attachments.imageLimits.maxImageBytes,
-			mediaTypes: ctx.attachments.imageLimits.mediaTypes
-		})
-	}), "dsh-image-gen: import route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: DELETE_ROUTE,
-		handler: (req, res) => serveDelete(req, res, { deleteWorkspaceImage: async (filePath) => {
-			const discovered = await getDshWorkspaceRoots().catch(() => []);
-			return deleteImageFromWorkspace(filePath, /* @__PURE__ */ new Set([
-				...knownWorkspaceRoots,
-				...discovered,
-				process.cwd()
-			]));
-		} })
-	}), "dsh-image-gen: delete route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: SAVE_WORKSPACE_ROUTE,
-		handler: (req, res) => serveSaveWorkspace(req, res, {
-			readImage: (ref) => ctx.attachments.readImage(ref),
-			saveToWorkspace: (options) => {
-				if (options.workspaceRoot) knownWorkspaceRoots.add(options.workspaceRoot);
-				return saveImageToWorkspace({
-					workspaceRoot: options.workspaceRoot,
-					folder: current().workspaceFolder,
-					attachmentId: options.attachmentId,
-					mediaType: options.mediaType,
-					data: options.data
-				});
-			},
-			getActiveWorkspaceRoot: () => Array.from(knownWorkspaceRoots)[0] || process.cwd(),
-			getAllowedWorkspaceRoots: async () => {
+	ctx.inject(["webServer"], (webCtx) => {
+		registerSubscriptionRoutes(webCtx, subscriptionManager);
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: SETTINGS_HEALTH_ROUTE,
+			handler: (req, res) => serveSettingsHealth(req, res, settingsHealth())
+		}), "dsh-image-gen: settings health");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: IMAGE_ROUTE,
+			handler: (req, res) => serveImage(req, res, { readImage: (ref) => ctx.attachments.readImage(ref) })
+		}), "dsh-image-gen: image route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: IMPORT_ROUTE,
+			handler: (req, res) => serveImport(req, res, {
+				saveImage: (image) => ctx.attachments.saveImage(image),
+				maxImageBytes: ctx.attachments.imageLimits.maxImageBytes,
+				mediaTypes: ctx.attachments.imageLimits.mediaTypes
+			})
+		}), "dsh-image-gen: import route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: DELETE_ROUTE,
+			handler: (req, res) => serveDelete(req, res, { deleteWorkspaceImage: async (filePath) => {
 				const discovered = await getDshWorkspaceRoots().catch(() => []);
-				return /* @__PURE__ */ new Set([
+				return deleteImageFromWorkspace(filePath, /* @__PURE__ */ new Set([
 					...knownWorkspaceRoots,
 					...discovered,
 					process.cwd()
-				]);
-			},
-			isSaveEnabled: () => current().saveToWorkspace !== false
-		})
-	}), "dsh-image-gen: save workspace route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: TEST_CONNECTION_ROUTE,
-		handler: (req, res) => serveTestConnection(req, res, {
-			resolveKey: (provider) => resolveApiKey(ctx, provider),
-			config: () => current(),
-			subscriptionManager
-		})
-	}), "dsh-image-gen: test connection route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: CANVAS_STATE_ROUTE,
-		handler: (req, res) => serveCanvasState(req, res, {
-			mirror: canvasMirror,
-			maxBodyBytes: Math.ceil(ctx.attachments.imageLimits.maxImageBytes * 1.4) + 262144,
-			maxImageBytes: ctx.attachments.imageLimits.maxImageBytes
-		})
-	}), "dsh-image-gen: canvas state route");
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: CANVAS_ASSET_ROUTE,
-		handler: (req, res) => serveCanvasAsset(req, res, {
-			maxImageBytes: ctx.attachments.imageLimits.maxImageBytes,
-			saveImage: (image) => ctx.attachments.saveImage(image)
-		})
-	}), "dsh-image-gen: canvas asset route");
+				]));
+			} })
+		}), "dsh-image-gen: delete route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: SAVE_WORKSPACE_ROUTE,
+			handler: (req, res) => serveSaveWorkspace(req, res, {
+				readImage: (ref) => ctx.attachments.readImage(ref),
+				saveToWorkspace: (options) => {
+					if (options.workspaceRoot) knownWorkspaceRoots.add(options.workspaceRoot);
+					return saveImageToWorkspace({
+						workspaceRoot: options.workspaceRoot,
+						folder: current().workspaceFolder,
+						attachmentId: options.attachmentId,
+						mediaType: options.mediaType,
+						data: options.data
+					});
+				},
+				getActiveWorkspaceRoot: () => Array.from(knownWorkspaceRoots)[0] || process.cwd(),
+				getAllowedWorkspaceRoots: async () => {
+					const discovered = await getDshWorkspaceRoots().catch(() => []);
+					return /* @__PURE__ */ new Set([
+						...knownWorkspaceRoots,
+						...discovered,
+						process.cwd()
+					]);
+				},
+				isSaveEnabled: () => current().saveToWorkspace !== false
+			})
+		}), "dsh-image-gen: save workspace route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: TEST_CONNECTION_ROUTE,
+			handler: (req, res) => serveTestConnection(req, res, {
+				resolveKey: (provider) => resolveApiKey(ctx, provider),
+				config: () => current(),
+				subscriptionManager
+			})
+		}), "dsh-image-gen: test connection route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: CANVAS_STATE_ROUTE,
+			handler: (req, res) => serveCanvasState(req, res, {
+				mirror: canvasMirror,
+				maxBodyBytes: Math.ceil(ctx.attachments.imageLimits.maxImageBytes * 1.4) + 262144,
+				maxImageBytes: ctx.attachments.imageLimits.maxImageBytes
+			})
+		}), "dsh-image-gen: canvas state route");
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: CANVAS_ASSET_ROUTE,
+			handler: (req, res) => serveCanvasAsset(req, res, {
+				maxImageBytes: ctx.attachments.imageLimits.maxImageBytes,
+				saveImage: (image) => ctx.attachments.saveImage(image)
+			})
+		}), "dsh-image-gen: canvas asset route");
+	});
 	ctx.inject(["systemPrompt"], (promptCtx) => {
-		promptCtx.systemPrompt.context({
+		promptCtx.effect(() => promptCtx.systemPrompt.context({
+			name: "dsh-image-gen:image-defaults",
+			order: 59,
+			text: () => imageToolContext(current())
+		}), "dsh-image-gen: image tool context");
+		promptCtx.effect(() => promptCtx.systemPrompt.context({
 			name: "dsh-image-gen:canvas",
 			order: 60,
 			text: () => canvasMirror.digest()
-		});
+		}), "dsh-image-gen: canvas context");
 	});
 	registerCanvasTools(ctx, canvasMirror, { persistSelectionImage: (image) => ctx.attachments.saveImage({
 		data: image.data,
 		mediaType: image.mediaType,
 		name: "canvas-selection"
 	}) });
-	ctx.effect(() => ctx.webServer.register({
-		kind: "exact",
-		path: STUDIO_ROUTE,
-		handler: (req, res) => serveStudio(req, res, {
-			describe: async () => {
-				const base = await describeStudio(ctx, current(), subscriptionManager);
-				const workspaces = await getDshWorkspacesFull().catch(() => []);
-				const activeRoot = Array.from(knownWorkspaceRoots)[0] || workspaces[0]?.path || process.cwd();
-				return {
-					...base,
-					workspaceRoot: activeRoot,
-					workspaces
-				};
-			},
-			generate: (input, signal) => {
-				const fallbackRoot = Array.from(knownWorkspaceRoots)[0] || process.cwd();
-				return generateFromStudio(ctx, current(), input, signal, fallbackRoot, subscriptionManager);
-			},
-			maxBodyBytes: Math.ceil(ctx.attachments.imageLimits.maxImageBytes * 1.4 * 5) + 262144
-		})
-	}), "dsh-image-gen: studio route");
-	const serveInspiration = createInspirationRoute();
-	ctx.effect(() => ctx.webServer.register({
-		kind: "prefix",
-		path: INSPIRATION_ROUTE,
-		handler: (req, res) => {
-			const originalUrl = req.url ?? "/";
-			req.url = originalUrl.startsWith("/plugins/dsh-image-gen/inspiration") ? originalUrl.slice(34) || "/" : originalUrl;
-			return serveInspiration(req, res);
-		}
-	}), "dsh-image-gen: inspiration route");
+	ctx.inject(["webServer"], (webCtx) => {
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: STUDIO_ROUTE,
+			handler: (req, res) => serveStudio(req, res, {
+				describe: async () => {
+					const base = await describeStudio(ctx, current(), subscriptionManager);
+					const workspaces = await getDshWorkspacesFull().catch(() => []);
+					const activeRoot = Array.from(knownWorkspaceRoots)[0] || workspaces[0]?.path || process.cwd();
+					return {
+						...base,
+						workspaceRoot: activeRoot,
+						workspaces
+					};
+				},
+				generate: (input, signal) => {
+					const fallbackRoot = Array.from(knownWorkspaceRoots)[0] || process.cwd();
+					return generateFromStudio(ctx, current(), input, signal, fallbackRoot, subscriptionManager);
+				},
+				maxBodyBytes: Math.ceil(ctx.attachments.imageLimits.maxImageBytes * 1.4 * 5) + 262144
+			})
+		}), "dsh-image-gen: studio route");
+		const serveInspiration = createInspirationRoute();
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "prefix",
+			path: INSPIRATION_ROUTE,
+			handler: (req, res) => {
+				const originalUrl = req.url ?? "/";
+				req.url = originalUrl.startsWith("/plugins/dsh-image-gen/inspiration") ? originalUrl.slice(34) || "/" : originalUrl;
+				return serveInspiration(req, res);
+			}
+		}), "dsh-image-gen: inspiration route");
+	});
 	/** Per-item generation request shared by generate_image and generate_images. */
-	const generateSingle = async (args, exec) => {
+	const generateSingle = async (rawArgs, exec) => {
+		const args = cleanImageToolArgs(rawArgs);
 		const active = resolveProvider(withProviderOverrides(current(), providerOverrideOf(args.provider), args.model));
 		if (active.provider === "comfyui") {
 			const workflow = selectComfyUIWorkflow(active, args.workflow);
@@ -29066,7 +29162,7 @@ function apply(ctx, config = {}) {
 					"grok-sub",
 					"google-sub"
 				],
-				description: "Optional provider for this call only (for example when the user asks to use a specific provider); omit to use the configured default. chatgpt-sub, grok-sub, and google-sub generate through the logged-in subscription account instead of an API key."
+				description: "OMIT unless the user explicitly requests a specific image provider. Omission uses the saved default; never guess a channel or fill this with an empty string. chatgpt-sub, grok-sub, and google-sub use logged-in subscription accounts instead of API keys."
 			},
 			model: {
 				type: "string",
@@ -29095,11 +29191,11 @@ function apply(ctx, config = {}) {
 					"2K",
 					"4K"
 				],
-				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K)."
+				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K). NEVER send this for chatgpt-sub; omit unused fields entirely."
 			},
 			size: {
 				type: "string",
-				description: "Optional dimensions or size tier for OpenAI, Seedream, or DashScope; WIDTHxHEIGHT maps to aspect_ratio for xAI Grok."
+				description: "Optional dimensions or size tier for OpenAI, Seedream, or DashScope; WIDTHxHEIGHT maps to aspect_ratio for xAI Grok. Do not combine with aspect_ratio; omit instead of using an empty string."
 			},
 			quality: {
 				type: "string",
@@ -29141,7 +29237,7 @@ function apply(ctx, config = {}) {
 					"grok-sub",
 					"google-sub"
 				],
-				description: "Optional provider for this call only, applied to every item; omit to use the configured default."
+				description: "OMIT unless the user explicitly requests a specific image provider. Uses the saved default for all items; never guess a channel or fill unused fields with empty strings."
 			},
 			model: {
 				type: "string",
@@ -29170,7 +29266,7 @@ function apply(ctx, config = {}) {
 					"2K",
 					"4K"
 				],
-				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K)."
+				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K). NEVER send this for chatgpt-sub; omit unused fields entirely."
 			},
 			size: {
 				type: "string",
@@ -29363,7 +29459,7 @@ function apply(ctx, config = {}) {
 					"grok-sub",
 					"google-sub"
 				],
-				description: "Optional provider for this call only (for example when the user asks to use a specific provider); omit to use the configured default. chatgpt-sub, grok-sub, and google-sub edit images through the logged-in subscription account instead of an API key."
+				description: "OMIT unless the user explicitly requests a specific image provider. Omission uses the saved default; never guess a channel or fill this with an empty string. Subscription providers edit through their logged-in accounts."
 			},
 			model: {
 				type: "string",
@@ -29415,7 +29511,7 @@ function apply(ctx, config = {}) {
 					"2K",
 					"4K"
 				],
-				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K)."
+				description: "Optional output resolution for Google Gemini, xAI Grok, Grok subscription (1K/2K), or Google subscription (1K/4K). NEVER send this for chatgpt-sub; omit unused fields entirely."
 			},
 			size: {
 				type: "string",
@@ -29431,8 +29527,9 @@ function apply(ctx, config = {}) {
 			}
 		},
 		output: imageOutput("Edited"),
-		async execute(args, exec) {
-			const active = resolveProvider(withProviderOverrides(current(), providerOverrideOf(args.provider), args.model));
+		async execute(rawArgs, exec) {
+			const args = cleanImageToolArgs(rawArgs);
+			const active = resolveProvider(withProviderOverrides(current(), providerOverrideOf(args.provider), args.model, "edit"));
 			const canvasSelection = args.source === "canvas_selection";
 			if (canvasSelection && (args.source_attachment_id !== void 0 || Array.isArray(args.source_attachment_ids) || args.source_path !== void 0 || Array.isArray(args.source_paths))) throw new Error("edit_image source=canvas_selection cannot be combined with source_attachment_id, source_attachment_ids, source_path, or source_paths; provide exactly one selector");
 			const sourceImages = canvasSelection ? await resolveCanvasSelectionReferences({

package/lib/types/client/index.d.ts

@@ -23,13 +23,14 @@ interface ImageSettings {
     openaiSizes?: Record<string, Record<string, string>>;
     openaiCompatBaseURL?: string;
     openaiCompatModel?: string;
+    openaiCompatEditModel?: string;
     openaiCompatSizes?: Record<string, Record<string, string>>;
     openaiCompatQuality?: string;
     openaiCompatQualityOptions?: string[];
     openaiCompatExtraBody?: Record<string, unknown>;
     openaiCompatEditSizeMode?: 'auto' | 'selected';
     /** Edit-request shape for the openai-compat relay (#41); see src/config.ts. */
-    openaiCompatEditFormat?: 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages';
+    openaiCompatEditFormat?: 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages' | 'siliconflowJson';
     /** Extra JSON fields merged into the JSON edit body; ignored in multipart mode. */
     openaiCompatEditExtra?: Record<string, unknown>;
     seedreamBaseURL?: string;

package/lib/types/config.d.ts

@@ -24,6 +24,8 @@ export interface Config {
     /** OpenAI-compatible relay settings; independent from the official OpenAI row. */
     openaiCompatBaseURL?: string;
     openaiCompatModel?: string;
+    /** Editing model; empty reuses the generation model. */
+    openaiCompatEditModel?: string;
     /** Default quality; empty preserves the upstream default by omitting the field. */
     openaiCompatQuality?: string;
     /** Quality values accepted by this relay; independent of resolution tiers. */
@@ -40,7 +42,7 @@ export interface Config {
       * `reference_images` field is a JSON array of base64 strings. Defaults to
       * `multipart`.
       */
-    openaiCompatEditFormat?: 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages';
+    openaiCompatEditFormat?: 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages' | 'siliconflowJson';
     /**
      * Extra JSON fields merged into the JSON edit body last (can override the
      * built-in defaults), e.g. SenseNova's `watermark`/`prompt_extend`.
@@ -173,7 +175,7 @@ export declare function resolveProvider(config: Config): {
  * Apply a per-call provider and/or model override on top of the saved config.
  * `model` is ignored for ComfyUI, whose per-call equivalent is `workflow`.
  */
-export declare function withProviderOverrides(config: Config, provider?: ImageProvider, model?: string): Config;
+export declare function withProviderOverrides(config: Config, provider?: ImageProvider, model?: string, mode?: 'generate' | 'edit'): Config;
 /**
  * One-time migration: when the legacy single OpenAI slot points at a relay
  * (non-official base URL) and the compat row is empty, move the relay config

package/lib/types/image-tool-context.d.ts

@@ -0,0 +1,3 @@
+/** Live, credential-free instructions for the conversation model's image tools. */
+import { type Config } from './config.js';
+export declare function imageToolContext(config: Config): string;

package/lib/types/openai-compatible.d.ts

@@ -32,7 +32,7 @@ export declare function generateOpenAICompatibleImage(input: {
     arkOptions?: ArkOutputOptions;
 }): Promise<GeneratedCompatibleImage>;
 /** How the edits endpoint expects its request body (#41). */
-export type CompatEditFormat = 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages' | 'xaiJson' | 'qwenJson';
+export type CompatEditFormat = 'multipart' | 'jsonImageUrlArray' | 'formReferenceImages' | 'siliconflowJson' | 'xaiJson' | 'qwenJson';
 export declare function editOpenAICompatibleImage(input: {
     apiKey: string;
     baseURL: string;

package/lib/types/shared.d.ts

@@ -200,6 +200,8 @@ export interface StudioProviderProfile {
     provider: StudioProvider;
     label: string;
     model: string;
+    /** Separate configured edit model; absent preserves the generation model. */
+    editModel?: string;
     configured: boolean;
     supportsEditing: boolean;
     ratioOptions: StudioOption[];
@@ -216,6 +218,8 @@ export interface StudioProviderProfile {
     watermarkOptions?: StudioOption[];
     watermarkCapabilities?: Record<'generate' | 'edit', StudioWatermarkCapability>;
 }
+/** A model picker/request must use the same operation-specific default. */
+export declare function studioModelForMode(profile: StudioProviderProfile, mode: 'generate' | 'edit'): string;
 export declare function studioWatermarkCapability(profile: StudioProviderProfile | undefined, mode: 'generate' | 'edit'): StudioWatermarkCapability;
 /** Lightweight workspace description exposed to the client for scoping. */
 export interface StudioWorkspaceInfo {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-image-gen",
-  "version": "0.9.2",
+  "version": "0.9.3",
   "description": "Bring ChatGPT-like image generation to DeepSeek Harness — Gemini, OpenAI, Seedream, DashScope, local ComfyUI & more.",
   "type": "module",
   "author": "shanliuling",