工作流 /
视频工具 /
Pika H3 Flex：文生视频（API）
Pika H3 Flex：文生视频（API）
Pika H3 Flex: Text to Video (API)
通过合作伙伴 API 使用 Pika H3 Flex 从提示词生成带同步音频的 CLIP，无需下载权重。一个节点即可覆盖文生视频、图生视频和首尾帧，只需连接可选的帧输入，支持 480p 至 4K 输出和 5 至 15 秒时长。当本地安装不可用时，非常适合在 MiniMax H3 上进行提示词探索。
视频工具
3 个节点 26 次使用 ComfyUI ≥ 0.44.1 ComfyUI 8 KB
节点构成
PikaH3FlexVideoNode
× 1
SaveVideo
× 1
MarkdownNote
× 1
官方标注的模型
Pika H3 FlexMiniMax H3
工作流 JSON
和上面「下载工作流 JSON」是同一份文件。
1{
2 "id": "a6bb0d53-0000-4000-8000-000000000000",
3 "revision": 0,
4 "last_node_id": 3,
5 "last_link_id": 1,
6 "nodes": [
7 {
8 "id": 1,
9 "type": "PikaH3FlexVideoNode",
10 "pos": [
11 120,
12 160
13 ],
14 "size": [
15 420,
16 320
17 ],
18 "flags": [],
19 "order": 0,
20 "mode": 0,
21 "inputs": [
22 {
23 "name": "first_frame",
24 "shape": 7,
25 "type": "IMAGE",
26 "link": null
27 },
28 {
29 "name": "last_frame",
30 "shape": 7,
31 "type": "IMAGE",
32 "link": null
33 }
34 ],
35 "outputs": [
36 {
37 "name": "VIDEO",
38 "type": "VIDEO",
39 "links": [
40 1
41 ]
42 }
43 ],
44 "properties": {
45 "Node name for S&R": "PikaH3FlexVideoNode"
46 },
47 "widgets_values": [
48 "integrated_multimodal_description: [Shot 1] Live-action, cinematic, a lone kite surfer cuts across a shallow turquoise lagoon at golden hour, the board throwing a long white wake. The camera tracks low alongside the board at fast speed as spray hits the lens. The surfer leans back, sheet in hand, carves toward the camera and then the kite pulls them into a jump.\n\noverall_soundscape: Wind rushes past the microphone while the board slices and slaps across shallow water, followed by a heavy splash on landing.\n\nnon_diegetic_music: A bright electronic pulse at a fast tempo with a rising synth line that cuts out at the splash.",
49 "768p",
50 "16:9",
51 5,
52 false,
53 42,
54 "randomize"
55 ],
56 "widgets_values_named": {
57 "prompt": "integrated_multimodal_description: [Shot 1] Live-action, cinematic, a lone kite surfer cuts across a shallow turquoise lagoon at golden hour, the board throwing a long white wake. The camera tracks low alongside the board at fast speed as spray hits the lens. The surfer leans back, sheet in hand, carves toward the camera and then the kite pulls them into a jump.\n\noverall_soundscape: Wind rushes past the microphone while the board slices and slaps across shallow water, followed by a heavy splash on landing.\n\nnon_diegetic_music: A bright electronic pulse at a fast tempo with a rising synth line that cuts out at the splash.",
58 "resolution": "768p",
59 "aspect_ratio": "16:9",
60 "duration": 5,
61 "rewrite_prompt": false,
62 "seed": 42,
63 "control_after_generate": "randomize"
64 },
65 "color": "#432",
66 "bgcolor": "#653"
67 },
68 {
69 "id": 2,
70 "type": "SaveVideo",
71 "pos": [
72 588.6214113724102,
73 168.48440058224696
74 ],
75 "size": [
76 453.8,
77 296.7
78 ],
79 "flags": [],
80 "order": 2,
81 "mode": 0,
82 "inputs": [
83 {
84 "name": "video",
85 "type": "VIDEO",
86 "link": 1
87 }
88 ],
89 "outputs": [
90 {
91 "name": "video",
92 "type": "VIDEO",
93 "links": null
94 }
95 ],
96 "properties": [],
97 "widgets_values": [
98 "api_pika_h3_flex_t2v",
99 "auto",
100 "auto",
101 "auto"
102 ],
103 "widgets_values_named": {
104 "filename_prefix": "api_pika_h3_flex_t2v",
105 "format": "auto",
106 "format.codec": "auto",
107 "codec": "auto"
108 }
109 },
110 {
111 "id": 3,
112 "type": "MarkdownNote",
113 "pos": [
114 -515.0371749112,
115 155.70708760395473
116 ],
117 "size": [
118 562.0569002470778,
119 958.8749280369451
120 ],
121 "flags": [],
122 "order": 1,
123 "mode": 0,
124 "inputs": [],
125 "outputs": [],
126 "properties": [],
127 "widgets_values": [
128 "## Pika H3 Flex: Text to Video (API)\n\n**Input Assets**\n- none — this template starts from text.\n\n**Pika H3 Flex through the partner API.** The same MiniMax H3 base the local Flex templates use, without downloading weights: write a prompt and get a finished clip with synchronized audio.\n\n---\n\n### What this graph does\n\n1. **Pika H3 Flex Video**: one node takes the prompt and the render settings. `first_frame` and `last_frame` are optional image inputs.\n2. **Save Video**: writes `api_pika_h3_flex_t2v.mp4`.\n\n### Prompt format\n\nUse the MiniMax H3 text mode structure, so picture and sound come from one description:\n\n```text\nintegrated_multimodal_description: [Shot 1] Live-action, cinematic, ... The camera tracks low alongside the board at fast speed ...\n\noverall_soundscape: ...\n\nnon_diegetic_music: ...\n```\n\n- `integrated_multimodal_description` holds the timeline, the camera vocabulary and any dialogue in `<d>[English] ...</d>`.\n- `overall_soundscape` covers wind, impacts and other diegetic sound, `non_diegetic_music` the audience-only score, `N/A` when there is none.\n- Keep every described event inside the requested duration.\n\n### Parameters\n\n| Widget | What it does |\n|---|---|\n| `prompt` | The structured H3 prompt. |\n| `resolution` | `480p` to `4k`; `1080p` and above render at 768p and upscale once. |\n| `aspect_ratio` | `auto` follows the first connected frame, otherwise the prompt. |\n| `duration` | 5 to 15 seconds; 5 s gives 124 frames, about 5.17 s. |\n| `rewrite_prompt` | Let the model expand the prompt. Off keeps seeds reproducible and leaves the structured prompt untouched. |\n| `seed` | Same seed gives similar, not identical, clips. |\n\n### Measured on this graph\n\nThe kite-surfer prompt rendered 1344 x 768, 124 frames, 5.17 s at 24 fps, with generated audio matching the splash.\n\n### Gotchas\n\n- Audio is generated with the picture, so the clip is cut-ready.\n- The same seed at a different resolution gives a different video.\n- Connect only `first_frame` for image to video, both frames for first-last-frame, neither for text to video.\n"
129 ],
130 "widgets_values_named": {
131 "text": "## Pika H3 Flex: Text to Video (API)\n\n**Input Assets**\n- none — this template starts from text.\n\n**Pika H3 Flex through the partner API.** The same MiniMax H3 base the local Flex templates use, without downloading weights: write a prompt and get a finished clip with synchronized audio.\n\n---\n\n### What this graph does\n\n1. **Pika H3 Flex Video**: one node takes the prompt and the render settings. `first_frame` and `last_frame` are optional image inputs.\n2. **Save Video**: writes `api_pika_h3_flex_t2v.mp4`.\n\n### Prompt format\n\nUse the MiniMax H3 text mode structure, so picture and sound come from one description:\n\n```text\nintegrated_multimodal_description: [Shot 1] Live-action, cinematic, ... The camera tracks low alongside the board at fast speed ...\n\noverall_soundscape: ...\n\nnon_diegetic_music: ...\n```\n\n- `integrated_multimodal_description` holds the timeline, the camera vocabulary and any dialogue in `<d>[English] ...</d>`.\n- `overall_soundscape` covers wind, impacts and other diegetic sound, `non_diegetic_music` the audience-only score, `N/A` when there is none.\n- Keep every described event inside the requested duration.\n\n### Parameters\n\n| Widget | What it does |\n|---|---|\n| `prompt` | The structured H3 prompt. |\n| `resolution` | `480p` to `4k`; `1080p` and above render at 768p and upscale once. |\n| `aspect_ratio` | `auto` follows the first connected frame, otherwise the prompt. |\n| `duration` | 5 to 15 seconds; 5 s gives 124 frames, about 5.17 s. |\n| `rewrite_prompt` | Let the model expand the prompt. Off keeps seeds reproducible and leaves the structured prompt untouched. |\n| `seed` | Same seed gives similar, not identical, clips. |\n\n### Measured on this graph\n\nThe kite-surfer prompt rendered 1344 x 768, 124 frames, 5.17 s at 24 fps, with generated audio matching the splash.\n\n### Gotchas\n\n- Audio is generated with the picture, so the clip is cut-ready.\n- The same seed at a different resolution gives a different video.\n- Connect only `first_frame` for image to video, both frames for first-last-frame, neither for text to video.\n"
132 },
133 "color": "#222",
134 "bgcolor": "#000"
135 }
136 ],
137 "links": [
138 [
139 1,
140 1,
141 0,
142 2,
143 0,
144 "VIDEO"
145 ]
146 ],
147 "groups": [],
148 "config": [],
149 "extra": {
150 "ds": {
151 "scale": 0.3479440172798536,
152 "offset": [
153 1885.4968955664744,
154 446.0019445904172
155 ]
156 },
157 "frontendVersion": "1.58.5"
158 },
159 "version": 0.4
160}