@henjicc/ai-sdk 0.5.2 → 0.6.1

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 0.6.1 - 2026-09-24
+
+- 修复百炼 Fun Duplex 实时识别把无句首标记的空文本中间帧误判为无效响应的问题；覆盖 Qwen Audio 3.1 streaming 与 Fun-ASR 共享协议。
+- 保持缺失文字字段、空最终结果、全程无有效 final 的严格校验及连接释放。
+
+## 0.6.0 - 2026-09-24
+
+- 新增 MiMo 2.6 Pro / Flash / Pro-UltraSpeed 能力；默认推荐 Pro / Flash，保留 2.5 ID（官方计划 10 月 21 日下线），不自动替换用户选择。
+- MiMo Chat / Responses 思考字段分离，兼容完整 reasoning_text 事件及历史思考回传，流中断不再伪装完成。
+- 新增大陆 GLM-5.3-FlashX 与官方价格；Flash / FlashX 开放 JSON 输出并保留思考。
+- 新增 KIE HappyHorse 1.1 文生、图生、多图参考模式，以及火山 Seedream 5.0 Flash 生成/编辑和图层拆分；价格按最新官方单价。
+- 百炼新增 Qwen-Audio 3.1 短音频、文件与实时 ASR，支持方言/即时热词与说话人输出；修复短音频 JSON 回退、上下文字段及文件声道数组。
+- Kimi K3 资料补齐缓存写入 TTL 价格。新增公开入口保持向后兼容，既有外部 ASR 消费方从 0.2.x 升级仍须满足 0.5.0 的宿主流式上传要求。
+
+
 ## 0.5.2 - 2026-09-21
 
 - `llm/streaming` 新增有类型定义的文本、JSON Object 与 JSON Schema 输出配置，并把它正式写入 Chat Completions 的 `response_format`。

package/README.md

@@ -1,7 +1,7 @@
 # @henjicc/ai-sdk
 
-痕迹AI 的多供应商模型 SDK：内含 8 个生成供应商、109 个图片/视频/音频模型，以及
-10 个 LLM 供应商预设和 15 个按需 ASR 模型。另有 12 个 FAL 图片工具使用独立按需入口，不进入默认 109 模型目录。预制 LLM 会按供应商与具体模型自动选择 Responses API 或 Chat Completions，宿主不需要暴露逐模型协议设置。SDK 负责目录、请求构建、媒体预处理、
+痕迹AI 的多供应商模型 SDK：内含 8 个生成供应商、111 个图片/视频/音频模型，以及
+10 个 LLM 供应商预设和 18 个按需 ASR 模型。另有 12 个 FAL 图片工具使用独立按需入口，不进入默认 111 模型目录。预制 LLM 会按供应商与具体模型自动选择 Responses API 或 Chat Completions，宿主不需要暴露逐模型协议设置。SDK 负责目录、请求构建、媒体预处理、
 供应商调用、轮询、SSE 与错误归一化；宿主只需注入网络、凭据、媒体读取和日志。
 
 ## 5 分钟快速开始
@@ -48,10 +48,10 @@ try {
 
 百炼工厂必须传 `baseUrl` 为实际地域/工作空间的 API 根地址，不包含端点路径。其他供应商可以覆盖根地址与 `credentialId`；SDK 不自动切换地域或账号。不同模型的向量不能混用；切换模型通常需要重建向量索引。Rerank 分数仅在本次请求内比较。
 
-SDK `0.5.2` 的正式分发渠道为公共 npm，无需配置 registry 或访问令牌：
+SDK `0.6.1` 的正式分发渠道为公共 npm，无需配置 registry 或访问令牌：
 
 ```bash
-npm install @henjicc/ai-sdk@0.5.2
+npm install @henjicc/ai-sdk@0.6.1
 ```
 
 然后提供 4 个宿主能力（`Transport` / `CredentialStore` / `MediaReader` / `Logger`），创建客户端：
@@ -148,11 +148,11 @@ LLM 预设默认使用 SDK 维护的官方地址，并按具体模型选择 Chat
 SDK 源码不得依赖 `@/`、`node:`、Electron、`import.meta.glob`、`eval`/`new Function`/`node:vm`；
 `npm run check:sdk` 会守住这些边界。Photoshop/受限宿主只做生成时应从
 `@henjicc/ai-sdk/generation` 导入 `createGenerationClient`；该入口不静态带入 LLM、Vercel AI SDK、
-Node 内置模块或 Fal 官方客户端，发布门禁会把它打成 IIFE 并在无网络生命周期中核对 109 个模型。
+Node 内置模块或 Fal 官方客户端，发布门禁会把它打成 IIFE 并在无网络生命周期中核对 111 个模型。
 
 ## 按需装配生成模型
 
-`@henjicc/ai-sdk/generation` 是兼容入口，默认始终装入 109 个模型。真正需要缩小 Photoshop/Tauri
+`@henjicc/ai-sdk/generation` 是兼容入口，默认始终装入 111 个模型。真正需要缩小 Photoshop/Tauri
 包体时，从不含任何内置 catalog/provider 的 `generation/core` 创建模块化客户端，并只导入完整 pack：
 
 ```ts
@@ -165,12 +165,12 @@ console.log(client.catalog.list().map((model) => model.meta.id)) // ['kie-z-imag
 
 完整单模型 pack 同时携带该模型的唯一真实 schema、provider adapter 与 provider-scoped 媒体预处理/
 上传策略；宿主不需要知道内部上传模块。`@henjicc/ai-sdk/provider-packs/kie` 可一次装入 KIE 的全部
-28 个模型；`provider-adapters/kie` 只装入 KIE 执行与上传策略、不装任何模型。所有 109 个单模型路径和
+29 个模型；`provider-adapters/kie` 只装入 KIE 执行与上传策略、不装任何模型。所有 111 个单模型路径和
 8 个供应商路径由 catalog 生成器自动产出并由 bundle 门禁穷举，新增模型不会靠手工维护 exports。
 
 每个 `models/<provider>/<model>` 子路径也导出名为 `model` 的低层定义，供目录分析或高级自定义组合；
 直接传裸 `model` 而不传同文件的 `pack` 不保证媒体上传或供应商执行完整，普通宿主应使用 `pack`。
-包根 `createAIClient` 仍默认 109 模型；若确实需要根 client 的 chat 与按需生成共存，可显式传：
+包根 `createAIClient` 仍默认 111 模型；若确实需要根 client 的 chat 与按需生成共存，可显式传：
 
 ```ts
 const client = createAIClient({
@@ -239,7 +239,7 @@ cancelLlmChatTask('chat-1')
 
 ### 可选模型分发包与统一能力筛选
 
-当前 12 个 FAL 图片工具不会混入默认 109 模型。宿主可选择单个完整工具模型 pack，也可按用户任务装入三个聚合包：
+当前 12 个 FAL 图片工具不会混入默认 111 模型。宿主可选择单个完整工具模型 pack，也可按用户任务装入三个聚合包：
 
 - `tool-packs/fal-image-edit-tools`：3 个消除工具。
 - `tool-packs/fal-image-utility-tools`：6 个重打光、暗光增强、扩图、商品摄影、照片修复和背景移除工具。
@@ -274,9 +274,9 @@ console.log(utilities.map((item) => item.id))
 - 图片实用工具：`tool-models/fal/relighting`、`tool-models/fal/control-light`、`tool-models/fal/outpaint`、`tool-models/fal/product-photography`、`tool-models/fal/photo-restoration`、`tool-models/fal/pixelcut-background-removal`。
 - 多角度：`tool-models/fal/qwen-image-edit-2511-multiple-angles`、`tool-models/fal/perspective-change`、`tool-models/fal/flux-2-multiple-angles`。
 
-每个单模型入口只导出 `model`、`provider` 与完整 `pack`。三个聚合包只携带各自 3 / 6 / 3 个工具模型、Fal adapter 和 Fal CDN 上传，不携带其余 109 模型或 LLM。能力筛选与分发是两层：`search()` 只过滤已经导入的候选，不会让已经进入 bundle 的代码自动消失；缩小包体仍必须显式选择单模型/provider/collection pack。
+每个单模型入口只导出 `model`、`provider` 与完整 `pack`。三个聚合包只携带各自 3 / 6 / 3 个工具模型、Fal adapter 和 Fal CDN 上传，不携带其余 111 模型或 LLM。能力筛选与分发是两层：`search()` 只过滤已经导入的候选，不会让已经进入 bundle 的代码自动消失；缩小包体仍必须显式选择单模型/provider/collection pack。
 
-Henji-AI 在执行层按需装入工具 pack，但普通模型选择器、`@henjicc/ai-sdk/generation` 与默认能力发现仍只展示 109 个主目录模型。
+Henji-AI 在执行层按需装入工具 pack，但普通模型选择器、`@henjicc/ai-sdk/generation` 与默认能力发现仍只展示 111 个主目录模型。
 
 ## 两类模型的公共边界
 
@@ -324,14 +324,14 @@ SDK 统一处理媒体来源优先级、首个/求和聚合、固定倍率和参
 
 ### 参数类型 → 控件
 
-以下 13 种是 `RuntimeParamDef` 的完整公开联合；括号内是当前真实 109 catalog 的出现数量。未出现不代表
+以下 13 种是 `RuntimeParamDef` 的完整公开联合；括号内是当前真实 111 catalog 的出现数量。未出现不代表
 类型无效，而是当前目录没有对应模型。
 
-| type | 109 catalog | 消费方控件 | 关键运行时字段 |
+| type | 111 catalog | 消费方控件 | 关键运行时字段 |
 |---|---:|---|---|
-| `dropdown` | 317 | 下拉选择 | `options[].value`、`default`、`required?` |
+| `dropdown` | 325 | 下拉选择 | `options[].value`、`default`、`required?` |
 | `switch` | 93 | 布尔开关 | `default` |
-| `number` | 85 | 数值输入/步进器 | `min?`、`max?`、`step?`、`default` |
+| `number` | 86 | 数值输入/步进器 | `min?`、`max?`、`step?`、`default` |
 | `text` | 7 | 单行文本 | `maxLength?`、`default` |
 | `image-upload` | 6 | 图片上传 | `maxCount?`、`accept?`、`maxSize?`、`format?` |
 | `composite` | 4 | 宿主自定义组件钩子 | `valueType?`、`default`；panel/config 不在 SDK |
@@ -344,7 +344,7 @@ SDK 统一处理媒体来源优先级、首个/求和聚合、固定倍率和参
 | `aspect-ratio` | 0 | 比例选择 | `options[].value` |
 
 所有参数共有 `id`、`type`、`order`、`default`，并可带 `required`、`valueType`、API 映射、
-`transferKey`、`visible`、`disabled`。实际 109 catalog 的字段集合与数量由
+`transferKey`、`visible`、`disabled`。实际 111 catalog 的字段集合与数量由
 `packages/ai-sdk/tests/catalog-consumer-contract.test.ts` 穷举锁定，不以旧 ParamDef 文件或示例推断。
 
 ### 条件与媒体输入
@@ -443,7 +443,7 @@ features 与原始 tags。顶层查询维度默认 AND，也可设 `mode: 'any'`
 调用方随后交给对应执行 handle；SDK 不提供一个掩盖协议差异的通用 `generate()`。
 
 能力画像是运行时选择层，不是打包器。`createModelCapabilityDiscovery({ generationPacks: [...] })` 只会看
-传入的 pack；它既不会隐式导入默认 109，也不会从 bundle 删除已导入代码。真正的按需分发仍以 import
+传入的 pack；它既不会隐式导入默认 111，也不会从 bundle 删除已导入代码。真正的按需分发仍以 import
 单模型/provider/collection pack 为边界。
 
 ### ASR/OCR 等开放能力

package/dist/capabilities/speech-recognition/bailian/module.js

@@ -77,10 +77,18 @@ async function executeFunShort(preset, input, apiKey, apiBaseUrl, context) {
     const options = providerOptions(input);
     const content = [];
     if (options.context?.trim())
-        content.push({ type: 'text', text: options.context.trim() });
+        content.push({ type: 'input_text', text: options.context.trim() });
     const parameters = { format: formatFrom(input, options) };
     if (options.sampleRateHz !== undefined)
-        parameters.sample_rate = options.sampleRateHz;
+        parameters.sample_rate = String(options.sampleRateHz);
+    if (preset.modelId === 'qwen-audio-3.1-asr-flash') {
+        if (options.keepDialect !== undefined)
+            parameters.keep_dialect = options.keepDialect;
+        if (options.diarizationEnabled !== undefined)
+            parameters.speaker_diarization_enabled = options.diarizationEnabled;
+        if (options.vocabulary !== undefined)
+            parameters.vocabulary = options.vocabulary;
+    }
     if (options.vocabularyId)
         parameters.vocabulary_id = options.vocabularyId;
     const languageHints = [...new Set([...(input.hints ?? []), ...(input.language ? [input.language] : [])]
@@ -141,8 +149,12 @@ function asyncParameters(preset, input, options) {
         ...(options.vocabularyId ? { vocabulary_id: options.vocabularyId } : {}),
         ...(options.diarizationEnabled !== undefined ? { diarization_enabled: options.diarizationEnabled } : {}),
         ...(options.speakerCount !== undefined ? { speaker_count: options.speakerCount } : {}),
-        ...(options.channelId !== undefined ? { channel_id: options.channelId } : {}),
+        ...(options.channelId !== undefined ? { channel_id: [options.channelId] } : {}),
         ...(options.specialWordFilter ? { special_word_filter: options.specialWordFilter } : {}),
+        ...(preset.modelId === 'qwen-audio-3.1-asr-flash-filetrans' ? {
+            ...(options.keepDialect !== undefined ? { keep_dialect: options.keepDialect } : {}),
+            ...(options.vocabulary !== undefined ? { vocabulary: options.vocabulary } : {}),
+        } : {}),
     };
 }
 async function wait(ms, context) {

package/dist/capabilities/speech-recognition/bailian/parse.js

@@ -26,6 +26,7 @@ function parseWord(value) {
         return undefined;
     return {
         text: `${wordText}${text(value.punctuation) ?? ''}`,
+        ...(numberValue(value.speaker_id) !== undefined ? { speakerId: numberValue(value.speaker_id) } : {}),
         startMs: numberValue(value.begin_time),
         endMs: numberValue(value.end_time),
     };
@@ -41,6 +42,7 @@ export function parseBailianSentence(value) {
     const words = records(sentence.words).map(parseWord).filter((word) => word !== undefined);
     return {
         text: sentenceText,
+        ...(numberValue(sentence.speaker_id) !== undefined ? { speakerId: numberValue(sentence.speaker_id) } : {}),
         startMs: numberValue(sentence.begin_time),
         endMs: numberValue(sentence.end_time),
         ...(words.length ? { words } : {}),
@@ -48,6 +50,9 @@ export function parseBailianSentence(value) {
 }
 /** 解析 Fun-ASR 短音频 SSE，去重供应商可能回显的累计句子。 */
 export function parseFunShortSse(payload) {
+    // 短于一分钟时供应商可能忽略 SSE 请求头并直接返回 JSON。
+    if (payload.trimStart().startsWith('{'))
+        payload = `data:${payload.replace(/\r?\n/g, '')}`;
     const segments = [];
     let latestText = '';
     let durationMs;
@@ -66,18 +71,21 @@ export function parseFunShortSse(payload) {
             throw new AiRuntimeError('invalid_response', 'Bailian Fun-ASR returned malformed SSE JSON');
         }
         const root = record(parsed);
+        if (text(root?.code))
+            throw new AiRuntimeError('provider_task_failed', text(root?.message) ?? String(root?.code));
         const output = record(root?.output);
         latestText = text(output?.text) ?? latestText;
         const durationSeconds = numberValue(record(root?.usage)?.duration);
         if (durationSeconds !== undefined)
             durationMs = durationSeconds * 1_000;
-        for (const sentence of recordsOrOne(output?.sentence)) {
+        for (const sentence of recordsOrOne(output?.sentences ?? output?.sentence)) {
             if (sentence.sentence_end !== true)
                 continue;
             const segment = parseBailianSentence(sentence);
             if (!segment)
                 continue;
             const duplicate = segments.find((candidate) => candidate.startMs === segment.startMs
+                && candidate.speakerId === segment.speakerId
                 && (candidate.text === segment.text
                     || candidate.text.startsWith(segment.text)
                     || segment.text.startsWith(candidate.text)));

package/dist/capabilities/speech-recognition/bailian/presets.d.ts

@@ -14,4 +14,6 @@ export declare const bailianQwen3AsrFlash: BailianAsrPreset;
 export declare const bailianQwen3AsrFlash20260210: BailianAsrPreset;
 export declare const bailianFunAsr: BailianAsrPreset;
 export declare const bailianQwen3AsrFlashFiletrans: BailianAsrPreset;
-export declare const bailianNonRealtimeAsrPresets: readonly [BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset];
+export declare const bailianQwenAudio31AsrFlash: BailianAsrPreset;
+export declare const bailianQwenAudio31AsrFlashFiletrans: BailianAsrPreset;
+export declare const bailianNonRealtimeAsrPresets: readonly [BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset, BailianAsrPreset];

package/dist/capabilities/speech-recognition/bailian/presets.js

@@ -46,10 +46,14 @@ export const bailianQwen3AsrFlashFiletrans = definePreset({
     protocol: 'file-async',
     asyncInputField: 'file_url',
 });
+export const bailianQwenAudio31AsrFlash = definePreset({ modelId: 'qwen-audio-3.1-asr-flash', protocol: 'fun-short-sse', maxInlineBytes: 7_500_000 });
+export const bailianQwenAudio31AsrFlashFiletrans = definePreset({ modelId: 'qwen-audio-3.1-asr-flash-filetrans', protocol: 'file-async', asyncInputField: 'file_urls' });
 export const bailianNonRealtimeAsrPresets = [
     bailianFunAsrFlash20260615,
     bailianQwen3AsrFlash,
     bailianQwen3AsrFlash20260210,
     bailianFunAsr,
     bailianQwen3AsrFlashFiletrans,
+    bailianQwenAudio31AsrFlash,
+    bailianQwenAudio31AsrFlashFiletrans,
 ];

package/dist/capabilities/speech-recognition/bailian/realtime/presets.d.ts

@@ -10,4 +10,5 @@ export declare const bailianFunAsrRealtime: BailianRealtimeAsrPreset;
 export declare const bailianFunAsrRealtime20260228: BailianRealtimeAsrPreset;
 export declare const bailianQwen3AsrFlashRealtime: BailianRealtimeAsrPreset;
 export declare const bailianQwen3AsrFlashRealtime20260210: BailianRealtimeAsrPreset;
-export declare const bailianRealtimeAsrPresets: readonly [BailianRealtimeAsrPreset, BailianRealtimeAsrPreset, BailianRealtimeAsrPreset, BailianRealtimeAsrPreset];
+export declare const bailianQwenAudio31AsrFlashStreaming: BailianRealtimeAsrPreset;
+export declare const bailianRealtimeAsrPresets: readonly [BailianRealtimeAsrPreset, BailianRealtimeAsrPreset, BailianRealtimeAsrPreset, BailianRealtimeAsrPreset, BailianRealtimeAsrPreset];

package/dist/capabilities/speech-recognition/bailian/realtime/presets.js

@@ -24,9 +24,11 @@ export const bailianFunAsrRealtime = definePreset('fun-asr-realtime', 'fun-duple
 export const bailianFunAsrRealtime20260228 = definePreset('fun-asr-realtime-2026-02-28', 'fun-duplex');
 export const bailianQwen3AsrFlashRealtime = definePreset('qwen3-asr-flash-realtime', 'qwen-realtime');
 export const bailianQwen3AsrFlashRealtime20260210 = definePreset('qwen3-asr-flash-realtime-2026-02-10', 'qwen-realtime');
+export const bailianQwenAudio31AsrFlashStreaming = definePreset('qwen-audio-3.1-asr-flash-streaming', 'fun-duplex');
 export const bailianRealtimeAsrPresets = [
     bailianFunAsrRealtime,
     bailianFunAsrRealtime20260228,
     bailianQwen3AsrFlashRealtime,
     bailianQwen3AsrFlashRealtime20260210,
+    bailianQwenAudio31AsrFlashStreaming,
 ];

package/dist/capabilities/speech-recognition/bailian/realtime/protocol.js

@@ -65,6 +65,14 @@ export function buildFunStart(preset, input, taskId) {
         parameters.heartbeat = true;
     if (options.speechNoiseThreshold !== undefined)
         parameters.speech_noise_threshold = options.speechNoiseThreshold;
+    if (preset.modelId === 'qwen-audio-3.1-asr-flash-streaming') {
+        if (options.keepDialect !== undefined)
+            parameters.keep_dialect = options.keepDialect;
+        if (options.vadModel !== undefined)
+            parameters.vad_model = options.vadModel;
+        if (options.vocabulary !== undefined)
+            parameters.vocabulary = options.vocabulary;
+    }
     return JSON.stringify({
         header: { action: 'run-task', task_id: taskId, streaming: 'duplex' },
         payload: {
@@ -142,10 +150,10 @@ function parseFunEvent(root) {
         const segment = parseBailianSentence(sentence);
         const transcript = segment?.text ?? string(sentence?.text);
         if (!transcript) {
-            // 官方会用 text="" + sentence_begin=true/sentence_end=false 宣告新句开始；
-            // 这是合法的中间生命周期事件，不是识别结果。最终句仍必须带有效文本。
-            if (sentence?.sentence_begin === true && sentence?.sentence_end === false) {
-                return { kind: 'ignored', eventType: 'empty-sentence-begin' };
+            // 句首及后续中间帧均可能尚无文字；只有首帧带 sentence_begin。
+            // 必须有 text 字符串与明确的非终态，不能把缺字段的异常响应当作空帧。
+            if (typeof sentence?.text === 'string' && sentence.sentence_end === false) {
+                return { kind: 'ignored', eventType: sentence.sentence_begin === true ? 'empty-sentence-begin' : 'empty-intermediate' };
             }
             if (sentence?.sentence_end === true) {
                 throw new AiRuntimeError('invalid_response', 'Bailian Fun-ASR final result has no text');

package/dist/capabilities/speech-recognition/bailian/realtime/types.d.ts

@@ -10,6 +10,9 @@ export interface BailianRealtimeAsrOptions {
     turnDetection?: 'server_vad' | 'manual';
     vadThreshold?: number;
     vadSilenceDurationMs?: number;
+    keepDialect?: boolean;
+    vadModel?: 'near_meeting_16k' | 'far_field_meeting_16k';
+    vocabulary?: Readonly<Record<string, number>>;
 }
 export interface BailianRealtimeModuleOptions {
     funWebSocketUrl?: string;

package/dist/capabilities/speech-recognition/bailian/types.d.ts

@@ -8,6 +8,10 @@ export interface BailianAsrOptions {
     speakerCount?: number;
     channelId?: number;
     specialWordFilter?: string;
+    /** Qwen-Audio 3.1 保留方言表达。 */
+    keepDialect?: boolean;
+    /** Qwen-Audio 3.x 即时热词：权重 1–5 或 50（超级热词最多 50 个）。 */
+    vocabulary?: Readonly<Record<string, number>>;
 }
 export interface BailianAsrModuleOptions {
     /** DashScope `/api/v1` 根路径；工作空间端点可由消费方完整注入。 */

package/dist/capabilities/speech-recognition/index.d.ts

@@ -10,12 +10,14 @@ export interface SpeechRecognitionInput {
 }
 export interface SpeechRecognitionWord {
     text: string;
+    speakerId?: number;
     startMs?: number;
     endMs?: number;
     confidence?: number;
 }
 export interface SpeechRecognitionSegment {
     text: string;
+    speakerId?: number;
     startMs?: number;
     endMs?: number;
     confidence?: number;

package/dist/catalog/index.js

@@ -77,41 +77,43 @@ import catalogModel71 from './kie/grok-imagine-video.model.js';
 import catalogModel72 from './kie/grok-imagine.model.js';
 import catalogModel73 from './kie/hailuo-02.model.js';
 import catalogModel74 from './kie/hailuo-2-3.model.js';
-import catalogModel75 from './kie/kling-3.0-omni.model.js';
-import catalogModel76 from './kie/kling-3.0-turbo.model.js';
-import catalogModel77 from './kie/kling-3.0.model.js';
-import catalogModel78 from './kie/kling-v2-6.model.js';
-import catalogModel79 from './kie/minimax-h3.model.js';
-import catalogModel80 from './kie/nano-banana-2-lite.model.js';
-import catalogModel81 from './kie/nano-banana-2.model.js';
-import catalogModel82 from './kie/nano-banana-pro.model.js';
-import catalogModel83 from './kie/qwen-image-3.0.model.js';
-import catalogModel84 from './kie/seedance-1.5-pro.model.js';
-import catalogModel85 from './kie/seedance-2.0-fast.model.js';
-import catalogModel86 from './kie/seedance-2.0-mini.model.js';
-import catalogModel87 from './kie/seedance-2.0.model.js';
-import catalogModel88 from './kie/seedance-2.5.model.js';
-import catalogModel89 from './kie/seedance-v1.model.js';
-import catalogModel90 from './kie/seedream-4.0.model.js';
-import catalogModel91 from './kie/seedream-4.5.model.js';
-import catalogModel92 from './kie/seedream-5.0-lite.model.js';
-import catalogModel93 from './kie/seedream-5.0-pro.model.js';
-import catalogModel94 from './kie/z-image.model.js';
-import catalogModel95 from './modelscope/flux-1-krea-dev.model.js';
-import catalogModel96 from './modelscope/majicmix-realistic.model.js';
-import catalogModel97 from './modelscope/modelscope-custom.model.js';
-import catalogModel98 from './modelscope/qwen-image-edit-2509.model.js';
-import catalogModel99 from './modelscope/qwen-image.model.js';
-import catalogModel100 from './modelscope/sdxl-14-ckpt.model.js';
-import catalogModel101 from './modelscope/z-image-turbo.model.js';
-import catalogModel102 from './ppio/kling-3.0.model.js';
-import catalogModel103 from './ppio/minimax-hailuo-2.3.model.js';
-import catalogModel104 from './ppio/minimax-speech.model.js';
-import catalogModel105 from './ppio/wan-2.5-preview.model.js';
-import catalogModel106 from './ppio/wan-2.6.model.js';
-import catalogModel107 from './ppio/wan-2.7.model.js';
-import catalogModel108 from './volcengine/seedream-5.0-lite.model.js';
-import catalogModel109 from './volcengine/seedream-5.0-pro.model.js';
+import catalogModel75 from './kie/happyhorse-1.1.model.js';
+import catalogModel76 from './kie/kling-3.0-omni.model.js';
+import catalogModel77 from './kie/kling-3.0-turbo.model.js';
+import catalogModel78 from './kie/kling-3.0.model.js';
+import catalogModel79 from './kie/kling-v2-6.model.js';
+import catalogModel80 from './kie/minimax-h3.model.js';
+import catalogModel81 from './kie/nano-banana-2-lite.model.js';
+import catalogModel82 from './kie/nano-banana-2.model.js';
+import catalogModel83 from './kie/nano-banana-pro.model.js';
+import catalogModel84 from './kie/qwen-image-3.0.model.js';
+import catalogModel85 from './kie/seedance-1.5-pro.model.js';
+import catalogModel86 from './kie/seedance-2.0-fast.model.js';
+import catalogModel87 from './kie/seedance-2.0-mini.model.js';
+import catalogModel88 from './kie/seedance-2.0.model.js';
+import catalogModel89 from './kie/seedance-2.5.model.js';
+import catalogModel90 from './kie/seedance-v1.model.js';
+import catalogModel91 from './kie/seedream-4.0.model.js';
+import catalogModel92 from './kie/seedream-4.5.model.js';
+import catalogModel93 from './kie/seedream-5.0-lite.model.js';
+import catalogModel94 from './kie/seedream-5.0-pro.model.js';
+import catalogModel95 from './kie/z-image.model.js';
+import catalogModel96 from './modelscope/flux-1-krea-dev.model.js';
+import catalogModel97 from './modelscope/majicmix-realistic.model.js';
+import catalogModel98 from './modelscope/modelscope-custom.model.js';
+import catalogModel99 from './modelscope/qwen-image-edit-2509.model.js';
+import catalogModel100 from './modelscope/qwen-image.model.js';
+import catalogModel101 from './modelscope/sdxl-14-ckpt.model.js';
+import catalogModel102 from './modelscope/z-image-turbo.model.js';
+import catalogModel103 from './ppio/kling-3.0.model.js';
+import catalogModel104 from './ppio/minimax-hailuo-2.3.model.js';
+import catalogModel105 from './ppio/minimax-speech.model.js';
+import catalogModel106 from './ppio/wan-2.5-preview.model.js';
+import catalogModel107 from './ppio/wan-2.6.model.js';
+import catalogModel108 from './ppio/wan-2.7.model.js';
+import catalogModel109 from './volcengine/seedream-5.0-flash.model.js';
+import catalogModel110 from './volcengine/seedream-5.0-lite.model.js';
+import catalogModel111 from './volcengine/seedream-5.0-pro.model.js';
 export * from './defineModel.js';
 export * from './model-index.js';
 export * from './validate.js';
@@ -228,5 +230,7 @@ export const catalog = [
     catalogModel107,
     catalogModel108,
     catalogModel109,
+    catalogModel110,
+    catalogModel111,
 ];
 export const catalogIndex = createModelIndex(catalog);

package/dist/catalog/kie/happyhorse-1.1.model.d.ts

@@ -0,0 +1,2 @@
+export declare const kieHappyHorse11Model: import("../../index.js").ModelRuntimeDefinition;
+export default kieHappyHorse11Model;

package/dist/catalog/kie/happyhorse-1.1.model.js

@@ -0,0 +1,67 @@
+import { defineModel } from '../defineModel.js';
+import { hasUploadedImage, resolveKieImageSources } from './mediaSources.js';
+const RATIOS = ['16:9', '9:16', '1:1', '4:3', '3:4', '4:5', '5:4', '9:21', '21:9'];
+function duration(params) {
+    const value = params.kieHappyHorse11Duration ?? 5;
+    if (typeof value !== 'number' || !Number.isInteger(value) || value < 3 || value > 15) {
+        throw new Error('HappyHorse 1.1 时长必须是 3–15 秒的整数');
+    }
+    return value;
+}
+export const kieHappyHorse11Model = defineModel({
+    meta: {
+        id: 'kie-happyhorse-1.1', canonicalModelId: 'happyhorse-1.1', seriesId: 'happyhorse', seriesRank: 1.1,
+        provider: 'kie', type: 'video',
+        tags: ['text-to-video', 'image-to-video', 'reference-mode', 'multi-mode-switch', 'supports-multi-image', 'provider-kie'],
+        polling: { interval: 5000, maxAttempts: 180, expectedAttempts: 60 },
+    },
+    inputLimits: {
+        images: { max: 1 }, videos: { max: 0 }, audios: { max: 0 },
+        rules: [{ when: 'kieHappyHorse11Mode === "reference-to-video"', images: { min: 1, max: 9 } }],
+    },
+    params: [
+        { id: 'kieHappyHorse11Mode', type: 'dropdown', order: 1, default: 'text-image-to-video',
+            options: [{ value: 'text-image-to-video' }, { value: 'reference-to-video' }] },
+        { id: 'kieHappyHorse11AspectRatio', type: 'dropdown', order: 2, default: 'smart',
+            visible: { condition: params => params.kieHappyHorse11Mode === 'reference-to-video' || !hasUploadedImage(params) },
+            options: [{ value: 'smart' }, ...RATIOS.map(value => ({ value }))] },
+        { id: 'kieHappyHorse11Resolution', type: 'dropdown', order: 3, default: '1080p',
+            options: [{ value: '720p' }, { value: '1080p' }] },
+        { id: 'kieHappyHorse11Duration', type: 'number', order: 4, default: 5, min: 3, max: 15, step: 1 },
+    ],
+    endpoints: '/api/v1/jobs/createTask',
+    request: { builder: params => {
+            const images = resolveKieImageSources(params);
+            const reference = params.kieHappyHorse11Mode === 'reference-to-video';
+            if (images.length > (reference ? 9 : 1) || (reference && !images.length)) {
+                throw new Error(reference ? '参考模式需要 1–9 张图片' : '图生视频只接受 1 张图片');
+            }
+            const mode = reference ? 'reference-to-video' : images.length ? 'image-to-video' : 'text-to-video';
+            const prompt = typeof params.prompt === 'string' ? params.prompt.trim() : '';
+            if ((!prompt && mode !== 'image-to-video') || prompt.length > (mode === 'text-to-video' ? 4999 : 5000)) {
+                throw new Error('HappyHorse 1.1 提示词为空或超出长度限制');
+            }
+            const input = { duration: duration(params), resolution: params.kieHappyHorse11Resolution === '720p' ? '720p' : '1080p' };
+            if (prompt)
+                input.prompt = prompt;
+            if (mode === 'image-to-video')
+                input.image_urls = images;
+            else {
+                const raw = String(params.kieHappyHorse11AspectRatio ?? 'smart');
+                const hint = typeof params.__firstImageRatio === 'number' && params.__firstImageRatio > 0 ? params.__firstImageRatio : 16 / 9;
+                input.aspect_ratio = RATIOS.includes(raw) ? raw : RATIOS.reduce((best, ratio) => {
+                    const distance = (value) => { const [w, h] = value.split(':').map(Number); return Math.abs(w / h - hint); };
+                    return distance(ratio) < distance(best) ? ratio : best;
+                }, '16:9');
+                if (reference)
+                    input.reference_image = images;
+            }
+            return { model: `happyhorse-1-1/${mode}`, input };
+        } },
+    pricing: {
+        currency: '$',
+        calculator: params => duration(params) * (params.kieHappyHorse11Resolution === '720p' ? 0.1125 : 0.145),
+        description: '720p $0.1125/秒；1080p $0.145/秒；三种模式同价',
+    },
+});
+export default kieHappyHorse11Model;

package/dist/catalog/volcengine/seedream-5.0-flash.model.d.ts

@@ -0,0 +1,2 @@
+export declare const volcengineSeedream50FlashModel: import("../../index.js").ModelRuntimeDefinition;
+export default volcengineSeedream50FlashModel;

package/dist/catalog/volcengine/seedream-5.0-flash.model.js

@@ -0,0 +1,7 @@
+import { createSeedream50Model } from './seedream-5.0.js';
+export const volcengineSeedream50FlashModel = createSeedream50Model({
+    id: 'volcengine-seedream-5.0-flash', canonicalModelId: 'seedream-5.0-flash',
+    apiModel: 'doubao-seedream-5-0-flash-260915', paramPrefix: 'volcengineSeedream50Flash',
+    aliases: ['seedream-5-flash-official'], flash: true,
+});
+export default volcengineSeedream50FlashModel;

package/dist/catalog/volcengine/seedream-5.0-pro.model.d.ts

@@ -1,3 +1,2 @@
-/** 火山方舟官方 Seedream 5.0 Pro 图片生成与编辑模型（运行时契约） */
 export declare const volcengineSeedream50ProModel: import("../../index.js").ModelRuntimeDefinition;
 export default volcengineSeedream50ProModel;