diff --git a/es/tutorials/partner-nodes/minimax/minimax-h3.mdx b/es/tutorials/partner-nodes/minimax/minimax-h3.mdx index 253870ba17..4e59b0ae6b 100644 --- a/es/tutorials/partner-nodes/minimax/minimax-h3.mdx +++ b/es/tutorials/partner-nodes/minimax/minimax-h3.mdx @@ -2,13 +2,14 @@ title: "Generación de video con la API de MiniMax H3 en ComfyUI" description: "Usa los nodos API de MiniMax H3 en ComfyUI para flujos en la nube de texto a video, primer/último fotograma y referencia a video con audio estéreo nativo." sidebarTitle: "Minimax H3" -translationSourceHash: b6d84062 +translationSourceHash: e44fb20e translationFrom: tutorials/partner-nodes/minimax/minimax-h3.mdx translationBlockHashes: "_intro": a261dbf5 "What MiniMax H3 is good at": 4c84d1e7 "Example outputs": 77256fd5 - "MiniMax H3 API workflows": 9f3a305f + "MiniMax H3 API workflows": 31840aba + "Editing a video with the H3 Max nodes": 21b27918 --- import ReqHint from "/snippets/es/tutorials/partner-nodes/req-hint.mdx"; import UpdateReminder from "/snippets/es/tutorials/update-reminder.mdx"; @@ -224,7 +225,7 @@ Genera videos condicionados por imágenes, videos y audio de referencia a travé #### Consejos para el prompt -1. **Referencia por etiqueta**: haz referencia a cada entrada por su orden en el prompt, por ejemplo `Image 1`, `Video 1` o `Audio 1` +1. **Referencia por etiqueta**: haz referencia a cada entrada por su orden en el prompt, por ejemplo `Image 1`, `Video 1` o `Audio 1`. También se acepta la forma abreviada `@Image1`, `@Video1` y `@Audio1`, que se resuelve a las mismas etiquetas 2. **Límites**: hasta 9 imágenes de referencia, 3 videos de referencia y 3 clips de audio de referencia. Cada video o clip de audio debe durar entre 2 y 15 segundos; el total de video y audio de referencia está limitado a 15 segundos cada uno 3. **El audio necesita un ancla visual**: el audio de referencia no se puede usar sin al menos una imagen o un video de referencia 4. **Requisitos de video**: los videos de referencia deben tener entre 23.976 y 60 FPS @@ -278,3 +279,43 @@ Genera videos condicionados por medios de referencia con el modelo H3 Max, más Imagen de referencia para el flujo de trabajo, o usa tus propias imágenes. +## Editar un video con los nodos H3 Max + +H3 Max también incluye seis nodos de partner que parten de un video o de una imagen fija en lugar de un prompt, y editan ese material: Recast, Extend Video, Reframe Video, Insert Video, Lip Sync y Camera Control. No hay plantillas publicadas para ellos. Agrégalos desde la biblioteca de nodos, en el grupo **Partner Nodes > video > MiniMax**, y actualiza ComfyUI a la última versión si no aparecen. + +| Nodo | Qué hace | Medio de partida | +| :--- | :----------- | :------------- | +| **MiniMax H3 Max Recast** | Reemplaza a las personas de un video con personas de fotos de referencia, conservando el movimiento, la cámara, los cortes y el sonido de la fuente | Video de 5 a 30 segundos, 1 a 4 fotos | +| **MiniMax H3 Max Extend Video** | Continúa un video con material nuevo descrito en el prompt | Video de 1.625 a 60 segundos, 1 a 15 segundos de material nuevo | +| **MiniMax H3 Max Reframe Video** | Ajusta un video a una nueva relación de aspecto y genera el área que lo rodea, conservando el tiempo y el sonido | Video de 0.5 a 300 segundos | +| **MiniMax H3 Max Insert Video** | Reemplaza la sección entre dos marcas de tiempo con una escena recién generada | Video de hasta 60 segundos, más un tiempo de inicio y de reanudación | +| **MiniMax H3 Max Lip Sync** | Anima una imagen fija para que la boca siga una pista de audio | Una imagen, 5 a 899 segundos de audio | +| **MiniMax H3 Max Camera Control** | Orbita la cámara alrededor del sujeto de una imagen fija | Una imagen, más un ángulo y una distancia finales | + +El precio depende del nodo, el modelo, la resolución, la duración y las entradas de referencia. La insignia de precio muestra antes de ejecutar el nodo una tarifa por segundo, un rango aproximado o un total calculado. Por ejemplo, Extend Video admite H3 Max y H3 Max Turbo y calcula el total a partir de la duración, la resolución y el modelo seleccionados. Las tarifas por resolución están en la [tabla de precios de Minimax](/es/tutorials/partner-nodes/pricing). + +En los prompts de Recast, Extend Video e Insert Video, haz referencia a las entradas conectadas como `@Image1`, `@Audio1` o `@Video1`, numeradas en el orden de conexión. El nodo resuelve cada etiqueta a la referencia correspondiente antes de enviar la solicitud. Una etiqueta que apunta más allá de las referencias de un tipo admitido detiene la ejecución con un error, mientras que una etiqueta de un tipo que el nodo no admite, como el audio en Recast, se envía tal cual. + +### Recast + +Reemplaza a las personas de un video de origen con las personas de 1 a 4 fotos de referencia, conservando la cámara, los cortes y el sonido del original. El video de origen debe durar de 5 a 30 segundos, con cada toma de 15 segundos o menos y una relación de aspecto entre 0.4 y 2.5. Las fotos deben ser de al menos 256x256 píxeles. De forma predeterminada reemplazan a las personas principales de izquierda a derecha; usa el prompt para asignarlas una por una, por ejemplo `the man becomes @Image2`. Se renderiza a 768P de forma predeterminada o a 1080P, y **Enhance realism** añade una textura de piel más natural y funde a las nuevas personas en la escena. + +### Extend Video + +Agrega de 1 a 15 segundos de material nuevo al final de un video. El prompt describe lo que ocurre después, no el material de origen. Acepta los modelos Max y Turbo y un video de origen de 1.625 a 60 segundos, de hasta 50 MB, con una relación de aspecto entre 0.4 y 2.5. Hasta 3 audios de referencia opcionales pueden guiar el material nuevo, de al menos 2 segundos cada uno y 14.5 segundos en total. **Output** devuelve el video completo renderizado de nuevo a la resolución seleccionada (`extended`) o solo el material nuevo (`continuation`, que comienza con una repetición de 1.6 segundos del final del video de origen). La resolución va de 480P a 2K; a 2K, `extended` está limitado a 45 segundos de origen más material nuevo. + +### Reframe Video + +Reajusta un video a una nueva relación de aspecto, ya sea 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 o 9:21, generando el área alrededor del encuadre reajustado. Se conservan el tiempo y el sonido del video de origen. El video de origen puede durar de 0.5 a 300 segundos, y un prompt opcional guía el contenido de los alrededores generados. Se renderiza a 480P, 768P o 1080P. + +### Insert Video + +Reemplaza la sección de un video entre `start_time` y `resume_time` con una escena recién generada de 5 a 13 segundos. El video de origen puede durar hasta 60 segundos, y `resume_time` debe caer al menos 1.375 segundos antes de su final. Describe la nueva escena en el prompt, con referencias, o ambas cosas: hasta 9 imágenes de referencia y 3 videos de referencia de 2 a 15 segundos cada uno, 15 segundos en total. **Color match** está activado de forma predeterminada y funde la nueva escena en el video de origen. Se renderiza a 480P o 768P. + +### Lip Sync + +Anima una imagen fija para que los movimientos de su boca sigan una pista de audio, y el video dura lo mismo que el audio. El audio puede durar de 5 a 899 segundos. La imagen debe ser de al menos 256x256 píxeles con una relación de aspecto entre 0.4 y 2.5. Se renderiza a 480P, 768P, 1080P o 2K. **Transcription** está desactivado de forma predeterminada; actívalo para transcribir el audio primero y usar el texto para guiar la sincronización. + +### Camera Control + +Genera un video a partir de una imagen fija con la cámara moviéndose a un nuevo ángulo y distancia alrededor del sujeto. Define la cámara final con `azimuth` (órbita horizontal, los valores positivos se mueven a la derecha), `elevation` (los valores positivos miran hacia abajo, los negativos hacia arriba) y `distance` (1 mantiene la distancia, por debajo de 1 se acerca, por encima de 1 se aleja). Una entrada de información de cámara de los nodos **Create Camera Info** o **Compose Camera Angle Prompt** establece el ángulo en una vista 3D en su lugar. Un prompt vacío mantiene la escena congelada para que solo se mueva la cámara. La duración es de 1 a 15 segundos a 480P, 768P o 1080P. El azimut es un objetivo más que un resultado exacto: una solicitud de 90 grados a menudo termina entre 30 y 90 grados. diff --git a/ja/tutorials/partner-nodes/minimax/minimax-h3.mdx b/ja/tutorials/partner-nodes/minimax/minimax-h3.mdx index be25de34c8..3a14fe6b6c 100644 --- a/ja/tutorials/partner-nodes/minimax/minimax-h3.mdx +++ b/ja/tutorials/partner-nodes/minimax/minimax-h3.mdx @@ -2,13 +2,14 @@ title: "ComfyUIでのMiniMax H3 APIビデオ生成" description: "ComfyUI で MiniMax H3 API ノードを使い、クラウドワークフローでテキストから動画、先頭・末尾フレームから動画、参照から動画を、ネイティブのステレオオーディオ付きで生成します。" sidebarTitle: "Minimax H3" -translationSourceHash: b6d84062 +translationSourceHash: e44fb20e translationFrom: tutorials/partner-nodes/minimax/minimax-h3.mdx translationBlockHashes: "_intro": a261dbf5 "What MiniMax H3 is good at": 4c84d1e7 "Example outputs": 77256fd5 - "MiniMax H3 API workflows": 9f3a305f + "MiniMax H3 API workflows": 31840aba + "Editing a video with the H3 Max nodes": 21b27918 --- import ReqHint from "/snippets/ja/tutorials/partner-nodes/req-hint.mdx"; @@ -225,7 +226,7 @@ MiniMax H3 APIを通じて、参照画像、参照ビデオ、参照オーディ #### プロンプト作成のヒント -1. **タグによる参照**:プロンプト内の順序で各入力を参照します。たとえば、`Image 1`、`Video 1`、`Audio 1` のように指定します。 +1. **タグによる参照**:プロンプト内の順序で各入力を参照します。たとえば、`Image 1`、`Video 1`、`Audio 1` のように指定します。`@Image1`、`@Video1`、`@Audio1` の省略記法も使用でき、同じタグとして解決されます。 2. **制限**:参照画像は最大9枚、参照ビデオは最大3本、参照オーディオクリップは最大3つまで指定できます。各ビデオまたはオーディオクリップは2〜15秒である必要があります。また、参照ビデオと参照オーディオは、それぞれ合計15秒までに制限されます。 3. **オーディオには視覚的なアンカーが必要**:参照オーディオは、少なくとも1つの参照画像または参照ビデオがないと使用できません。 4. **ビデオの要件**:参照ビデオは23.976〜60 FPSである必要があります。 @@ -279,3 +280,43 @@ MiniMaxは、参照駆動型生成(R2V)向けの公式の[フル参照モー ワークフロー用の参照画像です。または、ご自身の画像を使用できます。 +## H3 Maxノードでビデオを編集する + +H3 Maxには、プロンプトだけでなくビデオや静止画を起点として、その素材を編集する6つのパートナーノードも用意されています。Recast、Extend Video、Reframe Video、Insert Video、Lip Sync、Camera Controlです。これらのノードにテンプレートは付属していません。**Partner Nodes > video > MiniMax** のノードライブラリから追加してください。見つからない場合はComfyUIを最新バージョンにアップデートしてください。 + +| ノード | 機能 | 開始素材 | +| :--- | :----------- | :------------- | +| **MiniMax H3 Max Recast** | ビデオ内の人物を参照写真の人物に置き換え、ソースの動き、カメラ、カット、サウンドを維持します | 5〜30秒のビデオ、1〜4枚の写真 | +| **MiniMax H3 Max Extend Video** | プロンプトで記述した新しい映像でビデオを続けます | 1.625〜60秒のビデオ、1〜15秒の新しい映像 | +| **MiniMax H3 Max Reframe Video** | ビデオを新しいアスペクト比に収め、その周囲の領域を生成します。タイミングとサウンドは維持されます | 0.5〜300秒のビデオ | +| **MiniMax H3 Max Insert Video** | 2つのタイムスタンプの間の区間を、新しく生成したシーンに置き換えます | 最大60秒のビデオと、開始時刻および再開時刻 | +| **MiniMax H3 Max Lip Sync** | 静止画をアニメーションさせ、口の動きをオーディオトラックに追従させます | 画像1枚、5〜899秒のオーディオ | +| **MiniMax H3 Max Camera Control** | 静止画の被写体の周囲をカメラが周回します | 画像1枚と、終了角度および距離 | + +料金はノード、モデル、解像度、再生時間、参照入力によって決まります。価格バッジには、実行前に秒あたりの料金、おおよその範囲、または計算された合計が表示されます。たとえば Extend Video は H3 Max と H3 Max Turbo の両方に対応し、選択した再生時間、解像度、モデルから合計を計算します。解像度ごとの料金は[Minimaxの料金表](/ja/tutorials/partner-nodes/pricing)に記載されています。 + +Recast、Extend Video、Insert Videoのプロンプトでは、接続した参照を `@Image1`、`@Audio1`、`@Video1` として参照します。番号は接続順です。ノードはリクエストを送信する前に各タグを対応する参照に解決します。対応する種類の参照数を超えるタグがあると実行がエラーで停止し、ノードが受け付けない種類のタグ(例: Recast のオーディオ)はそのまま送信されます。 + +### Recast + +ソースビデオの人物を1〜4枚の参照写真の人物に置き換えます。カメラ、カット、オリジナルのサウンドは維持されます。ソースビデオは5〜30秒で、各ショットは15秒以下、アスペクト比は0.4〜2.5である必要があります。写真は最低256x256ピクセル必要です。デフォルトでは左から右へ主要な人物を置き換えます。プロンプトで1人ずつ割り当てることもでき、例えば `the man becomes @Image2` のように指定します。デフォルトでは768P、または1080Pでレンダリングされ、**Enhance realism** を有効にすると肌のテクスチャがより自然になり、新しい人物がシーンに馴染みます。 + +### Extend Video + +ビデオの末尾に1〜15秒の新しい映像を追加します。プロンプトにはソース映像ではなく、次に何が起きるかを記述します。MaxおよびTurboモデルに対応し、ソースビデオは1.625〜60秒、最大50 MB、アスペクト比は0.4〜2.5です。最大3つのオプションの参照オーディオで新しい映像を誘導できます。各2秒以上、合計14.5秒です。**Output** では、選択した解像度で再レンダリングされたビデオ全体(`extended`)か、新しい映像のみ(`continuation`。冒頭にソースの末尾が1.6秒繰り返されます)を返します。解像度は480P〜2Kです。2Kの場合、`extended` はソースと新規映像を合わせて45秒までに制限されます。 + +### Reframe Video + +ビデオを新しいアスペクト比(16:9、9:16、1:1、4:3、3:4、21:9、9:21のいずれか)に収め直し、収め直したフレームの周囲の領域を生成します。ソースのタイミングとサウンドは維持されます。ソースビデオは0.5〜300秒で、オプションのプロンプトで生成される周囲の内容を誘導できます。480P、768P、または1080Pでレンダリングします。 + +### Insert Video + +ビデオの `start_time` から `resume_time` までの区間を、新しく生成した5〜13秒のシーンに置き換えます。ソースビデオは最大60秒で、`resume_time` は末尾から少なくとも1.375秒前である必要があります。新しいシーンはプロンプトで、参照で、またはその両方で記述します。最大9枚の参照画像と3本の参照ビデオ(各2〜15秒、合計15秒)を使用できます。**Color match** はデフォルトで有効で、新しいシーンをソースに馴染ませます。480Pまたは768Pでレンダリングします。 + +### Lip Sync + +静止画をアニメーションさせ、口の動きがオーディオトラックに追従するようにします。ビデオの長さはオーディオと同じになります。オーディオは5〜899秒です。画像は最低256x256ピクセル、アスペクト比は0.4〜2.5である必要があります。480P、768P、1080P、または2Kでレンダリングします。**Transcription** はデフォルトでオフです。オンにすると先にオーディオを文字起こしし、そのテキストを使って同期を誘導します。 + +### Camera Control + +静止画から、カメラが被写体の周囲の新しい角度と距離へ移動するビデオを生成します。終了時のカメラは `azimuth`(水平方向の周回、正の値で右へ移動)、`elevation`(正の値で見下ろし、負の値で見上げる)、`distance`(1で距離を維持、1未満で近づき、1より大きいと引く)で設定します。**Create Camera Info** または **Compose Camera Angle Prompt** ノードからのカメラ情報入力を使うと、代わりに3Dビュー上で角度を設定できます。プロンプトを空にするとシーンが固定され、カメラだけが動きます。再生時間は1〜15秒で、480P、768P、1080Pに対応します。azimuthは正確な結果ではなく目標値です。90度を指定しても、実際には30〜90度の間に収まることがよくあります。 diff --git a/ko/tutorials/partner-nodes/minimax/minimax-h3.mdx b/ko/tutorials/partner-nodes/minimax/minimax-h3.mdx index d16b975973..345c05f80b 100644 --- a/ko/tutorials/partner-nodes/minimax/minimax-h3.mdx +++ b/ko/tutorials/partner-nodes/minimax/minimax-h3.mdx @@ -2,13 +2,14 @@ title: "ComfyUI에서 MiniMax H3 API 비디오 생성" description: "ComfyUI에서 MiniMax H3 API 노드로 클라우드 워크플로를 실행해 텍스트, 첫/마지막 프레임, 레퍼런스 기반 비디오를 네이티브 스테레오 오디오와 함께 생성하세요." sidebarTitle: "MiniMax H3" -translationSourceHash: b6d84062 +translationSourceHash: e44fb20e translationFrom: tutorials/partner-nodes/minimax/minimax-h3.mdx translationBlockHashes: "_intro": a261dbf5 "What MiniMax H3 is good at": 4c84d1e7 "Example outputs": 77256fd5 - "MiniMax H3 API workflows": 9f3a305f + "MiniMax H3 API workflows": 31840aba + "Editing a video with the H3 Max nodes": 21b27918 --- import ReqHint from "/snippets/ko/tutorials/partner-nodes/req-hint.mdx"; import UpdateReminder from "/snippets/ko/tutorials/update-reminder.mdx"; @@ -224,7 +225,7 @@ MiniMax H3 API를 통해 레퍼런스 이미지, 비디오, 오디오를 조건 #### 프롬프트 팁 -1. **태그로 레퍼런스 지정**: 프롬프트에서 각 입력을 순서대로 참조하세요. 예: `Image 1`, `Video 1`, `Audio 1` +1. **태그로 레퍼런스 지정**: 프롬프트에서 각 입력을 순서대로 참조하세요. 예: `Image 1`, `Video 1`, `Audio 1`. `@Image1`, `@Video1`, `@Audio1` 단축 표기도 허용되며 동일한 태그로 처리됩니다 2. **제한 사항**: 최대 9개의 레퍼런스 이미지, 3개의 레퍼런스 비디오, 3개의 레퍼런스 오디오 클립을 사용할 수 있습니다. 각 비디오 또는 오디오 클립은 2~15초여야 하며, 레퍼런스 비디오와 오디오의 총 길이는 각각 15초로 제한됩니다. 3. **오디오에는 시각적 앵커가 필요합니다**: 레퍼런스 오디오는 최소 한 개의 레퍼런스 이미지 또는 비디오가 없으면 사용할 수 없습니다. 4. **비디오 요구 사항**: 레퍼런스 비디오는 23.976~60 FPS여야 합니다. @@ -278,3 +279,43 @@ MiniMax는 레퍼런스 기반 생성(R2V)을 위한 공식 [전체 레퍼런스 워크플로용 레퍼런스 이미지입니다. 또는 직접 만든 이미지를 사용하세요. +## H3 Max 노드로 비디오 편집하기 + +H3 Max는 프롬프트만이 아니라 비디오나 정지 이미지에서 시작해 그 소재를 편집하는 파트너 노드 6개도 함께 제공합니다. Recast, Extend Video, Reframe Video, Insert Video, Lip Sync, Camera Control입니다. 이 노드들에는 템플릿이 제공되지 않습니다. **Partner Nodes > video > MiniMax** 아래의 노드 라이브러리에서 추가하고, 노드가 보이지 않으면 ComfyUI를 최신 버전으로 업데이트하세요. + +| 노드 | 기능 | 시작 미디어 | +| :--- | :----------- | :------------- | +| **MiniMax H3 Max Recast** | 비디오 속 인물을 레퍼런스 사진의 인물로 교체하며, 소스의 움직임, 카메라, 컷, 사운드를 유지합니다 | 5~30초 비디오, 사진 1~4장 | +| **MiniMax H3 Max Extend Video** | 프롬프트에 설명된 새 푸티지로 비디오를 이어서 만듭니다 | 1.625~60초 비디오, 1~15초의 새 푸티지 | +| **MiniMax H3 Max Reframe Video** | 비디오를 새 화면 비율에 맞추고 그 주변 영역을 생성하며, 타이밍과 사운드를 유지합니다 | 0.5~300초 비디오 | +| **MiniMax H3 Max Insert Video** | 두 타임스탬프 사이 구간을 새로 생성된 장면으로 교체합니다 | 최대 60초 비디오, 시작 시간과 재개 시간 | +| **MiniMax H3 Max Lip Sync** | 정지 이미지를 애니메이션하여 입 모양이 오디오 트랙을 따라가게 합니다 | 이미지 1장, 5~899초 오디오 | +| **MiniMax H3 Max Camera Control** | 정지 이미지 속 피사체 주위로 카메라를 회전시킵니다 | 이미지 1장, 종료 각도와 거리 | + +요금은 노드, 모델, 해상도, 재생 시간, 레퍼런스 입력에 따라 달라집니다. 가격 배지는 노드를 실행하기 전에 초당 요율, 대략적인 범위 또는 계산된 합계를 보여줍니다. 예를 들어 Extend Video는 H3 Max와 H3 Max Turbo를 모두 지원하며 선택한 재생 시간, 해상도, 모델로 합계를 계산합니다. 해상도별 요율은 [Minimax 가격표](/ko/tutorials/partner-nodes/pricing)에 있습니다. + +Recast, Extend Video, Insert Video의 프롬프트에서는 연결된 레퍼런스를 `@Image1`, `@Audio1`, `@Video1`로 지칭하며, 번호는 연결 순서를 따릅니다. 노드는 요청을 보내기 전에 각 태그를 대응하는 레퍼런스로 해석합니다. 해당 유형의 레퍼런스 개수를 넘는 태그가 있으면 오류와 함께 실행이 중단되고, 노드가 지원하지 않는 유형의 태그(예: Recast의 오디오)는 그대로 전송됩니다. + +### Recast + +소스 비디오의 인물을 1~4장의 레퍼런스 사진 속 인물로 교체하고, 원본의 카메라, 컷, 사운드를 유지합니다. 소스 비디오는 5~30초여야 하며, 그 안의 모든 샷은 15초 이하여야 하고 화면 비율은 0.4에서 2.5 사이여야 합니다. 사진은 최소 256x256 픽셀이어야 합니다. 기본적으로 왼쪽에서 오른쪽 순서로 주요 인물을 교체하며, 프롬프트로 인물을 하나씩 지정할 수 있습니다. 예를 들어 `the man becomes @Image2`와 같이 작성합니다. 기본적으로 768P 또는 1080P로 렌더링되며, **Enhance realism**은 피부 질감을 더 자연스럽게 만들고 새 인물을 장면에 자연스럽게 어우러지게 합니다. + +### Extend Video + +비디오 끝에 1~15초의 새 푸티지를 추가합니다. 프롬프트는 소스 푸티지가 아니라 다음에 일어나는 일을 설명합니다. Max 및 Turbo 모델과 1.625~60초, 최대 50 MB, 화면 비율 0.4에서 2.5 사이의 소스 비디오를 받습니다. 최대 3개의 선택적 레퍼런스 오디오로 새 푸티지를 유도할 수 있으며, 각 오디오는 최소 2초이고 합계 14.5초입니다. **Output**은 선택한 해상도로 전체 비디오를 다시 렌더링한 결과(`extended`) 또는 새 푸티지만(`continuation`, 소스 끝부분을 1.6초 반복하는 것으로 시작) 반환합니다. 해상도는 480P에서 2K까지이며, 2K에서는 `extended`가 소스와 새 푸티지를 합쳐 45초로 제한됩니다. + +### Reframe Video + +비디오를 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, 9:21 중 하나의 새 화면 비율로 다시 맞추고, 리프레임된 프레임 주변 영역을 생성합니다. 소스의 타이밍과 사운드는 유지됩니다. 소스 비디오는 0.5~300초일 수 있으며, 선택적 프롬프트가 생성되는 주변부의 내용을 안내합니다. 480P, 768P, 1080P로 렌더링됩니다. + +### Insert Video + +비디오에서 `start_time`과 `resume_time` 사이 구간을 5~13초의 새로 생성된 장면으로 교체합니다. 소스 비디오는 최대 60초까지 가능하며, `resume_time`은 비디오 끝에서 최소 1.375초 앞에 있어야 합니다. 프롬프트로 새 장면을 설명하거나 레퍼런스를 사용하거나 둘 다 할 수 있습니다. 최대 9장의 레퍼런스 이미지와 2~15초의 레퍼런스 비디오 3개(합계 15초)를 사용할 수 있습니다. **Color match**는 기본적으로 켜져 있으며 새 장면을 소스에 자연스럽게 어우러지게 합니다. 480P 또는 768P로 렌더링됩니다. + +### Lip Sync + +정지 이미지를 애니메이션하여 입 움직임이 오디오 트랙을 따라가게 하며, 비디오 길이는 오디오와 같습니다. 오디오는 5~899초일 수 있습니다. 이미지는 최소 256x256 픽셀이고 화면 비율이 0.4에서 2.5 사이여야 합니다. 480P, 768P, 1080P, 2K로 렌더링됩니다. **Transcription**은 기본적으로 꺼져 있으며, 켜면 오디오를 먼저 전사하고 그 텍스트를 사용해 동기화를 안내합니다. + +### Camera Control + +정지 이미지에서 카메라가 피사체 주위의 새 각도와 거리로 이동하는 비디오를 생성합니다. `azimuth`(수평 회전, 양수 값은 오른쪽으로 이동), `elevation`(양수 값은 아래를, 음수 값은 위를 바라봄), `distance`(1은 거리 유지, 1 미만은 더 가까이, 1 초과는 뒤로 물러남)로 종료 카메라를 설정합니다. **Create Camera Info** 또는 **Compose Camera Angle Prompt** 노드에서 온 카메라 정보 입력을 사용하면 대신 3D 뷰에서 각도를 설정합니다. 프롬프트를 비워 두면 장면이 고정된 채 카메라만 움직입니다. 재생 시간은 1~15초이고 480P, 768P, 1080P를 지원합니다. 방위각은 정확한 결과가 아니라 목표값입니다. 90도 요청은 흔히 30도에서 90도 사이에 도달합니다. diff --git a/tutorials/partner-nodes/minimax/minimax-h3.mdx b/tutorials/partner-nodes/minimax/minimax-h3.mdx index 8c54aecf0b..3b3ba46809 100644 --- a/tutorials/partner-nodes/minimax/minimax-h3.mdx +++ b/tutorials/partner-nodes/minimax/minimax-h3.mdx @@ -217,7 +217,7 @@ Generate videos conditioned on reference images, videos, and audio through the M #### Prompting tips -1. **Reference by tag**: Refer to each input by its order in the prompt, for example `Image 1`, `Video 1`, or `Audio 1` +1. **Reference by tag**: Refer to each input by its order in the prompt, for example `Image 1`, `Video 1`, or `Audio 1`. The `@Image1`, `@Video1`, and `@Audio1` shorthand is accepted as well and resolves to the same tags 2. **Limits**: Up to 9 reference images, 3 reference videos, and 3 reference audio clips. Each video or audio clip must be 2-15 seconds; total reference video and audio is capped at 15 seconds each 3. **Audio needs a visual anchor**: Reference audio cannot be used without at least one reference image or video 4. **Video requirements**: Reference videos must be 23.976-60 FPS @@ -271,3 +271,44 @@ Generate videos conditioned on reference media on the faster H3 Max model. The R Reference image for the workflow, or use your own images. + +## Editing a video with the H3 Max nodes + +H3 Max also ships six partner nodes that start from a video or a still image instead of a prompt alone, and edit that material: Recast, Extend Video, Reframe Video, Insert Video, Lip Sync, and Camera Control. No template ships with them. Add them from the node library under **Partner Nodes > video > MiniMax**, and update ComfyUI to the latest version if they are missing. + +| Node | What it does | Starting media | +| :--- | :----------- | :------------- | +| **MiniMax H3 Max Recast** | Replaces the people in a video with people from reference photos, keeping the motion, camera, cuts, and sound of the source | 5-30 second video, 1-4 photos | +| **MiniMax H3 Max Extend Video** | Continues a video with new footage described in the prompt | 1.625-60 second video, 1-15 seconds of new footage | +| **MiniMax H3 Max Reframe Video** | Fits a video into a new aspect ratio and generates the area around it, keeping timing and sound | 0.5-300 second video | +| **MiniMax H3 Max Insert Video** | Replaces the section between two timestamps with a newly generated scene | Video up to 60 seconds, plus a start and resume time | +| **MiniMax H3 Max Lip Sync** | Animates a still image so the mouth follows an audio track | One image, 5-899 seconds of audio | +| **MiniMax H3 Max Camera Control** | Orbits the camera around the subject in a still image | One image, plus an end angle and distance | + +Pricing depends on the node, model, resolution, duration, and reference inputs. The price badge can show a per-second rate, an approximate range, or a calculated total before you run the node; Extend Video, for example, supports both H3 Max and H3 Max Turbo and computes its total from the selected duration, resolution, and model. Rates per resolution are in the [Minimax pricing table](/tutorials/partner-nodes/pricing). + +In the Recast, Extend Video, and Insert Video prompts, refer to the connected references as `@Image1`, `@Audio1`, or `@Video1`, numbered in connection order. The node resolves each tag to the matching reference before sending the request. A tag that points past the references of a supported type stops the run with an error, while a tag for a type the node does not accept, such as audio on Recast, is passed through unchanged. + +### Recast + +Replaces the people in a source video with the people from 1 to 4 reference photos, keeping the camera, cuts, and sound of the original. The source video must run 5 to 30 seconds, with every shot inside it 15 seconds or shorter and an aspect ratio between 0.4 and 2.5. Photos must be at least 256x256 pixels. By default they replace the main people from left to right; use the prompt to assign them one by one, for example `the man becomes @Image2`. Renders at 768P by default or 1080P, and **Enhance realism** adds more natural skin texture and blends the new people into the scene. + +### Extend Video + +Adds 1 to 15 seconds of new footage to the end of a video. The prompt describes what happens next, not the source footage. Accepts the Max and Turbo models and a source video of 1.625 to 60 seconds, up to 50 MB, with an aspect ratio between 0.4 and 2.5. Up to 3 optional reference audios can steer the new footage, at least 2 seconds each and 14.5 seconds in total. **Output** returns either the whole video re-rendered at the selected resolution (`extended`) or only the new footage (`continuation`, which opens with a 1.6 second repeat of the source ending). Resolution runs from 480P to 2K; at 2K, `extended` is capped at 45 seconds of source plus new footage. + +### Reframe Video + +Refits a video into a new aspect ratio, either 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, or 9:21, by generating the area around the refitted frame. The timing and sound of the source are kept. Source video can run 0.5 to 300 seconds, and an optional prompt guides what the generated surroundings contain. Renders at 480P, 768P, or 1080P. + +### Insert Video + +Replaces the section of a video between `start_time` and `resume_time` with a newly generated scene of 5 to 13 seconds. The source video can run up to 60 seconds, and `resume_time` must fall at least 1.375 seconds before its end. Describe the new scene in the prompt, with references, or both: up to 9 reference images and 3 reference videos of 2 to 15 seconds each, 15 seconds in total. **Color match** is on by default and blends the new scene into the source. Renders at 480P or 768P. + +### Lip Sync + +Animates a still image so that its mouth movements follow an audio track, and the video is as long as the audio. The audio can run 5 to 899 seconds. The image must be at least 256x256 pixels with an aspect ratio between 0.4 and 2.5. Renders at 480P, 768P, 1080P, or 2K. **Transcription** is off by default; turn it on to transcribe the audio first and use the text to guide the synchronization. + +### Camera Control + +Generates a video from a still image with the camera moving to a new angle and distance around the subject. Set the end camera with `azimuth` (horizontal orbit, positive values move right), `elevation` (positive values look down, negative look up), and `distance` (1 keeps the distance, below 1 moves closer, above 1 pulls back). A camera info input from the Create Camera Info or Compose Camera Angle Prompt node sets the angle on a 3D view instead. An empty prompt keeps the scene frozen so that only the camera moves. Duration is 1 to 15 seconds at 480P, 768P, or 1080P. Azimuth is a target rather than an exact result: a 90 degree request often lands between 30 and 90 degrees. diff --git a/zh/tutorials/partner-nodes/minimax/minimax-h3.mdx b/zh/tutorials/partner-nodes/minimax/minimax-h3.mdx index d376905d38..3c7be6eeaa 100644 --- a/zh/tutorials/partner-nodes/minimax/minimax-h3.mdx +++ b/zh/tutorials/partner-nodes/minimax/minimax-h3.mdx @@ -2,13 +2,14 @@ title: "在 ComfyUI 中使用 MiniMax H3 API 进行视频生成" description: "在 ComfyUI 中使用 MiniMax H3 API 节点,通过云端工作流进行文生视频、首尾帧视频和参考生视频,支持原生立体声音频。" sidebarTitle: "Minimax H3" -translationSourceHash: b6d84062 +translationSourceHash: e44fb20e translationFrom: tutorials/partner-nodes/minimax/minimax-h3.mdx translationBlockHashes: "_intro": a261dbf5 "What MiniMax H3 is good at": 4c84d1e7 "Example outputs": 77256fd5 - "MiniMax H3 API workflows": 9f3a305f + "MiniMax H3 API workflows": 31840aba + "Editing a video with the H3 Max nodes": 21b27918 --- import ReqHint from "/snippets/zh/tutorials/partner-nodes/req-hint.mdx"; import UpdateReminder from "/snippets/zh/tutorials/update-reminder.mdx"; @@ -224,7 +225,7 @@ MiniMax 为基础生成模式(T2VA、I2VA、FL2VA 和 L2VA)发布了官方 #### 提示词技巧 -1. **按标签引用**:在提示词中按顺序引用每个输入,例如 `Image 1`、`Video 1` 或 `Audio 1`。 +1. **按标签引用**:在提示词中按顺序引用每个输入,例如 `Image 1`、`Video 1` 或 `Audio 1`。也接受 `@Image1`、`@Video1` 和 `@Audio1` 简写形式,它们会解析为同样的标签。 2. **限制**:最多 9 张参考图像、3 个参考视频和 3 段参考音频。每个视频或音频片段必须为 2–15 秒;参考视频与参考音频的总时长上限均为 15 秒。 3. **音频需要视觉锚点**:如果没有至少一个参考图像或参考视频,就无法使用参考音频。 4. **视频要求**:参考视频的帧率必须为 23.976–60 FPS。 @@ -278,3 +279,43 @@ MiniMax 为参考驱动生成(R2V)发布了一份官方的 [全参考模式 该工作流的参考图像,也可以使用你自己的图像。 +## 使用 H3 Max 节点编辑视频 + +H3 Max 还附带六个合作节点,它们以视频或静态图像为起点,而不是仅凭提示词,并对这些素材进行编辑:Recast、Extend Video、Reframe Video、Insert Video、Lip Sync 和 Camera Control。这些节点不随附模板。请从节点库中的 **Partner Nodes > video > MiniMax** 分组添加它们;如果找不到这些节点,请将 ComfyUI 更新到最新版本。 + +| 节点 | 作用 | 起始素材 | +| :--- | :----------- | :------------- | +| **MiniMax H3 Max Recast** | 用参考照片中的人物替换视频中的人物,保留源视频的动作、相机、剪辑和声音 | 5-30 秒视频,1-4 张照片 | +| **MiniMax H3 Max Extend Video** | 用提示词描述的新画面延续视频 | 1.625-60 秒视频,1-15 秒新画面 | +| **MiniMax H3 Max Reframe Video** | 将视频适配到新的宽高比,并生成其周围的区域,保留时间节奏和声音 | 0.5-300 秒视频 | +| **MiniMax H3 Max Insert Video** | 用新生成的场景替换两个时间点之间的片段 | 最长 60 秒视频,以及开始时间和恢复时间 | +| **MiniMax H3 Max Lip Sync** | 让静态图像动起来,使口型跟随音频轨道 | 一张图像,5-899 秒音频 | +| **MiniMax H3 Max Camera Control** | 让相机围绕静态图像中的主体环绕 | 一张图像,以及结束角度和距离 | + +计费取决于节点、模型、分辨率、时长和参考输入。价格徽章会在运行节点前显示每秒费率、近似区间或计算出的总额;例如 Extend Video 同时支持 H3 Max 和 H3 Max Turbo,会根据所选时长、分辨率和模型计算总额。各分辨率费率见 [Minimax 价格表](/zh/tutorials/partner-nodes/pricing)。 + +在 Recast、Extend Video 和 Insert Video 的提示词中,已连接的参考按连接顺序编号,请用 `@Image1`、`@Audio1` 或 `@Video1` 这样的标签引用它们。节点会在发送请求前把每个标签解析为对应的参考。指向该类型参考数量之外的标签会让运行报错停止;而节点不支持的类型(例如 Recast 中的音频)标签会原样发送。 + +### Recast + +用 1 到 4 张参考照片中的人物替换源视频中的人物,保留原始视频的相机、剪辑和声音。源视频时长必须为 5 到 30 秒,其中每个镜头不超过 15 秒,宽高比在 0.4 到 2.5 之间。照片至少为 256x256 像素。默认情况下,照片从左到右依次替换主要人物;可在提示词中逐个指定,例如 `the man becomes @Image2`。默认以 768P 渲染,也可选择 1080P;**Enhance realism** 会生成更自然的皮肤纹理,并让新人物更好地融入场景。 + +### Extend Video + +在视频末尾添加 1 到 15 秒的新画面。提示词描述接下来发生的内容,而不是源画面。支持 Max 和 Turbo 模型,以及时长 1.625 到 60 秒、最大 50 MB、宽高比在 0.4 到 2.5 之间的源视频。最多 3 个可选参考音频可引导新画面,每个至少 2 秒,总计 14.5 秒。**Output** 返回按所选分辨率重新渲染的完整视频(`extended`),或仅返回新画面(`continuation`,开头会重复源视频结尾的 1.6 秒)。分辨率范围为 480P 到 2K;在 2K 下,`extended` 的源视频加新画面总时长上限为 45 秒。 + +### Reframe Video + +通过生成重新适配后的画面周围的区域,把视频适配为新的宽高比,可选 16:9、9:16、1:1、4:3、3:4、21:9 或 9:21。保留源视频的时间节奏和声音。源视频时长可为 0.5 到 300 秒,可选提示词可引导生成的环境内容。以 480P、768P 或 1080P 渲染。 + +### Insert Video + +用新生成的 5 到 13 秒场景替换视频中 `start_time` 与 `resume_time` 之间的片段。源视频最长 60 秒,且 `resume_time` 必须比视频结尾至少提前 1.375 秒。可在提示词中描述新场景、使用参考,或两者结合:最多 9 张参考图像和 3 个参考视频,每个参考视频 2 到 15 秒,总计 15 秒。**Color match** 默认开启,会让新场景融入源视频。以 480P 或 768P 渲染。 + +### Lip Sync + +让静态图像动起来,使其口型运动跟随音频轨道,视频时长与音频相同。音频时长可为 5 到 899 秒。图像至少为 256x256 像素,宽高比在 0.4 到 2.5 之间。以 480P、768P、1080P 或 2K 渲染。**Transcription** 默认关闭;开启后会先转录音频,并用文本引导同步。 + +### Camera Control + +从静态图像生成视频,相机会围绕主体移动到新的角度和距离。使用 `azimuth`(水平环绕,正值向右移动)、`elevation`(正值向下看,负值向上看)和 `distance`(1 保持距离,小于 1 拉近,大于 1 拉远)设置结束时的相机位置。来自 **Create Camera Info** 或 **Compose Camera Angle Prompt** 节点的相机信息输入则改为在 3D 视图中设置角度。空提示词会让场景保持静止,只有相机会移动。时长为 1 到 15 秒,分辨率为 480P、768P 或 1080P。Azimuth 是目标值而不是精确结果:请求 90 度时,结果通常落在 30 到 90 度之间。