@dickpy/dsh-imagegen 1.5.5 → 1.5.7

package/src/prompt-enhancer.ts

@@ -1,137 +1,151 @@
-/** OpenAI-compatible chat helpers used by the optional prompt-enhancement UI. */
-
-import { isLikelyImageModelId } from './model-catalog.ts'
-
-export interface PromptModelConfig {
-  apiUrl: string
-  apiKey: string
-  model: string
-}
-
-/** Credentials shared by OpenAI-compatible `/models` discovery. */
-export interface ModelListConfig {
-  apiUrl: string
-  apiKey: string
-}
-
-function endpoint(base: string, suffix: string): string {
-  return `${base.replace(/\/+$/, '')}${suffix}`
-}
-
-function headers(apiKey: string): HeadersInit {
-  return {
-    'content-type': 'application/json',
-    ...apiKey.trim() === '' ? {} : { authorization: `Bearer ${apiKey.trim()}` },
-  }
-}
-
-async function responseJson(response: Response): Promise<Record<string, unknown>> {
-  const body: unknown = await response.json().catch(() => undefined)
-  if (!response.ok || body === undefined || body === null || typeof body !== 'object') {
-    const message = body !== null && typeof body === 'object' && typeof (body as { error?: { message?: unknown } }).error?.message === 'string'
-      ? (body as { error: { message: string } }).error.message
-      : `HTTP ${response.status}`
-    throw new Error(message)
-  }
-  return body as Record<string, unknown>
-}
-
-type ModelRecord = Record<string, unknown> & { id: string }
-
-async function listModelRecords(config: ModelListConfig): Promise<ModelRecord[]> {
-  if (config.apiUrl.trim() === '') throw new Error('API URL is required')
-  const response = await fetch(endpoint(config.apiUrl, '/models'), { headers: headers(config.apiKey) })
-  const body = await responseJson(response)
-  const data = Array.isArray(body.data) ? body.data : []
-  return data.flatMap(item => {
-    if (item === null || typeof item !== 'object' || typeof (item as { id?: unknown }).id !== 'string') return []
-    const id = (item as { id: string }).id.trim()
-    return id === '' ? [] : [{ ...(item as Record<string, unknown>), id }]
-  })
-}
-
-function textOf(value: unknown): string[] {
-  if (typeof value === 'string') return [value]
-  if (!Array.isArray(value)) return []
-  return value.filter((item): item is string => typeof item === 'string')
-}
-
-function hasImageGenerationCapability(record: ModelRecord): boolean | undefined {
-  const capability = record.capabilities
-  if (capability !== null && typeof capability === 'object') {
-    const values = capability as Record<string, unknown>
-    for (const key of ['image_generation', 'imageGeneration', 'text_to_image', 'textToImage', 'image_gen']) {
-      if (typeof values[key] === 'boolean') return values[key]
-    }
-    const serialized = JSON.stringify(values).toLowerCase()
-    if (/image[ _-]?generation|text[ _-]?to[ _-]?image/.test(serialized)) return true
-  }
-
-  const taskText = [
-    ...textOf(record.task),
-    ...textOf(record.task_type),
-    ...textOf(record.taskType),
-    ...textOf(record.type),
-    ...textOf(record.model_type),
-    ...textOf(record.modelType),
-    ...textOf(record.tasks),
-    ...textOf(record.description),
-  ].join(' ').toLowerCase()
-  if (/image[ _-]?generation|text[ _-]?to[ _-]?image|image[ _-]?gen/.test(taskText)) return true
-  if (/^image(?:[ _-]?generation)?$/.test(taskText.trim())) return true
-  if (/embedding|rerank|moderation|transcri|speech|audio|video|chat[ _-]?completion/.test(taskText)) return false
-
-  for (const key of ['output_modalities', 'outputModalities', 'supported_output_modalities']) {
-    const modalities = textOf(record[key]).map(value => value.toLowerCase())
-    if (modalities.length > 0) return modalities.includes('image')
-  }
-  return undefined
-}
-
-function isImageModelRecord(record: ModelRecord): boolean {
-  return hasImageGenerationCapability(record) ?? isLikelyImageModelId(record.id)
-}
-
-/** List candidates exposed by an OpenAI-compatible endpoint. */
-export async function listOpenAIModels(config: ModelListConfig): Promise<string[]> {
-  return [...new Set((await listModelRecords(config)).map(record => record.id))]
-    .sort((a, b) => a.localeCompare(b))
-}
-
-/** List only models that advertise or conventionally represent image generation. */
-export async function listImageModels(config: ModelListConfig): Promise<string[]> {
-  return [...new Set((await listModelRecords(config)).filter(isImageModelRecord).map(record => record.id))]
-    .sort((a, b) => a.localeCompare(b))
-}
-
-/** List chat models exposed by an OpenAI-compatible endpoint. */
-export async function listPromptModels(config: PromptModelConfig): Promise<string[]> {
-  return listOpenAIModels(config)
-}
-
-/** Expand a concise image request into a production-ready image prompt. */
-export async function enhancePrompt(config: PromptModelConfig, prompt: string): Promise<string> {
-  if (config.apiUrl.trim() === '' || config.model.trim() === '') throw new Error('prompt enhancement model is not configured')
-  const response = await fetch(endpoint(config.apiUrl, '/chat/completions'), {
-    method: 'POST',
-    headers: headers(config.apiKey),
-    body: JSON.stringify({
-      model: config.model.trim(),
-      temperature: 0.7,
-      messages: [
-        {
-          role: 'system',
-          content: 'You are an expert image-prompt editor. Expand the user request into one vivid, specific image-generation prompt. Preserve intent and language. Add only useful visual detail: subject, composition, lighting, materials, color, camera/style and quality. Return only the finished prompt, with no preface or markdown.',
-        },
-        { role: 'user', content: prompt },
-      ],
-    }),
-  })
-  const body = await responseJson(response)
-  const choices = Array.isArray(body.choices) ? body.choices : []
-  const content = choices[0] !== null && typeof choices[0] === 'object'
-    ? (choices[0] as { message?: { content?: unknown } }).message?.content
-    : undefined
-  if (typeof content !== 'string' || content.trim() === '') throw new Error('chat model returned an empty prompt')
-  return content.trim()
-}
+/** OpenAI-compatible chat helpers used by the optional prompt-enhancement UI. */
+
+import { isLikelyImageModelId } from './model-catalog.ts'
+
+export interface PromptModelConfig {
+  apiUrl: string
+  apiKey: string
+  model: string
+}
+
+/** Credentials shared by OpenAI-compatible `/models` discovery. */
+export interface ModelListConfig {
+  apiUrl: string
+  apiKey: string
+}
+
+function endpoint(base: string, suffix: string): string {
+  return `${base.replace(/\/+$/, '')}${suffix}`
+}
+
+function headers(apiKey: string): HeadersInit {
+  return {
+    'content-type': 'application/json',
+    ...apiKey.trim() === '' ? {} : { authorization: `Bearer ${apiKey.trim()}` },
+  }
+}
+
+async function responseJson(response: Response): Promise<Record<string, unknown>> {
+  const body: unknown = await response.json().catch(() => undefined)
+  if (!response.ok || body === undefined || body === null || typeof body !== 'object') {
+    const message = body !== null && typeof body === 'object' && typeof (body as { error?: { message?: unknown } }).error?.message === 'string'
+      ? (body as { error: { message: string } }).error.message
+      : `HTTP ${response.status}`
+    throw new Error(message)
+  }
+  return body as Record<string, unknown>
+}
+
+type ModelRecord = Record<string, unknown> & { id: string }
+
+async function listModelRecords(config: ModelListConfig): Promise<ModelRecord[]> {
+  if (config.apiUrl.trim() === '') throw new Error('API URL is required')
+  const response = await fetch(endpoint(config.apiUrl, '/models'), { headers: headers(config.apiKey) })
+  const body = await responseJson(response)
+  const data = Array.isArray(body.data) ? body.data : []
+  return data.flatMap(item => {
+    if (item === null || typeof item !== 'object' || typeof (item as { id?: unknown }).id !== 'string') return []
+    const id = (item as { id: string }).id.trim()
+    return id === '' ? [] : [{ ...(item as Record<string, unknown>), id }]
+  })
+}
+
+function textOf(value: unknown): string[] {
+  if (typeof value === 'string') return [value]
+  if (!Array.isArray(value)) return []
+  return value.filter((item): item is string => typeof item === 'string')
+}
+
+function hasImageGenerationCapability(record: ModelRecord): boolean | undefined {
+  const capability = record.capabilities
+  if (capability !== null && typeof capability === 'object') {
+    const values = capability as Record<string, unknown>
+    for (const key of ['image_generation', 'imageGeneration', 'text_to_image', 'textToImage', 'image_gen']) {
+      if (typeof values[key] === 'boolean') return values[key]
+    }
+    const serialized = JSON.stringify(values).toLowerCase()
+    if (/image[ _-]?generation|text[ _-]?to[ _-]?image/.test(serialized)) return true
+  }
+
+  const taskText = [
+    ...textOf(record.task),
+    ...textOf(record.task_type),
+    ...textOf(record.taskType),
+    ...textOf(record.type),
+    ...textOf(record.model_type),
+    ...textOf(record.modelType),
+    ...textOf(record.tasks),
+    ...textOf(record.description),
+  ].join(' ').toLowerCase()
+  if (/image[ _-]?generation|text[ _-]?to[ _-]?image|image[ _-]?gen/.test(taskText)) return true
+  if (/^image(?:[ _-]?generation)?$/.test(taskText.trim())) return true
+  if (/embedding|rerank|moderation|transcri|speech|audio|video|chat[ _-]?completion/.test(taskText)) return false
+
+  for (const key of ['output_modalities', 'outputModalities', 'supported_output_modalities']) {
+    const modalities = textOf(record[key]).map(value => value.toLowerCase())
+    if (modalities.length > 0) return modalities.includes('image')
+  }
+  return undefined
+}
+
+function isImageModelRecord(record: ModelRecord): boolean {
+  return hasImageGenerationCapability(record) ?? isLikelyImageModelId(record.id)
+}
+
+/** List candidates exposed by an OpenAI-compatible endpoint. */
+export async function listOpenAIModels(config: ModelListConfig): Promise<string[]> {
+  return [...new Set((await listModelRecords(config)).map(record => record.id))]
+    .sort((a, b) => a.localeCompare(b))
+}
+
+/** List only models that advertise or conventionally represent image generation. */
+export async function listImageModels(config: ModelListConfig): Promise<string[]> {
+  return [...new Set((await listModelRecords(config)).filter(isImageModelRecord).map(record => record.id))]
+    .sort((a, b) => a.localeCompare(b))
+}
+
+/** List chat models exposed by an OpenAI-compatible endpoint. */
+export async function listPromptModels(config: PromptModelConfig): Promise<string[]> {
+  return listOpenAIModels(config)
+}
+
+/** Remove reasoning-model artifacts from a chat model's visible content:
+ *  complete `<think>…</think>` blocks first, then anything from a dangling
+ *  unclosed `<think>` to the end of the text. Reasoning models served through
+ *  OpenAI-compatible endpoints (MiniMax M3, DeepSeek R1, Qwen QVQ, …) inline
+ *  these blocks in `message.content`; leaking them into the prompt box both
+ *  pollutes the prompt and can push it past image models' length limits. */
+function stripReasoning(text: string): string {
+  const withoutClosed = text.replace(/<think>[\s\S]*?<\/think>/gi, '')
+  const dangling = /<think>/i.exec(withoutClosed)
+  return (dangling === null ? withoutClosed : withoutClosed.slice(0, dangling.index)).trim()
+}
+
+/** Expand a concise image request into a production-ready image prompt. */
+export async function enhancePrompt(config: PromptModelConfig, prompt: string): Promise<string> {
+  if (config.apiUrl.trim() === '' || config.model.trim() === '') throw new Error('prompt enhancement model is not configured')
+  const response = await fetch(endpoint(config.apiUrl, '/chat/completions'), {
+    method: 'POST',
+    headers: headers(config.apiKey),
+    body: JSON.stringify({
+      model: config.model.trim(),
+      temperature: 0.7,
+      messages: [
+        {
+          role: 'system',
+          content: 'You are an expert image-prompt editor. Expand the user request into one vivid, specific image-generation prompt. Preserve intent and language. Add only useful visual detail: subject, composition, lighting, materials, color, camera/style and quality. Return only the finished prompt, with no preface or markdown.',
+        },
+        { role: 'user', content: prompt },
+      ],
+    }),
+  })
+  const body = await responseJson(response)
+  const choices = Array.isArray(body.choices) ? body.choices : []
+  const content = choices[0] !== null && typeof choices[0] === 'object'
+    ? (choices[0] as { message?: { content?: unknown } }).message?.content
+    : undefined
+  if (typeof content !== 'string' || content.trim() === '') throw new Error('chat model returned an empty prompt')
+  const enhanced = stripReasoning(content)
+  if (enhanced === '') throw new Error('chat model returned only reasoning content (empty <think> payload)')
+  return enhanced
+}