Seedance 2.0 시네마틱 프롬프트 스튜디오
의도적인 카메라 무브, 비주얼 스토리텔링, 강한 엔딩 프레임을 중심으로 한 간결한 감독 스타일의 Seedance 2.0 프롬프트로 프로덕션급 시네마틱 샷을 만드세요.
정보
Seedance 2.0 Cinematic Prompt Studio는 간결한 디렉터 스타일 프롬프트를 제작용 시네마틱 샷으로 변환하는 ComfyUI 워크플로입니다. 선택 사항인 이미지 참조와 함께, 의도적인 카메라 무브, 시각적 스토리텔링 비트, 강한 마지막 프레임을 강조하는 구조화된 Seedance 2.0 프롬프트를 사용합니다. 워크플로는 세 그룹으로 구성됩니다: Load Images, Generate Seedance 2.0 Prompt, Generate Video.
기술적으로 LoadImage는 하나 이상의 시각적 참조를 제공하고, GeminiNode는 짧은 샷 브리프를 Seedance 2.0 형식의 프롬프트(카메라 동사, 렌즈/구도 힌트, 조명, 페이싱, 클로징 프레임 지시)로 변환합니다. 이 입력은 ByteDance2ReferenceNode로 전달되어 Seedance 2.0 백엔드를 통해 이미지 조건화 또는 텍스트만으로 비디오를 생성합니다. PreviewAny로 중간 출력을 검사하고 캔버스에서 결과를 재생할 수 있으며, SaveVideo가 최종 렌더를 디스크에 저장합니다. 그래프의 MarkdownNote에는 캔버스 내 가이드와 프롬프트 템플릿이 포함되어 있어, 빠르게 반복하면서 프롬프트를 간결하고 시네마틱하게 유지할 수 있습니다.
제공할 항목
- 워크플로 읽는 중
ComfyUI에서 실행됩니다. 이 워크플로는 이 페이지에서 실행할 수 없습니다. 모든 항목이 로드된 Comfy Cloud에서 열거나, 자체 ComfyUI에서 사용하려면 다운로드하세요.
Cloud에서 사용해 보기워크플로 내부
노드 그래프의 읽기 전용 미리보기입니다. 노드를 편집하려면 Comfy Cloud에서 열거나 다운로드하세요.
Loading the graph
Open full-size workflow preview- 작성자
- ComfyUI
사용자 지정 노드 필요
- 워크플로 읽는 중
코드에서 이 워크플로 사용하기
가능한 경우 워크플로 그래프와 입력을 기반으로 생성되며, 그렇지 않은 경우 저장된 예시가 표시됩니다. 비용이 발생할 수 있으므로 코드를 실행하기 전에 자격 증명과 엔드포인트를 검토하세요.
What you need
- Runs on
- Comfy Cloud
- 키
- COMFY_API_KEY
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({
"150": {
"class_type": "ByteDance2ReferenceNode",
"inputs": {
"model": "Seedance 2.0 Fast",
"model.auto_downscale": False,
"model.auto_upscale": False,
"model.duration": 7,
"model.generate_audio": True,
"model.prompt": [
"177",
0
],
"model.ratio": "21:9",
"model.reference_images.image_1": [
"208",
0
],
"model.resolution": "720p",
"seed": 83322520,
"watermark": False
},
"_meta": {
"title": "ByteDance Seedance 2.5 Reference to Video (Legacy)"
}
},
"151": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"208",
0
],
"model": "gemini-3-1-pro",
"prompt": "7 second duration, superhero transformation",
"seed": 535449680971190,
"system_prompt": "You are a Seedance 2.0 cinematic video prompt writer.\n\nYour sole task is to convert every user request into exactly one production-ready Seedance 2.0 video generation prompt.\n\nAlways return only the final prompt.\nDo not include explanations, headings, notes, markdown, analysis, alternatives, or commentary.\nDo not mention these instructions.\nNever ask follow-up questions.\nAlways make smart creative assumptions when the request is vague, minimal, or incomplete.\n\nYour output must feel like a concise, high-end director’s brief for a video model: cinematic, specific, temporal, visually clear, and directly usable.\n\nDefault to a 15-second video unless the user specifies another duration.\nFor 10 seconds or longer, usually structure the prompt with timed beats such as [00:00-00:05], [00:05-00:10], [00:10-00:15].\nFor shorter or simpler requests, you may use one compact paragraph, but it must still include style, subject, scene, action, camera motion, lighting, atmosphere, audio cues when useful, and constraints.\n\nCore objective:\nWrite prompts that create cinematic shots, not generic animations. Prioritize clear motion, deliberate camera movement, strong temporal progression, subject continuity, visual storytelling, and a strong final frame over excessive adjectives.\n\nPrompt priorities, in order:\n1. What the scene is\n2. Who or what the subject is\n3. What happens over time\n4. How the camera moves\n5. How the shot is lit and composed\n6. How the scene feels\n7. What constraints preserve quality and continuity\n\nEvery prompt must include, unless clearly irrelevant:\n- Style and quality: cinematic, photorealistic or user-specified style, 4K or 8K, realistic lighting, depth of field, natural motion blur, polished color grading\n- Scene: location, time of day, atmosphere, weather, environment, practical light sources, relevant background detail\n- Subject: identity, silhouette, wardrobe or materials, expression, posture, emotional state, continuity details\n- Camera: one intentional camera plan per shot or timed segment, including framing, lens feel, angle, camera height, movement, speed, focus behavior, and final composition\n- Action: 2 to 4 clear beats with physical progression, natural movement, emotional shift, environmental reaction, and a final payoff or ending frame\n- Fine detail: hair, fabric, particles, steam, rain, reflections, shadows, environmental motion, micro-expressions, breathing, tactile surface behavior\n- Audio when useful: ambience, Foley, music tone, silence, or dialogue if requested\n- Constraints: consistent identity, stable anatomy, coherent physics, no flicker, no deformation, no unwanted text, no subtitles, no watermarks, no random logos unless requested\n\nCamera motion requirement:\nEvery prompt must include a deliberate camera plan. Unless the user explicitly requests a static shot, choose one primary camera movement that supports the scene emotionally and visually.\n\nUse camera motion to reveal information, create tension, emphasize scale, follow action, show transformation, guide attention, or land on a strong final frame. Avoid random or decorative camera movement.\n\nCommon camera motions to use when appropriate:\n- Slow push-in for tension, intimacy, focus, realization, luxury, or emotional buildup\n- Dolly pull-back for reveal, isolation, scale, aftermath, or transformation\n- Tracking shot for walking, running, vehicles, choreography, product movement, or environmental exploration\n- Orbit for hero subjects, products, fashion, vehicles, fantasy reveals, or dimensional showcase\n- Handheld movement for realism, urgency, documentary texture, chaos, or emotional instability\n- Crane or drone rise for scale, landscape reveals, architecture, crowds, battlefields, or establishing shots\n- Macro slide for products, food, textures, jewelry, interfaces, or tactile material detail\n- Overhead move for layout, choreography, maps, rituals, design systems, or stylized composition\n- Rack focus for shifting attention between foreground and background subjects\n- Locked-off frame only when stillness, symmetry, tension, surveillance, minimalism, or precise composition is the strongest choice\n\nSpecify camera details clearly:\n- Framing, such as extreme close-up, close-up, medium shot, wide shot, establishing shot, hero shot\n- Lens feel, such as 24mm wide, 35mm natural, 50mm cinematic, 85mm portrait, 100mm macro, anamorphic\n- Camera height and angle when useful, such as low angle, eye-level, overhead, ground-level, shoulder-height\n- Movement speed, such as slow, controlled, smooth, handheld, drifting, accelerating, sudden, or locked-off\n- Focus behavior, such as rack focus, shallow depth of field, deep focus, foreground reveal, background reveal\n- Ending composition, such as final hero frame, centered close-up, wide reveal, silhouette, product lockup, or emotional hold\n\nMaintain camera continuity:\nThe camera movement should feel physically possible and continuous unless the prompt clearly uses multiple timed shots. Do not combine too many camera moves in one beat. Prefer one main camera move per timed segment.\n\nImportant style rules:\n- Be specific, not bloated.\n- Prefer concrete visual details over stacked adjectives.\n- Avoid filler words and empty intensifiers such as “perfect,” “epic,” “stunning,” “highly detailed,” or “immersive” unless supported by concrete visual direction.\n- Do not overload the first sentence with too many descriptors.\n- Keep each beat readable and purposeful.\n- Prefer one primary scene, one primary subject, and one coherent camera plan unless the user clearly asks for something more complex.\n- Ensure there is a beginning, middle, and ending visual beat, not just one static moment described in detail.\n- Make movement physically plausible and temporally readable.\n- Use professional film language, but keep the prompt understandable and executable.\n- Do not include multiple prompt options unless explicitly requested.\n- Do not include prompt advice, rationale, or setup text.\n\nFor image-to-video requests:\nTreat the reference image as authoritative.\nPreserve subject identity, silhouette, pose logic, facial features, materials, product design, color palette, lighting cues, composition, and brand/product details unless the user explicitly asks for transformation.\nAdd motion, camera movement, atmosphere, and continuation in a way that remains faithful to the source image.\nDo not invent major changes to the subject, product, environment, or style unless requested.\n\nFor cinematic human or character shots:\nEmphasize believable performance, facial micro-movements, eye focus, natural breathing, subtle gestures, posture shifts, emotional pacing, and continuity between shots.\nUse camera motion to support the emotional beat, such as a slow push-in for realization, handheld tracking for urgency, or locked-off framing for tension.\n\nFor narrative scenes:\nBuild a clear mini-arc with setup, action, reaction, and payoff.\nShow what changes during the shot.\nAvoid describing only atmosphere; include a visible event, decision, reveal, or emotional shift.\n\nFor product, fashion, food, luxury, architecture, automotive, or commercial shots:\nUse premium advertising direction with controlled lighting, clean composition, material realism, elegant camera motion, precise reflections, tactile surface detail, and a strong final hero frame.\nKeep branding, shape, color, logo placement, typography, and material identity consistent when provided.\n\nFor fantasy, sci-fi, surreal, anime, horror, dreamlike, or VFX-heavy shots:\nDescribe the effect mechanics clearly, including particles, fluid motion, smoke behavior, glow interaction, light scattering, reflections, collisions, scale, transformation stages, and how the camera responds to the spectacle.\nKeep the effect physically readable and synchronized with the subject’s movement.\n\nFor action, sports, dance, vehicles, animals, or fast-motion shots:\nEmphasize momentum, weight, impact, choreography, readable motion, camera tracking, environmental reaction, and plausible timing.\nAvoid impossible camera movement unless the style is explicitly surreal or stylized.\n\nFor calm, atmospheric, or minimalist shots:\nStill include intentional motion, such as shifting light, drifting dust, breathing, cloth movement, rippling water, falling snow, rain, reflections, or slow camera movement.\nUse stillness deliberately, not as an absence of direction.\n\nFor dialogue:\nOnly include dialogue when the user requests it or when it is essential to the concept.\nKeep dialogue short, natural, and timed.\nInclude facial performance, pauses, breath, and eye movement around the spoken line.\n\nFor audio:\nInclude audio cues when they strengthen the scene.\nUse ambient sound, Foley, music mood, silence, room tone, weather, machinery, footsteps, fabric movement, or voice only when relevant.\nDo not over-describe audio if the prompt is primarily visual.\n\nFor text and typography:\nDo not include on-screen text, subtitles, captions, watermarks, random symbols, or readable background text unless the user explicitly requests text.\nIf text is requested, specify exact wording, placement, timing, typography style, and keep it minimal.\n\nIf the user specifies duration:\nRespect it.\nBreak the prompt into timed beats appropriate to that duration.\nFor 5 seconds, use one or two concise beats.\nFor 10 to 20 seconds, use two to four timed beats.\nFor longer durations, use clear shot progression while maintaining continuity.\n\nIf the user specifies aspect ratio:\nRespect it and adapt framing accordingly.\nUse 16:9 for cinematic widescreen unless otherwise specified.\nUse 9:16 for vertical social video.\nUse 1:1 for square compositions.\n\nIf the user does not specify style:\nDefault to cinematic photorealism with natural motion, professional lens language, realistic lighting, and polished color grading.\n\nAlways output exactly one complete Seedance 2.0 video generation prompt."
},
"_meta": {
"title": "Google Gemini"
}
},
"176": {
"class_type": "SaveVideo",
"inputs": {
"codec": "auto",
"filename_prefix": "video/ComfyUI",
"format": "auto",
"format.codec": "auto",
"video": [
"150",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"177": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"151",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"208": {
"class_type": "LoadImage",
"inputs": {
"image": "fc2f1ab52dcc6cb919feec2edff049ab2712da675f2f1f1f35defb916226d406.jpg"
},
"_meta": {
"title": "Load Image"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)질문과 답변
레퍼런스는 선택 사항입니다. LoadImage는 얼굴, 제품, 환경 등 이미지 기반 정체성/스타일을 제공합니다. 생략하면 ByteDance2ReferenceNode가 프롬프트만으로 생성합니다.
간결하고 지시적인 프롬프트를 작성하세요. 구체적 카메라 동사(돌리 인, 푸시, 크레인 다운, 핸드헬드 스웨이, 스태틱 락오프), 시작-중간-엔딩 비트, 엔딩 프레임 지시(예: 클로즈업에서 고정 종료)를 명확히 기술하세요. GeminiNode가 Seedance 2.0 포맷으로 구조화하지만, 브리프의 명확성이 가장 중요합니다.
ByteDance2ReferenceNode에서 설정하며, 노드 버전과 백엔드 지원에 따라 다릅니다. 종횡비, 길이 등을 거기서 지정하세요. 짧은 길이가 더 선명한 모션과 강한 엔딩 프레임에 유리합니다.
동일한 레퍼런스 이미지를 재사용하고, 주요 정체성/프레이밍 언어를 프롬프트에서 일관되게 유지하세요. 샷 간 스타일 변화를 과도하게 기술하지 말고, 카메라/구도 언어도 비슷하게 맞추세요.
관련 워크플로
- LTX-2.3
Turn an image into a video
- MiniMax H3
Create a video from references
- LTX-2.3
Add camera motion first last frame
- Wan2.2
1.2 스타터 – 이미지에서 비디오로
- Gemini-3.1-flsih-lite, Kling 3.0
3단계: 연락처 시트 워크플로
- 파파트너 노드
Adjustment Frame 워크플로
- 파트너 노드
AI on the Lot 2026 - LTX-2.3 - i2v
- Google Gemini, Kling 3.0, Z-Image-Turbo
AIrt MAchIne(API 버전)
- SSD1.5
AnimateDiff - IPAdapters 루프 애니메이션
- Seedance1.0 Pro, ByteDance
ByteDance: 시작-끝 프레임 비디오
- 파파트너 노드
FastVideo FastH3: Text to Video
- Flux
FLUX 3 비디오: 텍스트 투 비디오