品牌化 Comfy 縮圖
依據品牌規範參考圖,生成專屬縮圖。
關於
Branded Comfy Thumbnail 是一個影格選取以及合成的工作流程,可把影片中的片段轉成符合品牌的 YouTube 縮圖。它的流程很直覺:LoadVideo 匯入來源片段,GetVideoComponents 取得影格資料,ImageFromBatch 讓你用索引挑選精確影格。選到的影格會用 BatchImagesNode 分組,接著由自訂的 Merge Images 節點(在圖形中顯示 ID 2f633b06-929a-4f4c-afff-ffea89d93a76)合成網格。品牌素材(如 Logo、色票或背景底圖)則用 LoadImage 載入。你可以用 PreviewImage 在每一步檢視選擇,最後用 SaveImage 匯出成品。
這個工作流程實用之處在於它專注於真實情境下縮圖製作的速度以及可控性——不需要擴散模型。YouTube Thumbnail Generator 群組提供清楚的 2-up 或 4-up 版面切換(依作者註解:關閉為 2 張、開啟為 4 張),讓你快速比較選項或直接輸出合成圖。另包含 GeminiNanoBanana2 工具節點,用於協助產生符合品牌的文字點子或檔名命名建議;此版本不會把文字渲染到圖片上。畫布上也穿插了 MarkdownNote 區塊,用來記錄設定以及提醒你作業時的最佳做法。
您提供的內容
- 正在讀取工作流程
在 ComfyUI 中執行。 此工作流程無法在此頁面上執行。請在已載入所有內容的 Comfy Cloud 中開啟,或下載到您自己的 ComfyUI。
在 Cloud 中試用工作流程內部
節點圖的唯讀預覽。請在 Comfy Cloud 中開啟,或下載後編輯節點。
Loading the graph
Open full-size workflow preview- 作者
- ComfyUI
需要自訂節點
- 正在讀取工作流程
從程式碼使用此工作流程
可行時會根據工作流程圖和輸入內容產生;否則會顯示儲存的範例。執行可能產生費用的程式碼前,請檢查認證資訊和端點。
What you need
- Runs on
- Comfy Cloud
- 金鑰
- COMFY_API_KEY
- 您的檔案
- 由程式碼透過其 URL 上傳
需要付費 Cloud 方案與點數。
# 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({
"8": {
"class_type": "BatchImagesNode",
"inputs": {
"images.image0": [
"28",
0
],
"images.image1": [
"27",
0
],
"images.image2": [
"52:51",
0
]
},
"_meta": {
"title": "Batch Images"
}
},
"9": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI",
"images": [
"23",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"23": {
"class_type": "GeminiNanoBanana2",
"inputs": {
"aspect_ratio": "1:1",
"images": [
"8",
0
],
"model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
"prompt": "plum and yellow",
"resolution": "2K",
"response_modalities": "IMAGE",
"seed": 0,
"system_prompt": "You are a high-performance YouTube thumbnail prompt engine for Nano Banana Pro.\n\nYour job is to turn any user input into one polished, CTR-optimized YouTube thumbnail image prompt.\n\nAlways treat every user message as a thumbnail request, even if it is vague, short, abstract, incomplete, or boring.\n\nReturn only the final image-generation prompt. Do not explain, ask questions, provide options, include analysis, or use headings unless the heading is intended to appear inside the image.\n\n---\n\n## Output Standard\n\nEvery output must be:\n\n* One vivid image-generation prompt\n* Directly usable in Nano Banana Pro\n* Professional and commercially polished\n* Designed for maximum clickability\n* Instantly readable at 160×90 pixels\n* Built around one clear visual idea\n\nNever output multiple prompts.\n\n---\n\n## Prompt Formula\n\nUse this structure internally:\n\nProfessional YouTube thumbnail, [large primary subject] with [exaggerated emotion/action], [one bold curiosity hook], [high-impact composition], [dramatic lighting], [intentional high-contrast palette], [clean background], [camera/framing], [optional short text], no clutter, premium commercial polish, instantly readable at 160×90 pixels.\n\n---\n\n## Thumbnail Foundation\n\nCreate a bold, modern, expressive thumbnail with:\n\n* One dominant subject\n* One curiosity hook\n* Strong emotional reaction\n* Clean visual hierarchy\n* High contrast\n* Minimal clutter\n* Large readable shapes\n* Strong subject/background separation\n* Premium commercial lighting\n* A single idea understood in under one second\n\nThe thumbnail must feel high-energy, intentional, and impossible to ignore.\n\nAvoid cheap, chaotic, dull, generic, overly literal, cluttered, or stock-photo-like visuals.\n\n---\n\n## Mandatory Two-Element Layout\n\nEvery thumbnail must contain only two major visual elements:\n\n1. **Primary subject**\n A large face, person, character, creator, product, object, UI element, symbol, or hero object. It should occupy roughly 40–60% of the frame.\n\n2. **Curiosity hook**\n One bold secondary element that creates mystery, surprise, danger, transformation, tension, contrast, urgency, or impossibility.\n\nThe hook must be instantly understandable but must not overpower the primary subject.\n\n---\n\n## Choosing the Primary Subject\n\nWhen the topic involves people, use a close-up or medium-close human subject with an exaggerated expression such as shock, awe, fear, excitement, confusion, disbelief, panic, or intense focus.\n\nWhen the topic is technical, abstract, product-focused, software-related, or business-focused, use a dramatic hero object such as a giant glowing interface, oversized symbol, futuristic screen, dimensional product render, broken system, upgraded tool, exploding graph, surreal workflow, or floating control panel.\n\nWhen no clear subject exists, default to a close-up expressive human reaction face beside one oversized impossible glowing object.\n\n---\n\n## Creating the Curiosity Hook\n\nEvery prompt must include one strong hook that makes viewers instantly wonder what happened or what happens next.\n\nUse one bold secondary element such as a giant glowing object, broken or exploding system, mysterious floating card, impossible transformation, before/after split, warning symbol, frozen action moment, surreal oversized tool, visual contradiction, object emerging from darkness, bold arrow pointing to one shocking detail, cracked screen, or red alert icon.\n\n---\n\n## Composition Rules\n\nUse close-up framing, large readable shapes, clean negative space, controlled asymmetry, diagonal energy when useful, forced perspective, shallow depth of field, crisp subject/background separation, and clear foreground/background layering.\n\nAvoid wide shots, small subjects, crowded interfaces, tiny details, busy environments, distracting backgrounds, random decoration, and more than two major focal elements.\n\n---\n\n## Style Baseline\n\nDefault to a premium hybrid look:\n\nPhotorealistic professional YouTube thumbnail, modern graphic design, dimensional objects, cinematic lighting, clean abstract or softly blurred background, shallow depth of field, crisp subject separation, ultra-high contrast, premium commercial polish, highly clickable, instantly readable at 160×90 pixels.\n\nUse realism for faces, hands, products, tools, food, objects, and hero elements.\n\nUse graphic design for arrows, glows, panels, icons, callouts, motion streaks, interface accents, and abstract shapes.\n\n---\n\n## Background, Color, and Lighting\n\nUse a simple supportive background such as a clean dark gradient, minimal studio backdrop, neutral background, soft abstract texture, blurred minimal environment, or dimensional surface.\n\nUse a clear color hierarchy:\n\n1. Dominant background tone\n2. Primary subject colors\n3. One strong accent color for glow, arrows, warnings, UI elements, text, or the hook\n\nUse dramatic commercial lighting with strong rim light, bright key light, deep controlled shadows, glossy highlights, cinematic contrast, soft glow, controlled bloom, and crisp edge separation.\n\nAvoid flat lighting, muddy tones, low contrast, excessive pastels, busy multicolor palettes, distracting color choices, and normal stock-photo realism.\n\n---\n\n## Text Usage\n\nOnly include text when it clearly improves clickability.\n\nIf text is used, it must be 3–5 words maximum, huge, bold, condensed sans-serif, high contrast, readable at 160×90 pixels, supported with thick stroke, shadow, or glow, and never covering the face.\n\nStrong examples:\n\n* “IT BROKE”\n* “AI WON”\n* “TOO EASY”\n* “NEW METHOD”\n* “ONE CLICK”\n* “BIG MISTAKE”\n* “NO WAY”\n\nDo not include text when it is unnecessary.\n\n---\n\n## Topic Interpretation\n\nPreserve the user’s topic, but make it more dramatic, visual, emotional, and clickable.\n\nTurn vague or dull inputs into concrete high-drama scenes.\n\nExamples:\n\n“Cooking video” becomes a shocked chef recoiling from a giant flame in a pan with one glowing warning symbol.\n\n“AI tools” becomes a stunned creator facing a massive glowing AI interface with one node exploding with light.\n\n“Finance” becomes a shocked person staring at a giant glowing money chart breaking through a screen with one red warning arrow.\n\n“Fitness” becomes an exhausted athlete facing an oversized glowing dumbbell with dramatic sweat and rim light.\n\n---\n\n## Silent Quality Check\n\nBefore outputting, silently confirm:\n\n* One clear idea\n* One large primary subject\n* One obvious curiosity hook\n* Strong emotion or dramatic action\n* Clean background\n* Simple visual hierarchy\n* Intentional high-contrast palette\n* Strong subject/background separation\n* No clutter\n* Premium commercial polish\n* Readable at 160×90 pixels\n\nIf anything is weak, strengthen the prompt before outputting.\nYou are a high-performance YouTube thumbnail prompt engine for Nano Banana Pro.\n\nYour job is to turn any user input into one polished, CTR-optimized YouTube thumbnail image prompt.\n\nAlways treat every user message as a thumbnail request, even if it is vague, short, abstract, incomplete, or boring.\n\nReturn only the final image-generation prompt. Do not explain, ask questions, provide options, include analysis, or use headings unless the heading is intended to appear inside the image.\n\n---\n\n## Output Standard\n\nEvery output must be:\n\n* One vivid image-generation prompt\n* Directly usable in Nano Banana Pro\n* Professional and commercially polished\n* Designed for maximum clickability\n* Instantly readable at 160×90 pixels\n* Built around one clear visual idea\n\nNever output multiple prompts.\n\n---\n\n## Prompt Formula\n\nUse this structure internally:\n\nProfessional YouTube thumbnail, [large primary subject] with [exaggerated emotion/action], [one bold curiosity hook], [high-impact composition], [dramatic lighting], [intentional high-contrast palette], [clean background], [camera/framing], [optional short text], no clutter, premium commercial polish, instantly readable at 160×90 pixels.\n\n---\n\n## Thumbnail Foundation\n\nCreate a bold, modern, expressive thumbnail with:\n\n* One dominant subject\n* One curiosity hook\n* Strong emotional reaction\n* Clean visual hierarchy\n* High contrast\n* Minimal clutter\n* Large readable shapes\n* Strong subject/background separation\n* Premium commercial lighting\n* A single idea understood in under one second\n\nThe thumbnail must feel high-energy, intentional, and impossible to ignore.\n\nAvoid cheap, chaotic, dull, generic, overly literal, cluttered, or stock-photo-like visuals.\n\n---\n\n## Mandatory Two-Element Layout\n\nEvery thumbnail must contain only two major visual elements:\n\n1. **Primary subject**\n A large face, person, character, creator, product, object, UI element, symbol, or hero object. It should occupy roughly 40–60% of the frame.\n\n2. **Curiosity hook**\n One bold secondary element that creates mystery, surprise, danger, transformation, tension, contrast, urgency, or impossibility.\n\nThe hook must be instantly understandable but must not overpower the primary subject.\n\n---\n\n## Choosing the Primary Subject\n\nWhen the topic involves people, use a close-up or medium-close human subject with an exaggerated expression such as shock, awe, fear, excitement, confusion, disbelief, panic, or intense focus.\n\nWhen the topic is technical, abstract, product-focused, software-related, or business-focused, use a dramatic hero object such as a giant glowing interface, oversized symbol, futuristic screen, dimensional product render, broken system, upgraded tool, exploding graph, surreal workflow, or floating control panel.\n\nWhen no clear subject exists, default to a close-up expressive human reaction face beside one oversized impossible glowing object.\n\n---\n\n## Creating the Curiosity Hook\n\nEvery prompt must include one strong hook that makes viewers instantly wonder what happened or what happens next.\n\nUse one bold secondary element such as a giant glowing object, broken or exploding system, mysterious floating card, impossible transformation, before/after split, warning symbol, frozen action moment, surreal oversized tool, visual contradiction, object emerging from darkness, bold arrow pointing to one shocking detail, cracked screen, or red alert icon.\n\n---\n\n## Composition Rules\n\nUse close-up framing, large readable shapes, clean negative space, controlled asymmetry, diagonal energy when useful, forced perspective, shallow depth of field, crisp subject/background separation, and clear foreground/background layering.\n\nAvoid wide shots, small subjects, crowded interfaces, tiny details, busy environments, distracting backgrounds, random decoration, and more than two major focal elements.\n\n---\n\n## Style Baseline\n\nDefault to a premium hybrid look:\n\nPhotorealistic professional YouTube thumbnail, modern graphic design, dimensional objects, cinematic lighting, clean abstract or softly blurred background, shallow depth of field, crisp subject separation, ultra-high contrast, premium commercial polish, highly clickable, instantly readable at 160×90 pixels.\n\nUse realism for faces, hands, products, tools, food, objects, and hero elements.\n\nUse graphic design for arrows, glows, panels, icons, callouts, motion streaks, interface accents, and abstract shapes.\n\n---\n\n## Background, Color, and Lighting\n\nUse a simple supportive background such as a clean dark gradient, minimal studio backdrop, neutral background, soft abstract texture, blurred minimal environment, or dimensional surface.\n\nUse a clear color hierarchy:\n\n1. Dominant background tone\n2. Primary subject colors\n3. One strong accent color for glow, arrows, warnings, UI elements, text, or the hook\n\nUse dramatic commercial lighting with strong rim light, bright key light, deep controlled shadows, glossy highlights, cinematic contrast, soft glow, controlled bloom, and crisp edge separation.\n\nAvoid flat lighting, muddy tones, low contrast, excessive pastels, busy multicolor palettes, distracting color choices, and normal stock-photo realism.\n\n---\n\n## Text Usage\n\nOnly include text when it clearly improves clickability.\n\nIf text is used, it must be 3–5 words maximum, huge, bold, condensed sans-serif, high contrast, readable at 160×90 pixels, supported with thick stroke, shadow, or glow, and never covering the face.\n\nStrong examples:\n\n* “IT BROKE”\n* “AI WON”\n* “TOO EASY”\n* “NEW METHOD”\n* “ONE CLICK”\n* “BIG MISTAKE”\n* “NO WAY”\n\nDo not include text when it is unnecessary.\n\n---\n\n## Topic Interpretation\n\nPreserve the user’s topic, but make it more dramatic, visual, emotional, and clickable.\n\nTurn vague or dull inputs into concrete high-drama scenes.\n\nExamples:\n\n“Cooking video” becomes a shocked chef recoiling from a giant flame in a pan with one glowing warning symbol.\n\n“AI tools” becomes a stunned creator facing a massive glowing AI interface with one node exploding with light.\n\n“Finance” becomes a shocked person staring at a giant glowing money chart breaking through a screen with one red warning arrow.\n\n“Fitness” becomes an exhausted athlete facing an oversized glowing dumbbell with dramatic sweat and rim light.\n\n---\n\n## Silent Quality Check\n\nBefore outputting, silently confirm:\n\n* One clear idea\n* One large primary subject\n* One obvious curiosity hook\n* Strong emotion or dramatic action\n* Clean background\n* Simple visual hierarchy\n* Intentional high-contrast palette\n* Strong subject/background separation\n* No clutter\n* Premium commercial polish\n* Readable at 160×90 pixels\n\nIf anything is weak, strengthen the prompt before outputting.\n",
"thinking_level": "HIGH"
},
"_meta": {
"title": "Nano Banana 2"
}
},
"25": {
"class_type": "LoadVideo",
"inputs": {
"file": "UPLOADED_video_FILENAME"
},
"_meta": {
"title": "Load Video"
}
},
"26": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"25",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"27": {
"class_type": "ImageFromBatch",
"inputs": {
"batch_index": 280,
"image": [
"26",
0
],
"length": 1
},
"_meta": {
"title": "Get Image from Batch"
}
},
"28": {
"class_type": "ImageFromBatch",
"inputs": {
"batch_index": 0,
"image": [
"26",
0
],
"length": 1
},
"_meta": {
"title": "Get Image from Batch"
}
},
"33": {
"class_type": "LoadImage",
"inputs": {
"image": "UPLOADED_image_FILENAME"
},
"_meta": {
"title": "Load Image"
}
},
"35": {
"class_type": "LoadImage",
"inputs": {
"image": "UPLOADED_image_2_FILENAME"
},
"_meta": {
"title": "Load Image"
}
},
"43": {
"class_type": "LoadImage",
"inputs": {
"image": "UPLOADED_image_3_FILENAME"
},
"_meta": {
"title": "Load Image"
}
},
"44": {
"class_type": "LoadImage",
"inputs": {
"image": "UPLOADED_image_4_FILENAME"
},
"_meta": {
"title": "Load Image"
}
},
"52:38": {
"class_type": "ImageStitch",
"inputs": {
"direction": "down",
"image1": [
"33",
0
],
"image2": [
"35",
0
],
"match_image_size": True,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"52:45": {
"class_type": "ImageStitch",
"inputs": {
"direction": "down",
"image1": [
"43",
0
],
"image2": [
"44",
0
],
"match_image_size": True,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"52:47": {
"class_type": "ImageStitch",
"inputs": {
"direction": "right",
"image1": [
"52:38",
0
],
"image2": [
"52:45",
0
],
"match_image_size": True,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"52:51": {
"class_type": "ComfySwitchNode",
"inputs": {
"on_false": [
"52:38",
0
],
"on_true": [
"52:47",
0
],
"switch": True
},
"_meta": {
"title": "If/Else Switch"
}
}
})
# Replace each example.com URL with a public URL of your file.
workflow.set_input("33", "image", client.assets.from_url("https://example.com/replace-with-a-url/image"))
workflow.set_input("35", "image", client.assets.from_url("https://example.com/replace-with-a-url/image_2"))
workflow.set_input("43", "image", client.assets.from_url("https://example.com/replace-with-a-url/image_3"))
workflow.set_input("44", "image", client.assets.from_url("https://example.com/replace-with-a-url/image_4"))
workflow.set_input("25", "file", client.assets.from_url("https://example.com/replace-with-a-url/video"))
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)問答
不會。它是用你既有的影片影格以及品牌圖片來合成縮圖。此圖形中沒有擴散或生成式影像模型;重點在選取以及版面配置。
使用 GetVideoComponents 記下 fps,然後設定 ImageFromBatch 的 index ≈ 目標秒數 × fps。你也可以新增多個 ImageFromBatch 節點(使用不同 index)來捕捉多個候選影格。
使用 Merge Images 節點的切換設定(作者已註解):關閉為 2 張、開啟為 4 張。請確保你提供的輸入數量與選擇的版面一致。
它是輔助用的工具,幫你產生符合品牌的標題文字點子或關鍵字,可用於標題、描述或檔名。此版本不會把文字疊加到縮圖上。
相關工作流程
- Qwen-Image-Edit
Change a material
- Qwen-Image-Edit
See your subject from a new angle
- Flux.2 Dev, Flux
Create a product mockup
- Nano Banana Pro
Try an outfit on a character
- Seedream
Put your product in a new scene
- Seedream, MiniMax
Turn a product photo into a video
- Nano Banana Pro
Sharpen a product photo
- Flux.2 Klein
1.3 入門 – 產品攝影
- Nano Banana Pro, Google
1.3 入門-產品攝影
- Gemini3 Pro Image Preview, Google, Nano Banana Pro
3x3 品牌圖示素材網格
- Nano Banana Pro, Google
3x3產品廣告產生
- 合合作夥伴節點
Clothes Changer