工作流 /
品牌与设计 /
YouTube缩略图生成器
YouTube缩略图生成器
YouTube Thumbnail Generator
使用 Nano Banana Pro 生成高对比度、可点击的 YouTube 缩略图。通过夸张的主体和醒目的色彩策略，打造极具吸引力的戏剧性构图。
品牌与设计
6 个节点 1 个输入 9 次使用 ComfyUI ≥ 0.20.3 enigmatic_e 9 KB
节点构成
MarkdownNote
× 1
SaveImage
× 1
LoadImage
× 1
BatchImagesNode
× 1
GeminiNanoBanana2
× 1
Note
× 1
官方标注的模型
Nano Banana Pro
工作流 JSON
和上面「下载工作流 JSON」是同一份文件。
1{
2 "id": "91dff0a2-fa31-4347-996a-f6955b179370",
3 "revision": 0,
4 "last_node_id": 24,
5 "last_link_id": 20,
6 "nodes": [
7 {
8 "id": 15,
9 "type": "MarkdownNote",
10 "pos": [
11 2090,
12 270
13 ],
14 "size": [
15 220,
16 130
17 ],
18 "flags": [],
19 "order": 0,
20 "mode": 0,
21 "inputs": [],
22 "outputs": [],
23 "title": "Enable multi-image input",
24 "properties": [],
25 "widgets_values": [
26 "Select the `Batch Images` node and press Ctrl-B to enable it. Then, add more load image nodes to provide input images."
27 ],
28 "color": "#222",
29 "bgcolor": "#000"
30 },
31 {
32 "id": 9,
33 "type": "SaveImage",
34 "pos": [
35 2900,
36 490
37 ],
38 "size": [
39 800,
40 520
41 ],
42 "flags": [],
43 "order": 5,
44 "mode": 0,
45 "inputs": [
46 {
47 "name": "images",
48 "type": "IMAGE",
49 "link": 20
50 }
51 ],
52 "outputs": [],
53 "properties": [],
54 "widgets_values": [
55 "ComfyUI"
56 ]
57 },
58 {
59 "id": 16,
60 "type": "LoadImage",
61 "pos": [
62 1700,
63 500
64 ],
65 "size": [
66 250,
67 320
68 ],
69 "flags": [],
70 "order": 1,
71 "mode": 0,
72 "inputs": [],
73 "outputs": [
74 {
75 "name": "IMAGE",
76 "type": "IMAGE",
77 "links": [
78 9
79 ]
80 },
81 {
82 "name": "MASK",
83 "type": "MASK",
84 "links": null
85 }
86 ],
87 "properties": {
88 "Node name for S&R": "LoadImage"
89 },
90 "widgets_values": [
91 "man_with_dog.png",
92 "image"
93 ]
94 },
95 {
96 "id": 8,
97 "type": "BatchImagesNode",
98 "pos": [
99 2090,
100 530
101 ],
102 "size": [
103 200,
104 90
105 ],
106 "flags": [],
107 "order": 3,
108 "mode": 4,
109 "inputs": [
110 {
111 "label": "image0",
112 "name": "images.image0",
113 "type": "IMAGE",
114 "link": 9
115 },
116 {
117 "label": "image1",
118 "name": "images.image1",
119 "type": "IMAGE",
120 "link": null
121 }
122 ],
123 "outputs": [
124 {
125 "name": "IMAGE",
126 "type": "IMAGE",
127 "links": [
128 19
129 ]
130 }
131 ],
132 "properties": {
133 "Node name for S&R": "BatchImagesNode"
134 },
135 "widgets_values": []
136 },
137 {
138 "id": 23,
139 "type": "GeminiNanoBanana2",
140 "pos": [
141 2380,
142 490
143 ],
144 "size": [
145 450,
146 610
147 ],
148 "flags": [],
149 "order": 4,
150 "mode": 0,
151 "showAdvanced": true,
152 "inputs": [
153 {
154 "name": "images",
155 "shape": 7,
156 "type": "IMAGE",
157 "link": 19
158 },
159 {
160 "name": "files",
161 "shape": 7,
162 "type": "GEMINI_INPUT_FILES",
163 "link": null
164 }
165 ],
166 "outputs": [
167 {
168 "name": "IMAGE",
169 "type": "IMAGE",
170 "links": [
171 20
172 ]
173 },
174 {
175 "name": "STRING",
176 "type": "STRING",
177 "links": null
178 },
179 {
180 "name": "thought_image",
181 "type": "IMAGE",
182 "links": null
183 }
184 ],
185 "properties": {
186 "Node name for S&R": "GeminiNanoBanana2"
187 },
188 "widgets_values": [
189 "how to feed dogs",
190 "Nano Banana 2 (Gemini 3.1 Flash Image)",
191 790992566483447,
192 "randomize",
193 "16:9",
194 "2K",
195 "IMAGE",
196 "MINIMAL",
197 "# 🎯 SYSTEM PROMPT — NANO BANANA THUMBNAIL ENGINE\n\nYou are a high-performance YouTube thumbnail generation engine.\n\nYour sole purpose is to transform ANY input into a scroll-stopping, high click-through rate (CTR) thumbnail using Nano Banana Pro.\n\nAlways assume the output is a professional YouTube thumbnail.\n\n---\n\n## ⚠️ CORE RULES (NON-NEGOTIABLE)\n\n- Always interpret input as a thumbnail request\n- Always optimize for maximum CTR (click-through rate)\n- Never produce dull, realistic, or flat imagery\n- Never create cluttered compositions\n- Never rely on small details — prioritize bold, readable shapes\n- Always exaggerate emotion, scale, and contrast\n- Always produce ONE strong, clear visual idea (not multiple concepts)\n\n---\n\n## 🧠 THUMBNAIL DESIGN PRINCIPLES\n\n### Visual Style\n- Hyper-saturated colors pushed beyond realism\n- High contrast with strong separation between subject and background\n- Photorealistic, commercial-grade lighting and polish\n\n### Composition\n- Subject occupies 40–60% of the frame\n- Two-element layout ONLY:\n 1. Primary subject (face or hero object)\n 2. Curiosity hook (secondary visual element)\n- Clear focal point — instantly readable at small size\n- Slight wide-angle or forced perspective for energy\n- Shallow depth of field (sharp subject, blurred background)\n\n### Lighting\n- Strong rim light or side light to separate subject\n- Cinematic contrast (bright highlights + deep shadows)\n\n### Color Strategy\nUse complementary color pairs for maximum pop:\n- Orange / Blue\n- Red / Cyan\n- Yellow / Purple\n\nBackgrounds must be:\n- Clean gradients\n- Soft blur environments\n- OR solid bold colors\n\nNever busy or detailed.\n\n---\n\n## 😱 SUBJECT STRATEGY\n\nPrefer:\n- Human face with exaggerated emotion:\n (shock, excitement, fear, awe, confusion)\n\nOR\n\n- A dramatic object presented at hero scale:\n (oversized, glowing, exploding, impossible physics)\n\nIf no subject is provided:\n→ Default to expressive human reaction\n\n---\n\n## 🎯 VISUAL HOOK STRATEGY\n\nEvery thumbnail must include a “curiosity hook”:\n- Something unusual, exaggerated, or unexpected\n- Creates immediate question in viewer’s mind\n- Examples:\n - Oversized object\n - Impossible scenario\n - Before/after contrast\n - Destruction / chaos moment\n - Transformation freeze-frame\n\n---\n\n## 🔤 TEXT RULES (ONLY IF IMPLIED)\n\n- Use 3–5 words MAX\n- Thick, bold sans-serif font\n- High contrast with outline or drop shadow\n- Never overlap subject’s face\n- Must be readable at 160×90 resolution\n\nIf text is not necessary → do NOT force it\n\n---\n\n## ⚡ INTERPRETATION RULES\n\nIf input is vague or abstract:\n→ Convert it into the most visually dramatic, specific scenario possible\n\nExamples:\n- \"Cooking video\" → chef shocked, flames exploding from pan\n- \"AI tools\" → person overwhelmed by glowing holographic screens\n- \"Finance\" → person reacting to giant money explosion\n\nAlways choose the most clickable version.\n\n---\n\n## 🧩 DEFAULT FALLBACK\n\nIf input is minimal:\n→ expressive face + oversized object + bold gradient background\n\n---\n\n## ✅ OUTPUT GOAL\n\nThe final image MUST:\n- Be instantly readable at small size\n- Have one clear subject and one clear hook\n- Trigger strong emotional reaction\n- Feel impossible to scroll past\n\n---\n\n## 🚀 GENERATION STYLE SUMMARY\n\n\"Photorealistic YouTube thumbnail, ultra high contrast, hyper-saturated colors, expressive subject, dramatic lighting, clean gradient or blurred background, shallow depth of field, bold composition, cinematic rim lighting, wide-angle energy, highly clickable, professional commercial quality\"\n\n---\n\nEND SYSTEM PROMPT"
198 ],
199 "color": "#432",
200 "bgcolor": "#653"
201 },
202 {
203 "id": 24,
204 "type": "Note",
205 "pos": [
206 1710,
207 280
208 ],
209 "size": [
210 340,
211 150
212 ],
213 "flags": [],
214 "order": 2,
215 "mode": 0,
216 "inputs": [],
217 "outputs": [],
218 "properties": [],
219 "widgets_values": [
220 "Add an image as a reference for the main subject or object you want in the thumbnail. You can include multiple reference images, like another thumbnail, to help guide the style. Enable batch image mode and copy/paste the Load Image node into Image 1. This makes it easy to keep adding more reference images as needed."
221 ],
222 "color": "#222",
223 "bgcolor": "#653"
224 }
225 ],
226 "links": [
227 [
228 9,
229 16,
230 0,
231 8,
232 0,
233 "IMAGE"
234 ],
235 [
236 19,
237 8,
238 0,
239 23,
240 0,
241 "IMAGE"
242 ],
243 [
244 20,
245 23,
246 0,
247 9,
248 0,
249 "IMAGE"
250 ]
251 ],
252 "groups": [
253 {
254 "id": 1,
255 "title": "Up to 14",
256 "bounding": [
257 2080,
258 450,
259 220,
260 350
261 ],
262 "color": "#3f789e",
263 "font_size": 24,
264 "flags": []
265 }
266 ],
267 "config": [],
268 "extra": {
269 "ds": {
270 "scale": 0.6415841584158416,
271 "offset": [
272 -1353.3333333333333,
273 80.48611111111109
274 ]
275 },
276 "frontendVersion": "1.43.2",
277 "workflowRendererVersion": "Vue-corrected",
278 "linearData": {
279 "inputs": [
280 [
281 "16",
282 "image"
283 ],
284 [
285 "23",
286 "prompt"
287 ]
288 ],
289 "outputs": [
290 "9"
291 ]
292 },
293 "linearMode": false,
294 "VHS_latentpreview": false,
295 "VHS_latentpreviewrate": 0,
296 "VHS_MetadataImage": true,
297 "VHS_KeepIntermediate": true
298 },
299 "version": 0.4
300}