음성 복제로 UGC 동영상 생성
이미지를 업로드하여 AI 프롬프트를 생성하고 음성을 합성합니다. Eleven Labs와 LTX Video를 사용하여 정확한 립싱크가 적용된 최종 동영상을 제작하세요.
워크플로 내부
소스 템플릿의 읽기 전용 미리보기입니다. Cloud에서 열어 노드를 살펴보고 변경할 수 있습니다.
코드에서 이 모델 호출하기
아래 요청은 Playground와 동일하게 준비된 그래프와 설정을 사용합니다. 제출할 때마다 유료 실행이 시작될 수 있으므로 결과를 알 수 없는 제출을 자동으로 재시도하지 마세요.
What you need
- Runs on
- Comfy Cloud
- 키
- COMFY_API_KEY
- 내 파일
- 코드가 해당 URL에서 업로드
유료 Cloud 요금제와 크레딧이 필요합니다.
# Python 3.10+: pip install comfy-sdk==0.4.0
import os
from comfy_sdk import Comfy
api_key = os.environ.get("COMFY_API_KEY", "").strip()
if not api_key: raise RuntimeError("Set COMFY_API_KEY")
os.environ.setdefault("COMFY_BASE_URL", "https://cloud.comfy.org")
client = Comfy(api_key=api_key)
workflow = client.workflows.from_json({
"3": {
"class_type": "ElevenLabsTextToSpeech",
"inputs": {
"apply_text_normalization": "auto",
"language_code": "",
"model": "eleven_v3",
"model.similarity_boost": 0.75,
"model.speed": 1,
"output_format": "mp3_44100_192",
"seed": 3333,
"stability": 0.5,
"text": [
"475",
0
],
"voice": [
"442",
0
]
},
"_meta": {
"title": "ElevenLabs Text to Speech"
}
},
"440": {
"class_type": "LoadImage",
"inputs": {
"image": "UPLOADED_image_FILENAME"
},
"_meta": {
"title": "Load Image"
}
},
"441": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"440",
0
],
"model": "gemini-3-1-pro",
"prompt": "A friendly, honest product review for a small amber jar of overnight repair cream, filmed like a casual morning video at home. She has just woken up on a rainy day, holds the jar up next to her cheek and talks straight to the camera like she is telling a friend about something she genuinely loves. Open with a quick, surprising hook about waking up with skin that looks rested even after a short night. Mention one clear benefit: it soaks in fast and is not greasy, so she can put it on and go straight to bed. Keep the tone warm, a little playful and relaxed, with a small laugh. End with a simple, light recommendation to try it, no hard sell and no prices.",
"seed": 967822922709019,
"system_prompt": "Role: You are a professional prompt writer for Eleven Labs v3 two-part generation.\n\nTask: Generate a two-part prompt from the user’s input.\n\nStrict Output Constraint (IMPORTANT):\n\nReturn the output as RAW PLAIN TEXT only.\nDO NOT use Markdown code blocks.\nDO NOT use any headers, labels, or bold text outside of the generated two-part output.\nUse ONLY the separator --- to divide the two parts.\nPart 1: spoken dialogue/performance text only, optimized for Eleven v3.\nPart 2: a structured scene breakdown in short field lines.\nDo NOT output anything before Part 1 or after Part 2.\n\nGlobal Length Rule:\n\nKeep the final generated prompt short enough to produce about 10 seconds of audio.\nAim for roughly 20 to 35 spoken words in Part 1 unless the user explicitly asks otherwise.\nCondense long inputs aggressively while preserving the core meaning, tone, and key message.\nPrioritize brevity, clarity, speakability, and a strong opening hook over completeness.\nRemove repetition, side details, filler, and nonessential context.\nIf the input is too long, compress it into a concise spoken version rather than preserving everything.\nDo not create Part 1 output that would likely exceed about 10 seconds of speech.\n\nPart 1 Rules:\n\nPart 1 must contain ONLY the words meant to be spoken aloud.\nDo NOT begin Part 1 with visual description, scene-setting narration, character description, or action description.\nStart immediately with the most attention-grabbing, high-impact spoken hook.\nThe hook should appear as early as possible, ideally in the first sentence or phrase.\nDo NOT describe what the character looks like, where they are, what they are wearing, what they are holding, or what is happening visually unless that information is spoken by the character as part of the dialogue itself.\nPart 1 should read like a real spoken performance, monologue, ad read, or direct address.\nPart 1 may use Eleven v3-compatible vocal tags in square brackets, such as [laughs], [whispers], [sighs], [excited], [curious], [sarcastic], [exhales], [crying], [snorts], [mischievously], [sings], when helpful.\nUse punctuation, ellipses, pauses, and selective capitalization to improve spoken delivery when useful.\nDo NOT use SSML.\nDo NOT use non-voice tags such as music, ambience, sound effects, camera notes, or technical markup.\nCompress long user text into the shortest strong spoken version that preserves the selling point or core message.\nPrefer one clear hook, one core benefit, and one short closing beat.\n\nPart 2 Rules:\n\nPart 2 must transform the same idea into a compact structured format.\nPart 2 must strongly bias toward UGC-style action and framing and prnounce the character is talking into the camera.\nPart 2 must describe the subject as if they are casually filming a real user-generated testimonial, ad, or selfie-style clip.\nPrioritize natural human behavior such as:\ntalking directly to camera\nhandheld or phone-like framing\ncasual presenting of product\nnatural gestures\nauthentic facial expression\nslight body movement\nconversational delivery\nAvoid cinematic, polished commercial, theatrical, or overly staged action unless the user explicitly asks for that.\nKeep the action grounded, believable, and creator-like rather than scripted or broadcast-like.\nEach line should follow this pattern: field name, colon, content.\nUse only concise production-style descriptors.\nKeep Part 2 aligned with Part 1 and do not introduce unrelated ideas.\nPrefer fields such as:\nscene:\ncharacter:\naction:\ncamera:\ntone:\nUse action: to emphasize the UGC performance behavior most strongly.\nUse camera: to emphasize casual capture style, such as handheld, selfie framing, direct-to-camera, slight natural movement, or eye-level creator framing.\nKeep wardrobe and appearance brief unless they are essential to the concept.\nKeep Part 2 compact and production-friendly.\n\nUGC Bias Rules for Part 2:\n\nDefault to direct-to-camera testimonial energy unless the user asks otherwise.\nDefault to natural handheld realism unless the user asks otherwise.\nDefault to real-person creator behavior instead of actor blocking unless the user asks otherwise.\nPrefer verbs like:\ntalking, holding, showing, gesturing, reacting, smiling, leaning in, presenting, pointing\nAvoid verbs like:\nposing, performing choreography, dramatically turning, stylized modeling\nMake the structured breakdown help the model generate believable UGC action, not just static appearance.\n\nConsistency Rules:\n\nPart 1 and Part 2 must describe the same scene and intent.\nPart 2 is a clean structural breakdown of Part 1.\nDo not make Part 1 and Part 2 feel like separate concepts.\nPreserve the user’s intent, subject, tone, setting, and spoken energy across both parts.\n\nSafety and Quality Rules:\n\nDo not introduce sensitive, explicit, hateful, political, or unsafe content that was not already in the user input.\nKeep vocal tags contextually appropriate and naturally speakable.\nUse expressive enhancements sparingly and intentionally.\nEnsure the final output always contains exactly two parts separated by ---."
},
"_meta": {
"title": "Google Gemini"
}
},
"442": {
"class_type": "ElevenLabsVoiceSelector",
"inputs": {
"voice": "Bella (female, american)"
},
"_meta": {
"title": "ElevenLabs Voice Selector"
}
},
"443": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"441",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"445": {
"class_type": "SaveAudioMP3",
"inputs": {
"audio": [
"3",
0
],
"filename_prefix": "audio/text2speech2video",
"quality": "V0"
},
"_meta": {
"title": "Save Audio (MP3) (DEPRECATED)"
}
},
"473": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"476",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"474": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"475",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"475": {
"class_type": "RegexExtract",
"inputs": {
"case_insensitive": True,
"dotall": False,
"group_index": 1,
"mode": "First Group",
"multiline": False,
"regex_pattern": "(?sm)^(.*?)^---$(.*)",
"string": [
"441",
0
]
},
"_meta": {
"title": "Extract Text"
}
},
"476": {
"class_type": "RegexExtract",
"inputs": {
"case_insensitive": True,
"dotall": False,
"group_index": 1,
"mode": "First Group",
"multiline": False,
"regex_pattern": "(?s)(?<=\\n---\\n)([\\s\\S]*)$",
"string": [
"441",
0
]
},
"_meta": {
"title": "Extract Text"
}
},
"479": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video/ComfyUI",
"format": "auto",
"format.codec": "auto",
"video": [
"478:312",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"478:285": {
"class_type": "RandomNoise",
"inputs": {
"noise_seed": 42
},
"_meta": {
"title": "RandomNoise"
}
},
"478:286": {
"class_type": "RandomNoise",
"inputs": {
"noise_seed": 3333
},
"_meta": {
"title": "RandomNoise"
}
},
"478:287": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"478:309",
1
],
"video_latent": [
"478:296",
0
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"478:288": {
"class_type": "KSamplerSelect",
"inputs": {
"sampler_name": "euler_cfg_pp"
},
"_meta": {
"title": "KSamplerSelect"
}
},
"478:289": {
"class_type": "ManualSigmas",
"inputs": {
"sigmas": "0.85, 0.7250, 0.4219, 0.0"
},
"_meta": {
"title": "ManualSigmas"
}
},
"478:290": {
"class_type": "CFGGuider",
"inputs": {
"cfg": 1,
"model": [
"478:601",
0
],
"negative": [
"478:292",
1
],
"positive": [
"478:292",
0
]
},
"_meta": {
"title": "CFG Guider"
}
},
"478:291": {
"class_type": "SamplerCustomAdvanced",
"inputs": {
"guider": [
"478:315",
0
],
"latent_image": [
"478:326",
0
],
"noise": [
"478:286",
0
],
"sampler": [
"478:298",
0
],
"sigmas": [
"478:308",
0
]
},
"_meta": {
"title": "SamplerCustomAdvanced"
}
},
"478:292": {
"class_type": "LTXVCropGuides",
"inputs": {
"latent": [
"478:309",
0
],
"negative": [
"478:307",
1
],
"positive": [
"478:307",
0
]
},
"_meta": {
"title": "LTXVCropGuides"
}
},
"478:293": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-22b-distilled-lora-384.safetensors",
"model": [
"478:317",
0
],
"strength_model": 0.5
},
"_meta": {
"title": "Load LoRA"
}
},
"478:294": {
"class_type": "ResizeImagesByLongerEdge",
"inputs": {
"images": [
"478:297",
0
],
"longer_edge": 1536
},
"_meta": {
"title": "Resize Images by Longer Edge (DEPRECATED)"
}
},
"478:295": {
"class_type": "LTXVLatentUpsampler",
"inputs": {
"samples": [
"478:309",
0
],
"upscale_model": [
"478:313",
0
],
"vae": [
"478:317",
2
]
},
"_meta": {
"title": "LTXVLatentUpsampler"
}
},
"478:296": {
"class_type": "LTXVImgToVideoInplace",
"inputs": {
"bypass": [
"478:305",
0
],
"image": [
"478:334",
0
],
"latent": [
"478:295",
0
],
"strength": 1,
"vae": [
"478:317",
2
]
},
"_meta": {
"title": "LTXVImgToVideoInplace"
}
},
"478:297": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"440",
0
],
"resize_type": "scale dimensions",
"resize_type.crop": "center",
"resize_type.height": [
"478:324",
0
],
"resize_type.width": [
"478:330",
0
],
"scale_method": "lanczos"
},
"_meta": {
"title": "Resize Image/Mask"
}
},
"478:298": {
"class_type": "KSamplerSelect",
"inputs": {
"sampler_name": "euler_ancestral_cfg_pp"
},
"_meta": {
"title": "KSamplerSelect"
}
},
"478:299": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a/2",
"values.a": [
"478:330",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"478:301": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a/2",
"values.a": [
"478:324",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"478:302": {
"class_type": "EmptyLTXVLatentVideo",
"inputs": {
"batch_size": 1,
"height": [
"478:301",
1
],
"length": [
"478:329",
1
],
"width": [
"478:299",
1
]
},
"_meta": {
"title": "Empty LTXV Latent Video"
}
},
"478:303": {
"class_type": "LTXVAudioVAEDecode",
"inputs": {
"audio_vae": [
"478:335",
0
],
"samples": [
"478:311",
1
]
},
"_meta": {
"title": "LTXV Audio VAE Decode"
}
},
"478:304": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a",
"values.a": [
"478:323",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"478:305": {
"class_type": "PrimitiveBoolean",
"inputs": {
"value": False
},
"_meta": {
"title": "Switch to Text to Video?"
}
},
"478:306": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"478:318",
0
],
"text": [
"478:319",
0
]
},
"_meta": {
"title": "CLIP Text Encode (Prompt)"
}
},
"478:307": {
"class_type": "LTXVConditioning",
"inputs": {
"frame_rate": [
"478:304",
0
],
"negative": [
"478:314",
0
],
"positive": [
"478:306",
0
]
},
"_meta": {
"title": "LTXVConditioning"
}
},
"478:308": {
"class_type": "ManualSigmas",
"inputs": {
"sigmas": "1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0"
},
"_meta": {
"title": "ManualSigmas"
}
},
"478:309": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"478:291",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"478:310": {
"class_type": "SamplerCustomAdvanced",
"inputs": {
"guider": [
"478:290",
0
],
"latent_image": [
"478:287",
0
],
"noise": [
"478:285",
0
],
"sampler": [
"478:288",
0
],
"sigmas": [
"478:289",
0
]
},
"_meta": {
"title": "SamplerCustomAdvanced"
}
},
"478:311": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"478:310",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"478:312": {
"class_type": "CreateVideo",
"inputs": {
"audio": [
"478:303",
0
],
"bit_depth": "auto",
"codec": "none",
"color_space": "sRGB",
"fps": [
"478:304",
0
],
"images": [
"478:316",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"478:313": {
"class_type": "LatentUpscaleModelLoader",
"inputs": {
"model_name": "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"
},
"_meta": {
"title": "Load Latent Upscale Model"
}
},
"478:314": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"478:318",
0
],
"text": "pc game, console game, video game, cartoon, childish, ugly"
},
"_meta": {
"title": "CLIP Text Encode (Prompt)"
}
},
"478:315": {
"class_type": "CFGGuider",
"inputs": {
"cfg": 1,
"model": [
"478:601",
0
],
"negative": [
"478:307",
1
],
"positive": [
"478:307",
0
]
},
"_meta": {
"title": "CFG Guider"
}
},
"478:316": {
"class_type": "VAEDecodeTiled",
"inputs": {
"overlap": 64,
"samples": [
"478:311",
0
],
"temporal_overlap": 4,
"temporal_size": 4096,
"tile_size": 768,
"vae": [
"478:317",
2
]
},
"_meta": {
"title": "VAE Decode (Tiled)"
}
},
"478:317": {
"class_type": "CheckpointLoaderSimple",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors"
},
"_meta": {
"title": "Load Checkpoint"
}
},
"478:318": {
"class_type": "LTXAVTextEncoderLoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors",
"device": "default",
"text_encoder": "gemma_3_12B_it_fp4_mixed.safetensors"
},
"_meta": {
"title": "Load LTXV Audio Text Encoder"
}
},
"478:319": {
"class_type": "PrimitiveStringMultiline",
"inputs": {
"value": [
"476",
0
]
},
"_meta": {
"title": "Prompt"
}
},
"478:323": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 24
},
"_meta": {
"title": "Frame Rate"
}
},
"478:324": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 720
},
"_meta": {
"title": "Height"
}
},
"478:325": {
"class_type": "LTXVImgToVideoInplace",
"inputs": {
"bypass": [
"478:305",
0
],
"image": [
"478:334",
0
],
"latent": [
"478:302",
0
],
"strength": 0.7,
"vae": [
"478:317",
2
]
},
"_meta": {
"title": "LTXVImgToVideoInplace"
}
},
"478:326": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"478:327",
0
],
"video_latent": [
"478:325",
0
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"478:327": {
"class_type": "SetLatentNoiseMask",
"inputs": {
"mask": [
"478:333",
0
],
"samples": [
"478:328",
0
]
},
"_meta": {
"title": "Set Latent Noise Mask"
}
},
"478:328": {
"class_type": "LTXVAudioVAEEncode",
"inputs": {
"audio": [
"478:332",
0
],
"audio_vae": [
"478:335",
0
]
},
"_meta": {
"title": "LTXV Audio VAE Encode"
}
},
"478:329": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a * b + 1",
"values.a": [
"478:331",
0
],
"values.b": [
"478:323",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"478:330": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 1280
},
"_meta": {
"title": "Width"
}
},
"478:331": {
"class_type": "PrimitiveFloat",
"inputs": {
"value": 12
},
"_meta": {
"title": "Duration"
}
},
"478:332": {
"class_type": "TrimAudioDuration",
"inputs": {
"audio": [
"3",
0
],
"duration": [
"478:331",
0
],
"start_index": 0
},
"_meta": {
"title": "Trim Audio Duration"
}
},
"478:333": {
"class_type": "SolidMask",
"inputs": {
"height": [
"478:324",
0
],
"value": 0,
"width": [
"478:330",
0
]
},
"_meta": {
"title": "Create Solid Mask"
}
},
"478:334": {
"class_type": "LTXVPreprocess",
"inputs": {
"image": [
"478:294",
0
],
"img_compression": 18
},
"_meta": {
"title": "LTXV Preprocess"
}
},
"478:335": {
"class_type": "LTXVAudioVAELoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors"
},
"_meta": {
"title": "Load LTXV Audio VAE"
}
},
"478:601": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-id-lora-talkvid-3k.safetensors",
"model": [
"478:293",
0
],
"strength_model": 1
},
"_meta": {
"title": "Load LoRA"
}
}
})
workflow.set_input("440", "image", client.assets.from_url("https://comfy.org/catalog-media/75f8228c-37f2-4dcd-bd25-b02f020d9a6a/33ce8181-b949-4c83-93ce-1613d967d887"))
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)
for output in job.get_outputs("479"):
output.to_file(output.name)
for output in job.get_outputs("445"):
output.to_file(output.name)예시 사용해 보기
이 워크플로 정보
이 ComfyUI 튜토리얼 워크플로는 단일 이미지를 짧은 립싱크 UGC 스타일 비디오로 변환하며, 복제 또는 선택한 음성을 사용할 수 있습니다. 세 단계로 구성됩니다: 업로드한 이미지에서 프롬프트 생성, ElevenLabs로 음성 합성, LTX‑2.3으로 비디오 생성. Create Prompt 그룹은 LoadImage에서 이미지를 받아 GeminiNode로 전달해 표현 태그가 포함된 연설 스크립트와 장면 설명 두 가지 텍스트를 자동 생성합니다. RegexExtract가 Gemini 응답을 파싱해 “speech”와 “scene” 섹션을 분리해 후속 노드로 전달합니다.
오디오 단계에서는 ElevenLabsVoiceSelector로 프리셋 음성을 선택하거나 ElevenLabsInstantVoiceClone으로 직접 음색을 복제할 수 있으며, 최종 내레이션은 ElevenLabsTextToSpeech에서 합성하고 필요시 SaveAudioMP3로 저장할 수 있습니다. Create Video 그룹은 장면 설명과 오디오를 LTX‑2.3 비디오 노드(UUID: 98fb87e2-23b5-4ecb-aacc-365912414a12)에 전달해 립싱크가 정확한 토킹 스타일 비디오를 생성합니다. PreviewAny로 미리보기하고 SaveVideo로 저장합니다. 이 구성은 이미지에서 자동으로 스크립트를 작성하고, ElevenLabs + LTX로 음성-입술 정렬을 보장해 UGC, 제품 설명, 후기, 소셜 포스트에 실용적입니다.
질문과 답변
네. 프롬프트 생성을 위한 GeminiNode에 Gemini API 키, 음성 선택/복제 및 TTS용 ElevenLabs 노드에 ElevenLabs API 키가 필요합니다. Gemini가 없다면 speech/scene 프롬프트를 수동으로 작성해 해당 단계를 건너뛸 수 있습니다.
ElevenLabsTextToSpeech에서 생성된 내레이션을 LTX‑2.3 비디오 노드에 직접 전달해 오디오에 맞춰 입술 움직임을 정렬합니다. 최종 TTS 파일을 그대로 사용하는 것이 정확한 립싱크를 보장합니다.
네. LoadAudio로 내레이션을 입력해 Create Video 그룹의 TTS 오디오 위치에 연결하세요. 이 경우 ElevenLabsTextToSpeech 단계는 건너뜁니다.
GeminiNode는 연설 스크립트와 장면 설명을 모두 반환합니다. RegexExtract는 이 응답을 speech 프롬프트(ElevenLabsTextToSpeech용)와 scene 프롬프트(LTX‑2.3 비디오 노드용)로 깔끔하게 분리해 파이프라인의 신뢰성을 높입니다.