工作流 /
视频 /
Kling 4.0:参考生视频

Kling 4.0:参考生视频
Kling 4.0: Reference to Video
从两张参考图像中选取一个角色和一个地点,在两者之间拍摄一段三镜头时尚片段:一个带快速变焦的极特写、一个带甩镜的中厅转身,以及一个在拱形窗口下120度环绕。每个剪辑都落在同一房间内的新位置,五秒内包含两个硬切,并为该空间生成音频。
视频
5 个节点 2 个输入 327 次使用 ComfyUI ≥ 0.43.0 ComfyUI 13 KB
节点构成
LoadImage
× 2
Kling4ReferenceToVideoNode
× 1
SaveVideo
× 1
MarkdownNote
× 1
官方标注的模型
Kling 4.0
工作流 JSON
和上面「下载工作流 JSON」是同一份文件。
1{
2 "id": "4b1453ad-6dfe-4760-affa-d9fd5feff4a5",
3 "revision": 0,
4 "last_node_id": 6,
5 "last_link_id": 6,
6 "nodes": [
7 {
8 "id": 2,
9 "type": "LoadImage",
10 "pos": [
11 130.14818245894412,
12 524.4448937007801
13 ],
14 "size": [
15 282.798828125,
16 313.99999999999994
17 ],
18 "flags": [],
19 "order": 0,
20 "mode": 0,
21 "inputs": [],
22 "outputs": [
23 {
24 "name": "IMAGE",
25 "type": "IMAGE",
26 "links": [
27 6
28 ]
29 },
30 {
31 "name": "MASK",
32 "type": "MASK",
33 "links": null
34 }
35 ],
36 "properties": {
37 "Node name for S&R": "LoadImage"
38 },
39 "widgets_values": [
40 "kling4_r2v_brutalist_hall.png",
41 "image"
42 ],
43 "widgets_values_named": {
44 "image": "kling4_r2v_env_brutalist.png",
45 "upload": "image"
46 }
47 },
48 {
49 "id": 3,
50 "type": "Kling4ReferenceToVideoNode",
51 "pos": [
52 458.2975606427351,
53 129.5957324002115
54 ],
55 "size": [
56 400,
57 580
58 ],
59 "flags": [],
60 "order": 3,
61 "mode": 0,
62 "inputs": [
63 {
64 "name": "model.reference_images.image_1",
65 "shape": 7,
66 "type": "IMAGE",
67 "link": 5
68 },
69 {
70 "name": "model.reference_images.image_2",
71 "shape": 7,
72 "type": "IMAGE",
73 "link": 6
74 },
75 {
76 "name": "model.reference_images.image_3",
77 "shape": 7,
78 "type": "IMAGE",
79 "link": null
80 },
81 {
82 "name": "model.reference_videos.video_1",
83 "shape": 7,
84 "type": "VIDEO",
85 "link": null
86 },
87 {
88 "name": "model.reference_videos.video_2",
89 "shape": 7,
90 "type": "VIDEO",
91 "link": null
92 },
93 {
94 "name": "model.first_frame",
95 "shape": 7,
96 "type": "IMAGE",
97 "link": null
98 },
99 {
100 "name": "model.last_frame",
101 "shape": 7,
102 "type": "IMAGE",
103 "link": null
104 }
105 ],
106 "outputs": [
107 {
108 "name": "VIDEO",
109 "type": "VIDEO",
110 "links": [
111 3
112 ]
113 }
114 ],
115 "properties": {
116 "Node name for S&R": "Kling4ReferenceToVideoNode"
117 },
118 "widgets_values": [
119 "kling-4.0",
120 "Three shots, hard cuts on the beat. Fashion-editorial / MV title fragment inside the brutalist hall from @Image2; the woman from @Image1 keeps her orange-red bob, freckles, burgundy lips, camel leather trench, black turtleneck, wide black trousers and boots exactly as they are. Do not change the hall: dark concrete walls, stone bench, arched window, hard golden god-ray and dust in the beam.\n\n0-1.5s: extreme close-up of her face filling the frame, god-ray rim light on her hair and cheek; she looks dead into camera with a small confident almost-smile, then a tiny decisive chin tilt; fast snap zoom-in on her eyes then hard hold.\n1.5-3.2s: hard cut - she is now standing mid-hall inside the golden beam beside the stone bench (new position); medium-wide; she turns on her heel, trench coat flares, and walks two sharp steps toward camera; whip-pan left with her motion then stops.\n3.2-5s: hard cut - centered in the thickest part of the god-ray under the arched window (new position); medium three-quarter; both arms relaxed at her sides, no pockets; camera makes a smooth 120° orbit right around her as she holds a still fashion stare into lens, dust sparkling in the beam, ending locked on a clean beauty frame.\n\nEach shot has its own camera move; never one continuous slow pan. Sound: coat leather rustle, boot heel on stone, soft hall reverb; no dialogue, no subtitles.",
121 "720p",
122 "adaptive",
123 5,
124 42,
125 "randomize",
126 "randomize"
127 ],
128 "widgets_values_named": {
129 "model": "kling-4.0",
130 "model.prompt": "Three shots, hard cuts on the beat. Fashion-editorial / MV title fragment inside the brutalist hall from @Image2; the woman from @Image1 keeps her orange-red bob, freckles, burgundy lips, camel leather trench, black turtleneck, wide black trousers and boots exactly as they are. Do not change the hall: dark concrete walls, stone bench, arched window, hard golden god-ray and dust in the beam.\n\n0-1.5s: extreme close-up of her face filling the frame, god-ray rim light on her hair and cheek; she looks dead into camera with a small confident almost-smile, then a tiny decisive chin tilt; fast snap zoom-in on her eyes then hard hold.\n1.5-3.2s: hard cut - she is now standing mid-hall inside the golden beam beside the stone bench (new position); medium-wide; she turns on her heel, trench coat flares, and walks two sharp steps toward camera; whip-pan left with her motion then stops.\n3.2-5s: hard cut - centered in the thickest part of the god-ray under the arched window (new position); medium three-quarter; both arms relaxed at her sides, no pockets; camera makes a smooth 120° orbit right around her as she holds a still fashion stare into lens, dust sparkling in the beam, ending locked on a clean beauty frame.\n\nEach shot has its own camera move; never one continuous slow pan. Sound: coat leather rustle, boot heel on stone, soft hall reverb; no dialogue, no subtitles.",
131 "model.resolution": "720p",
132 "model.aspect_ratio": "adaptive",
133 "model.duration": 5,
134 "model.seed": 42,
135 "control_after_generate#1": "randomize",
136 "control_after_generate": "randomize"
137 },
138 "color": "#432",
139 "bgcolor": "#653"
140 },
141 {
142 "id": 4,
143 "type": "SaveVideo",
144 "pos": [
145 894.4100252007463,
146 131.18629344855975
147 ],
148 "size": [
149 660.735211995881,
150 594.8164837417409
151 ],
152 "flags": [],
153 "order": 4,
154 "mode": 0,
155 "inputs": [
156 {
157 "name": "video",
158 "type": "VIDEO",
159 "link": 3
160 }
161 ],
162 "outputs": [
163 {
164 "name": "video",
165 "type": "VIDEO",
166 "links": null
167 }
168 ],
169 "properties": [],
170 "widgets_values": [
171 "video/kling4_r2v",
172 "mp4",
173 "auto",
174 "auto"
175 ],
176 "widgets_values_named": {
177 "filename_prefix": "video/kling4.0_r2v",
178 "format": "mp4",
179 "format.codec": "auto",
180 "codec": "auto"
181 }
182 },
183 {
184 "id": 5,
185 "type": "LoadImage",
186 "pos": [
187 120.99635063305095,
188 128.52833996828463
189 ],
190 "size": [
191 282.798828125,
192 314
193 ],
194 "flags": [],
195 "order": 1,
196 "mode": 0,
197 "inputs": [],
198 "outputs": [
199 {
200 "name": "IMAGE",
201 "type": "IMAGE",
202 "links": [
203 5
204 ]
205 },
206 {
207 "name": "MASK",
208 "type": "MASK",
209 "links": []
210 }
211 ],
212 "properties": {
213 "Node name for S&R": "LoadImage"
214 },
215 "widgets_values": [
216 "kling4_r2v_woman_copper_bob.png",
217 "image"
218 ],
219 "widgets_values_named": {
220 "image": "kling4_r2v_char.png",
221 "upload": "image"
222 }
223 },
224 {
225 "id": 6,
226 "type": "MarkdownNote",
227 "pos": [
228 -321.20891403611324,
229 131.25625019065765
230 ],
231 "size": [
232 395.6800529886141,
233 626.5317167071183
234 ],
235 "flags": [],
236 "order": 2,
237 "mode": 0,
238 "inputs": [],
239 "outputs": [],
240 "properties": [],
241 "widgets_values": [
242 "**Input Assets**\n\n- [kling4_r2v_woman_copper_bob.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/input/kling4_r2v_woman_copper_bob.png) — character reference: face, hair, wardrobe\n- [kling4_r2v_brutalist_hall.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/input/kling4_r2v_brutalist_hall.png) — location reference: concrete hall, arched window, god-ray\n\n## Two references, two jobs\n\n`@Image1` owns the person: copper-red bob, freckles, burgundy lips, camel leather trench, black turtleneck, wide black trousers, boots — each named, each pinned with `keep … exactly as they are`. `@Image2` owns the room: dark concrete walls, stone bench, arched window, the hard golden beam and the dust in it, plus `do not change the hall` so the model does not redecorate the location it was given.\n\nA reference pins an identity and a place; it does not pin blocking. That is why the prompt asks for the positions in words — close-up in place, then `standing mid-hall beside the stone bench (new position)`, then `centred under the arched window (new position)`. Without that sentence the model keeps her standing exactly where the reference image has her and the cut lands on the same framing twice.\n\n## Three shots in five seconds\n\n- **0–1.5 s** — extreme close-up of her face filling the frame, rim light from the beam on hair and cheek, a small confident almost-smile, a tiny chin tilt, then a snap zoom into the eyes and a hard hold.\n- **1.5–3.2 s** — hard cut to mid-hall inside the beam: she turns on her heel, the trench flares, two sharp steps toward camera, and a whip-pan that travels with her motion and stops.\n- **3.2–5 s** — hard cut under the arched window: medium three-quarter, arms relaxed, a smooth 120° orbit right around her, dust sparkling in the beam, ending on a clean beauty frame.\n\nEvery shot owns its camera move — the shot list says `never one continuous slow pan` explicitly, because left alone the model averages three moves into one drift. Keep the shots few and give each one a different idea; 720p, five seconds, seed 42.\n\n## Report Issue\n\nNote: Please update ComfyUI first ([guide](https://docs.comfy.org/installation/update_comfyui)) and prepare required models. Desktop/Cloud updates follow stable releases, so some nightly-supported models may not be available yet.\n\n- Cannot run / runtime errors: [ComfyUI/issues](https://github.com/comfyanonymous/ComfyUI/issues)\n- UI / frontend issues: [ComfyUI_Frontend/issues](https://github.com/Comfy-Org/ComfyUI_Frontend/issues)\n- Workflow issues: [workflow_templates/issues](https://github.com/Comfy-Org/workflow_templates/issues)"
243 ],
244 "widgets_values_named": {
245 "text": "**Input Assets**\n\n- [kling4_r2v_woman_copper_bob.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/input/kling4_r2v_woman_copper_bob.png) — character reference: face, hair, wardrobe\n- [kling4_r2v_brutalist_hall.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/input/kling4_r2v_brutalist_hall.png) — location reference: concrete hall, arched window, god-ray\n\n## Two references, two jobs\n\n`@Image1` owns the person: copper-red bob, freckles, burgundy lips, camel leather trench, black turtleneck, wide black trousers, boots — each named, each pinned with `keep … exactly as they are`. `@Image2` owns the room: dark concrete walls, stone bench, arched window, the hard golden beam and the dust in it, plus `do not change the hall` so the model does not redecorate the location it was given.\n\nA reference pins an identity and a place; it does not pin blocking. That is why the prompt asks for the positions in words — close-up in place, then `standing mid-hall beside the stone bench (new position)`, then `centred under the arched window (new position)`. Without that sentence the model keeps her standing exactly where the reference image has her and the cut lands on the same framing twice.\n\n## Three shots in five seconds\n\n- **0–1.5 s** — extreme close-up of her face filling the frame, rim light from the beam on hair and cheek, a small confident almost-smile, a tiny chin tilt, then a snap zoom into the eyes and a hard hold.\n- **1.5–3.2 s** — hard cut to mid-hall inside the beam: she turns on her heel, the trench flares, two sharp steps toward camera, and a whip-pan that travels with her motion and stops.\n- **3.2–5 s** — hard cut under the arched window: medium three-quarter, arms relaxed, a smooth 120° orbit right around her, dust sparkling in the beam, ending on a clean beauty frame.\n\nEvery shot owns its camera move — the shot list says `never one continuous slow pan` explicitly, because left alone the model averages three moves into one drift. Keep the shots few and give each one a different idea; 720p, five seconds, seed 42.\n\n## Report Issue\n\nNote: Please update ComfyUI first ([guide](https://docs.comfy.org/installation/update_comfyui)) and prepare required models. Desktop/Cloud updates follow stable releases, so some nightly-supported models may not be available yet.\n\n- Cannot run / runtime errors: [ComfyUI/issues](https://github.com/comfyanonymous/ComfyUI/issues)\n- UI / frontend issues: [ComfyUI_Frontend/issues](https://github.com/Comfy-Org/ComfyUI_Frontend/issues)\n- Workflow issues: [workflow_templates/issues](https://github.com/Comfy-Org/workflow_templates/issues)"
246 },
247 "color": "#222",
248 "bgcolor": "#000",
249 "title": "Note: Two References, Three Shots"
250 }
251 ],
252 "links": [
253 [
254 3,
255 3,
256 0,
257 4,
258 0,
259 "VIDEO"
260 ],
261 [
262 5,
263 5,
264 0,
265 3,
266 0,
267 "IMAGE"
268 ],
269 [
270 6,
271 2,
272 0,
273 3,
274 1,
275 "IMAGE"
276 ]
277 ],
278 "groups": [],
279 "config": [],
280 "extra": {
281 "ds": {
282 "scale": 0.380954549593041,
283 "offset": [
284 2074.1824010044447,
285 490.2012743759599
286 ]
287 },
288 "frontendVersion": "1.57.1"
289 },
290 "version": 0.4
291}