dsh-image-gen 0.6.8 → 0.6.9

package/lib/index.js

@@ -6077,6 +6077,33 @@ const DEFAULT_ZHIPU_BASE_URL = "https://open.bigmodel.cn/api/paas/v4";
 const DEFAULT_COMFYUI_BASE_URL = "http://127.0.0.1:8188";
 const DEFAULT_COMFYUI_TIMEOUT_MS = 3e5;
 const DEFAULT_COMFYUI_WORKFLOW_LABEL = "API workflow";
+/** Content types Ark's Seedream endpoint can return. */
+const ARK_OUTPUT_FORMATS = ["png", "jpeg"];
+/** Whether Ark stamps an "AI generated" watermark on the result. */
+const ARK_BACKGROUND_MODES = ["opaque", "transparent"];
+/**
+* Map the Ark output controls onto request-body fields.
+*
+* `background` is opt-in per call site because Ark restricts `transparent` to
+* image-to-image with a single alpha-bearing reference and rejects the whole
+* request otherwise:
+*
+*   text-to-image + transparent → 400 InvalidParameter
+*     "transparent background requires exactly one input image"
+*
+* So the generation endpoint must never carry it, and that caller passes
+* `{ background: false }`. `background: opaque` is never emitted anywhere: it
+* is already Ark's default, so sending it would change nothing while adding a
+* field only the edit path can act on.
+*/
+function arkOutputBody(options, { background = true } = {}) {
+	if (options === void 0) return {};
+	const body = {};
+	if (options.outputFormat !== void 0) body.output_format = options.outputFormat;
+	if (options.watermark !== void 0) body.watermark = options.watermark;
+	if (background && options.background === "transparent") body.background = "transparent";
+	return body;
+}
 /** Default model names. */
 const DEFAULT_GOOGLE_MODEL = "gemini-3.1-flash-image";
 const DEFAULT_OPENAI_MODEL = "gpt-image-2";
@@ -6170,6 +6197,9 @@ const Config = z.object({
 	openaiCompatEditExtra: z.dict(z.any()).default({}),
 	seedreamBaseURL: z.string().default(DEFAULT_SEEDREAM_BASE_URL),
 	seedreamModel: z.string().default(DEFAULT_SEEDREAM_MODEL),
+	seedreamOutputFormat: z.union(ARK_OUTPUT_FORMATS).default("jpeg"),
+	seedreamWatermark: z.boolean().default(true),
+	seedreamBackground: z.union(ARK_BACKGROUND_MODES).default("opaque"),
 	dashscopeEndpoint: z.string().default(DEFAULT_DASHSCOPE_ENDPOINT),
 	dashscopeModel: z.string().default(DEFAULT_DASHSCOPE_MODEL),
 	xaiBaseURL: z.string().default(DEFAULT_XAI_BASE_URL),
@@ -6214,13 +6244,21 @@ function resolveProvider(config) {
 				editExtra: config.openaiCompatEditExtra ?? {}
 			};
 		}
-		case "seedream": return {
-			provider: "seedream",
-			apiKeyEnv: SEEDREAM_API_KEY_ENV,
-			model: config.seedreamModel ?? "doubao-seedream-5-0-260128",
-			baseURL: config.seedreamBaseURL ?? "https://ark.cn-beijing.volces.com/api/v3",
-			imageSize: "2K"
-		};
+		case "seedream": {
+			const seedreamBackground = config.seedreamBackground ?? "opaque";
+			return {
+				provider: "seedream",
+				apiKeyEnv: SEEDREAM_API_KEY_ENV,
+				model: config.seedreamModel ?? "doubao-seedream-5-0-260128",
+				baseURL: config.seedreamBaseURL ?? "https://ark.cn-beijing.volces.com/api/v3",
+				imageSize: "2K",
+				arkOptions: {
+					outputFormat: seedreamBackground === "transparent" ? "png" : config.seedreamOutputFormat ?? "jpeg",
+					watermark: config.seedreamWatermark ?? true,
+					background: seedreamBackground
+				}
+			};
+		}
 		case "dashscope": return {
 			provider: "dashscope",
 			apiKeyEnv: DASHSCOPE_API_KEY_ENV,
@@ -8118,7 +8156,10 @@ async function generateOpenAICompatibleImage(input) {
 			model: input.model,
 			prompt: input.prompt,
 			size: input.size,
-			...input.provider === "seedream" ? { response_format: "url" } : {}
+			...input.provider === "seedream" ? {
+				response_format: "url",
+				...arkOutputBody(input.arkOptions, { background: false })
+			} : {}
 		})
 	}), input.provider, input);
 }
@@ -8306,6 +8347,7 @@ async function editSeedreamImage(input) {
 			prompt: input.prompt,
 			image: input.sourceImages.map(toDataUrl$1),
 			...input.size === void 0 || input.size.length === 0 ? {} : { size: input.size },
+			...arkOutputBody(input.arkOptions),
 			response_format: "b64_json"
 		})
 	});
@@ -20624,7 +20666,8 @@ async function generateFromStudio(ctx, config, input, signal, fallbackWorkspaceR
 				sourceImages,
 				size,
 				maxBytes: ctx.attachments.imageLimits.maxImageBytes,
-				signal
+				signal,
+				arkOptions: wired.arkOptions
 			}) : await generateOpenAICompatibleImage({
 				provider: "seedream",
 				apiKey: credential,
@@ -20633,7 +20676,8 @@ async function generateFromStudio(ctx, config, input, signal, fallbackWorkspaceR
 				prompt: input.prompt,
 				size,
 				maxBytes: ctx.attachments.imageLimits.maxImageBytes,
-				signal
+				signal,
+				arkOptions: wired.arkOptions
 			});
 			output = size;
 		} else {
@@ -21675,6 +21719,7 @@ function apply(ctx, config = {}) {
 				}), active.provider, active.model, size, current(), exec, knownWorkspaceRoots);
 			}
 			const size = args.size ?? active.imageSize;
+			const arkOptions = active.provider === "seedream" ? active.arkOptions : void 0;
 			return saveGenerated(ctx, await generateOpenAICompatibleImage({
 				provider: active.provider,
 				apiKey: credential,
@@ -21683,7 +21728,8 @@ function apply(ctx, config = {}) {
 				prompt: args.prompt,
 				size,
 				maxBytes: ctx.attachments.imageLimits.maxImageBytes,
-				signal: exec.signal
+				signal: exec.signal,
+				...arkOptions === void 0 ? {} : { arkOptions }
 			}), active.provider, active.model, size, current(), exec, knownWorkspaceRoots);
 		},
 		presentResult: (_args, result) => imagePresentation(result)
@@ -21862,7 +21908,8 @@ function apply(ctx, config = {}) {
 				sourceImages,
 				size,
 				maxBytes: ctx.attachments.imageLimits.maxImageBytes,
-				signal: exec.signal
+				signal: exec.signal,
+				arkOptions: active.arkOptions
 			}), active.provider, active.model, size, current(), exec, knownWorkspaceRoots);
 			return saveGenerated(ctx, await editDashScopeImage({
 				apiKey: credential,

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

@@ -28,6 +28,12 @@ interface ImageSettings {
     openaiCompatEditExtra?: Record<string, unknown>;
     seedreamBaseURL?: string;
     seedreamModel?: string;
+    /** Ark `output_format`; `png` is lossless and keeps an alpha channel. */
+    seedreamOutputFormat?: 'png' | 'jpeg';
+    /** Ark `watermark`; off removes the baked-in "AI generated" mark. */
+    seedreamWatermark?: boolean;
+    /** Ark `background`; `transparent` needs an alpha-bearing edit reference. */
+    seedreamBackground?: 'opaque' | 'transparent';
     dashscopeEndpoint?: string;
     dashscopeModel?: string;
     xaiBaseURL?: string;
@@ -81,7 +87,7 @@ interface ImageCardFace {
     locale?: LocaleService | undefined;
     promoted: boolean;
 }
-type SettingsCardProps = PropsRuntime<'settings.plugin.item'> & InjectFace<SettingsFace>;
+type SettingsCardProps = PropsRuntime<'settings.plugins.tab'> & InjectFace<SettingsFace>;
 type ImageCardProps = PropsRuntime<'tool.call.toolview'> & InjectFace<ImageCardFace>;
 interface ImageResultNodeProps {
     node: {

package/lib/types/config.d.ts

@@ -1,7 +1,7 @@
 /** User-facing configuration for supported image providers. */
 import z from '@deepseek-ai/schemastery';
-import { DEFAULT_DASHSCOPE_ENDPOINT, DEFAULT_DASHSCOPE_MODEL, DEFAULT_COMFYUI_BASE_URL, DEFAULT_COMFYUI_TIMEOUT_MS, DEFAULT_COMFYUI_WORKFLOW_LABEL, DEFAULT_GOOGLE_ENDPOINT, DEFAULT_GOOGLE_MODEL, DEFAULT_OPENAI_BASE_URL, DEFAULT_OPENAI_MODEL, DEFAULT_SEEDREAM_BASE_URL, DEFAULT_SEEDREAM_MODEL, DEFAULT_XAI_BASE_URL, DEFAULT_XAI_MODEL, DEFAULT_ZHIPU_BASE_URL, DEFAULT_ZHIPU_MODEL, DASHSCOPE_API_KEY_ENV, GOOGLE_API_KEY_ENV, IMAGE_PROVIDERS, OPENAI_API_KEY_ENV, OPENAI_COMPAT_API_KEY_ENV, SEEDREAM_API_KEY_ENV, XAI_API_KEY_ENV, ZHIPU_API_KEY_ENV, activeComfyUIWorkflow, resolveComfyUIWorkflows, type ComfyUIWorkflowEntry, type ImageProvider } from './shared.js';
-export { DEFAULT_DASHSCOPE_ENDPOINT, DEFAULT_DASHSCOPE_MODEL, DEFAULT_COMFYUI_BASE_URL, DEFAULT_COMFYUI_TIMEOUT_MS, DEFAULT_COMFYUI_WORKFLOW_LABEL, DEFAULT_GOOGLE_ENDPOINT, DEFAULT_GOOGLE_MODEL, DEFAULT_OPENAI_BASE_URL, DEFAULT_OPENAI_MODEL, DEFAULT_SEEDREAM_BASE_URL, DEFAULT_SEEDREAM_MODEL, DEFAULT_XAI_BASE_URL, DEFAULT_XAI_MODEL, DEFAULT_ZHIPU_BASE_URL, DEFAULT_ZHIPU_MODEL, DASHSCOPE_API_KEY_ENV, GOOGLE_API_KEY_ENV, IMAGE_PROVIDERS, OPENAI_API_KEY_ENV, OPENAI_COMPAT_API_KEY_ENV, SEEDREAM_API_KEY_ENV, XAI_API_KEY_ENV, ZHIPU_API_KEY_ENV, activeComfyUIWorkflow, resolveComfyUIWorkflows, type ComfyUIWorkflowEntry, type ImageProvider, };
+import { ARK_BACKGROUND_MODES, ARK_OUTPUT_FORMATS, DEFAULT_DASHSCOPE_ENDPOINT, DEFAULT_DASHSCOPE_MODEL, DEFAULT_COMFYUI_BASE_URL, DEFAULT_COMFYUI_TIMEOUT_MS, DEFAULT_COMFYUI_WORKFLOW_LABEL, DEFAULT_GOOGLE_ENDPOINT, DEFAULT_GOOGLE_MODEL, DEFAULT_OPENAI_BASE_URL, DEFAULT_OPENAI_MODEL, DEFAULT_SEEDREAM_BASE_URL, DEFAULT_SEEDREAM_MODEL, DEFAULT_XAI_BASE_URL, DEFAULT_XAI_MODEL, DEFAULT_ZHIPU_BASE_URL, DEFAULT_ZHIPU_MODEL, DASHSCOPE_API_KEY_ENV, GOOGLE_API_KEY_ENV, IMAGE_PROVIDERS, OPENAI_API_KEY_ENV, OPENAI_COMPAT_API_KEY_ENV, SEEDREAM_API_KEY_ENV, XAI_API_KEY_ENV, ZHIPU_API_KEY_ENV, activeComfyUIWorkflow, resolveComfyUIWorkflows, type ArkBackgroundMode, type ArkOutputFormat, type ArkOutputOptions, type ComfyUIWorkflowEntry, type ImageProvider } from './shared.js';
+export { ARK_BACKGROUND_MODES, ARK_OUTPUT_FORMATS, DEFAULT_DASHSCOPE_ENDPOINT, DEFAULT_DASHSCOPE_MODEL, DEFAULT_COMFYUI_BASE_URL, DEFAULT_COMFYUI_TIMEOUT_MS, DEFAULT_COMFYUI_WORKFLOW_LABEL, DEFAULT_GOOGLE_ENDPOINT, DEFAULT_GOOGLE_MODEL, DEFAULT_OPENAI_BASE_URL, DEFAULT_OPENAI_MODEL, DEFAULT_SEEDREAM_BASE_URL, DEFAULT_SEEDREAM_MODEL, DEFAULT_XAI_BASE_URL, DEFAULT_XAI_MODEL, DEFAULT_ZHIPU_BASE_URL, DEFAULT_ZHIPU_MODEL, DASHSCOPE_API_KEY_ENV, GOOGLE_API_KEY_ENV, IMAGE_PROVIDERS, OPENAI_API_KEY_ENV, OPENAI_COMPAT_API_KEY_ENV, SEEDREAM_API_KEY_ENV, XAI_API_KEY_ENV, ZHIPU_API_KEY_ENV, activeComfyUIWorkflow, resolveComfyUIWorkflows, type ArkBackgroundMode, type ArkOutputFormat, type ArkOutputOptions, type ComfyUIWorkflowEntry, type ImageProvider, };
 /** Default workspace subfolder that receives generated image files. */
 export declare const DEFAULT_WORKSPACE_FOLDER = "dsh-image-gen";
 /** Google tool-level controls. */
@@ -34,6 +34,22 @@ export interface Config {
     openaiCompatEditExtra?: Record<string, unknown>;
     seedreamBaseURL?: string;
     seedreamModel?: string;
+    /**
+     * Ark `output_format`. Ark defaults to `jpeg`, which is lossy and cannot
+     * carry an alpha channel; `png` is lossless, so it survives later editing
+     * (e.g. background removal) without JPEG ringing around the subject.
+     */
+    seedreamOutputFormat?: ArkOutputFormat;
+    /**
+     * Ark `watermark`. Ark defaults to `true`, which bakes an "AI generated"
+     * mark into the bottom-right corner.
+     */
+    seedreamWatermark?: boolean;
+    /**
+     * Ark `background`. Only the edit path can honour `transparent`, and only
+     * when every reference image already carries an alpha channel.
+     */
+    seedreamBackground?: ArkBackgroundMode;
     dashscopeEndpoint?: string;
     dashscopeModel?: string;
     xaiBaseURL?: string;
@@ -85,6 +101,7 @@ export declare function resolveProvider(config: Config): {
     model: string;
     baseURL: string;
     imageSize: string;
+    arkOptions: ArkOutputOptions;
 } | {
     provider: 'dashscope';
     apiKeyEnv: string;

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

@@ -1,5 +1,6 @@
 /** OpenAI Images API and compatible response adapter. */
 import type { ImageMediaType } from '@deepseek-ai/dsh-attachment';
+import { type ArkOutputOptions } from './shared.js';
 export interface GeneratedCompatibleImage {
     data: Uint8Array;
     mediaType: ImageMediaType;
@@ -17,6 +18,8 @@ export declare function generateOpenAICompatibleImage(input: {
     size: string;
     maxBytes: number;
     signal: AbortSignal;
+    /** Ark-only output controls; ignored by every other provider. */
+    arkOptions?: ArkOutputOptions;
 }): Promise<GeneratedCompatibleImage>;
 /** How the edits endpoint expects its request body (#41). */
 export type CompatEditFormat = 'multipart' | 'jsonImageUrlArray';

package/lib/types/seedream.d.ts

@@ -1,6 +1,7 @@
 /** Volcengine Ark Seedream image-editing adapter. */
 import type { ImageMediaType } from '@deepseek-ai/dsh-attachment';
 import type { GeneratedCompatibleImage } from './openai-compatible.js';
+import { type ArkOutputOptions } from './shared.js';
 /** Edit one image through Ark ImageGenerations using a data-URL reference. */
 export declare function editSeedreamImage(input: {
     apiKey: string;
@@ -14,4 +15,10 @@ export declare function editSeedreamImage(input: {
     size?: string;
     maxBytes: number;
     signal: AbortSignal;
+    /**
+     * Ark output controls. The edit path is the only one Ark lets request a
+     * transparent background, and it requires every reference image to carry an
+     * alpha channel — see `arkOutputBody`.
+     */
+    arkOptions?: ArkOutputOptions;
 }): Promise<GeneratedCompatibleImage>;

package/lib/types/shared.d.ts

@@ -247,6 +247,45 @@ export declare const DEFAULT_COMFYUI_BASE_URL = "http://127.0.0.1:8188";
 export declare const DEFAULT_COMFYUI_TIMEOUT_MS = 300000;
 export declare const DEFAULT_COMFYUI_WORKFLOW_LABEL = "API workflow";
 export declare const MAX_COMFYUI_WORKFLOW_BYTES: number;
+/** Content types Ark's Seedream endpoint can return. */
+export declare const ARK_OUTPUT_FORMATS: readonly ["png", "jpeg"];
+/** Whether Ark stamps an "AI generated" watermark on the result. */
+export declare const ARK_BACKGROUND_MODES: readonly ["opaque", "transparent"];
+export type ArkOutputFormat = typeof ARK_OUTPUT_FORMATS[number];
+export type ArkBackgroundMode = typeof ARK_BACKGROUND_MODES[number];
+/**
+ * Ark (Seedream) output controls, shared by the generate and edit paths.
+ *
+ * Every field mirrors an Ark request-body field of the same meaning, and every
+ * default below reproduces Ark's own default — so an untouched configuration
+ * sends exactly what it sent before these options existed.
+ */
+export interface ArkOutputOptions {
+    /** `output_format`. Ark defaults to `jpeg`; `png` is lossless and keeps an alpha channel. */
+    outputFormat?: ArkOutputFormat;
+    /** `watermark`. Ark defaults to `true`, which bakes an "AI generated" mark into the image. */
+    watermark?: boolean;
+    /** `background`. Ark defaults to `opaque`. */
+    background?: ArkBackgroundMode;
+}
+/**
+ * Map the Ark output controls onto request-body fields.
+ *
+ * `background` is opt-in per call site because Ark restricts `transparent` to
+ * image-to-image with a single alpha-bearing reference and rejects the whole
+ * request otherwise:
+ *
+ *   text-to-image + transparent → 400 InvalidParameter
+ *     "transparent background requires exactly one input image"
+ *
+ * So the generation endpoint must never carry it, and that caller passes
+ * `{ background: false }`. `background: opaque` is never emitted anywhere: it
+ * is already Ark's default, so sending it would change nothing while adding a
+ * field only the edit path can act on.
+ */
+export declare function arkOutputBody(options: ArkOutputOptions | undefined, { background }?: {
+    background?: boolean;
+}): Record<string, unknown>;
 /** Default model names. */
 export declare const DEFAULT_GOOGLE_MODEL = "gemini-3.1-flash-image";
 export declare const DEFAULT_OPENAI_MODEL = "gpt-image-2";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-image-gen",
-  "version": "0.6.8",
+  "version": "0.6.9",
   "description": "Bring ChatGPT-like image generation to DeepSeek Harness — Gemini, OpenAI, Seedream, DashScope, local ComfyUI & more.",
   "type": "module",
   "author": "shanliuling",
@@ -66,7 +66,7 @@
   "engines": {
     "node": "^22.19.0 || >=24.0.0"
   },
-  "license": "MIT",
+  "license": "Apache-2.0",
   "keywords": [
     "deepseek",
     "deepseek-harness",

package/skills/install-dsh-image-gen/SKILL.md

@@ -16,7 +16,7 @@ description: Install, configure, verify, or remove the dsh-image-gen Bundle in a
 4. If pnpm blocks the Git dependency's `prepare` script, explain that the allowance executes repository code during installation. Add only the exact package key pnpm reports to the profile's `pnpm-workspace.yaml`, then retry after the user approves.
 5. Verify `dsh --profile <profile> --dump-config` contains the `dsh-image-gen` layer and `image-gen` row.
 6. Start the profile. If port 3080 is already in use, identify the existing DSH process before stopping anything.
-7. In Settings > Plugins > Plugin configuration > Image generation, select a cloud Provider and save its API key, or select Local ComfyUI, enter its URL, and import a ComfyUI API Format workflow JSON containing `{{prompt}}`. Never print, read back, or commit a key.
+7. In Settings > Plugins > Image generation (DSH 0.1.5 and earlier: Settings > Plugins > Plugin Configuration > Image generation), select a cloud Provider and save its API key, or select Local ComfyUI, enter its URL, and import a ComfyUI API Format workflow JSON containing `{{prompt}}`. Never print, read back, or commit a key.
 8. Test with “生成一张戴墨镜的赛博朋克猫图片”. Confirm one `generate_image` call succeeds and its image renders in the conversation.
 
 For removal, run `dsh plugin --profile <profile> remove dsh-image-gen`. Do not delete the user's credentials unless they explicitly request it.