工作流 /
音频 /
种子音频 1.0：文本+音频转音频
种子音频 1.0：文本+音频转音频
Seed Audio 1.0: Text + Audio to Audio
上传一个参考音频CLIP并编写文本提示词，将声音克隆到新场景中。使用最多3个参考声音生成对话、背景音乐和音效。
音频
4 个节点 1 个输入 45 次使用 ComfyUI ≥ 0.27.1 ComfyUI 6 KB
节点构成
ByteDanceSeedAudio
× 1
SaveAudioAdvanced
× 1
MarkdownNote
× 1
LoadAudio
× 1
官方标注的模型
Seed Audio 1.0
工作流 JSON
和上面「下载工作流 JSON」是同一份文件。
1{
2 "id": "6d029989-7e4b-4077-acb4-e5c45eec2b1d",
3 "revision": 0,
4 "last_node_id": 8,
5 "last_link_id": 6,
6 "nodes": [
7 {
8 "id": 3,
9 "type": "ByteDanceSeedAudio",
10 "pos": [
11 -243.23859960711826,
12 -386.98716660616896
13 ],
14 "size": [
15 501.90625,
16 572.375
17 ],
18 "flags": [],
19 "order": 2,
20 "mode": 0,
21 "inputs": [
22 {
23 "label": "reference_audio_1",
24 "name": "reference_mode.reference_audio_1",
25 "shape": 7,
26 "type": "AUDIO",
27 "link": 5
28 },
29 {
30 "label": "reference_audio_2",
31 "name": "reference_mode.reference_audio_2",
32 "shape": 7,
33 "type": "AUDIO",
34 "link": null
35 },
36 {
37 "label": "reference_audio_3",
38 "name": "reference_mode.reference_audio_3",
39 "shape": 7,
40 "type": "AUDIO",
41 "link": null
42 }
43 ],
44 "outputs": [
45 {
46 "name": "AUDIO",
47 "type": "AUDIO",
48 "links": [
49 2
50 ]
51 }
52 ],
53 "properties": {
54 "Node name for S&R": "ByteDanceSeedAudio"
55 },
56 "widgets_values": [
57 "Quiet café ambience, soft room tone, faint distant chatter, no background music.\n\nElena (woman in her late twenties, warm mid-range voice, soft but clear, calm and friendly, voiced by @Audio1) says gently:\n\"Hi, I'm Elena. Thanks for joining me today. I'll speak naturally, warmly, and clearly, just like my reference voice.\"\n\nElena (voiced by @Audio1), a little warmer, says:\n\"Listen for my pitch, my pauses, and how I finish each sentence softly but clearly.\"\n\nEnds in silence.",
58 "audio reference",
59 "24000",
60 0,
61 0,
62 0,
63 1266743122,
64 "randomize"
65 ],
66 "color": "#432",
67 "bgcolor": "#653"
68 },
69 {
70 "id": 4,
71 "type": "SaveAudioAdvanced",
72 "pos": [
73 286.54824314203955,
74 -386.8170900771985
75 ],
76 "size": [
77 621.46875,
78 216
79 ],
80 "flags": [],
81 "order": 3,
82 "mode": 0,
83 "inputs": [
84 {
85 "name": "audio",
86 "type": "AUDIO",
87 "link": 2
88 }
89 ],
90 "outputs": [
91 {
92 "name": "audio",
93 "type": "AUDIO",
94 "links": null
95 }
96 ],
97 "properties": [],
98 "widgets_values": [
99 "audio/Seed_Audio_1.0_r2a",
100 "mp3",
101 "V0"
102 ]
103 },
104 {
105 "id": 5,
106 "type": "MarkdownNote",
107 "pos": [
108 -1201.9415604737128,
109 -388.79810918508224
110 ],
111 "size": [
112 476.71875,
113 629.765625
114 ],
115 "flags": [],
116 "order": 0,
117 "mode": 0,
118 "inputs": [],
119 "outputs": [],
120 "title": "Note: TA2A Prompt Guide",
121 "properties": [],
122 "widgets_values": [
123 "Seed Audio 1.0: TA2A (Text + Audio to Audio) Workflow Guide\n\n## What This Workflow Does\nLoad Audio → ByteDance Seed Audio 1.0 (audio reference) → Save Audio.\n\nTA2A = text prompt + reference audio. Clone voice(s) from up to 3 reference clips, then generate a new scene with those voices plus ambience, music, and SFX.\n\nBest for: keeping a consistent character voice, multi-speaker scenes with known voices, dubbing when you already have a voice sample.\n\n## How to Use\n1. Load Audio: add your reference clip(s) (≤30s each).\n2. Connect AUDIO → reference_audio_1 (and _2, _3 if needed, in order, no gaps).\n3. Keep reference_mode on \"audio reference\".\n4. In text_prompt, tag each speaker with voiced by @AudioN matching the connected clip.\n5. Run the workflow.\n\n@Audio1 = reference_audio_1 · @Audio2 = reference_audio_2 · @Audio3 = reference_audio_3\n\nNeed a reference clip first? Use the T2A workflow to generate a clean single-speaker sample, then use it here.\n\n## Target Language\nThe model supports English and Chinese. It may mix languages if you do not specify.\n\nTo lock the output language, add a bracket tag at the top:\n- [Language: English only.]\n- [Language: Chinese only.]\n\nAlso state the language in each character's voice traits, for example: speaks English only, or speaks Chinese only. Write dialogue in your target language.\n\n## Avoid Scene Text Being Spoken\nWrap non-dialogue in [square brackets]. Only quoted text after says / whispers should be spoken.\n\n## Prompt Guide\n1. [Language: ...] optional but recommended\n2. [Environment / BGM / SFX: ...]\n3. Name (voice traits, voiced by @Audio1) says: \"dialogue\"\n4. Name (voiced by @Audio1), [emotion], says: \"next line\"\n5. Name2 (voice traits, voiced by @Audio2) replies: \"dialogue\"\n6. [Outro]\n\nRules:\n- Every @AudioN in the prompt must have a matching connected clip\n- Same speaker reuses the same @AudioN on later lines\n- Max 3 reference clips · ≤30s each · connect 1→2→3 without gaps\n\nLimits: prompt ≤3000 chars · output ≤2 min · English & Chinese\n\n## Example (single reference)\n[Language: English only.]\n[Environment: quiet café ambience, soft room tone, no music.]\n\nElena (warm mid-range voice, calm and friendly, speaks English only, voiced by @Audio1) says gently:\n\"Hi, thanks for joining me today.\"\n\nElena (voiced by @Audio1), a little warmer, says:\n\"I'll speak naturally, just like my reference voice.\"\n\n[Silence.]"
124 ],
125 "color": "#222",
126 "bgcolor": "#000"
127 },
128 {
129 "id": 8,
130 "type": "LoadAudio",
131 "pos": [
132 -687.3165865045286,
133 -386.89512185518817
134 ],
135 "size": [
136 366.625,
137 152
138 ],
139 "flags": [],
140 "order": 1,
141 "mode": 0,
142 "inputs": [],
143 "outputs": [
144 {
145 "name": "AUDIO",
146 "type": "AUDIO",
147 "links": [
148 5
149 ]
150 }
151 ],
152 "properties": {
153 "Node name for S&R": "LoadAudio"
154 },
155 "widgets_values": [
156 "seed_audio_ref_audio1.mp3",
157 null,
158 ""
159 ]
160 }
161 ],
162 "links": [
163 [
164 2,
165 3,
166 0,
167 4,
168 0,
169 "AUDIO"
170 ],
171 [
172 5,
173 8,
174 0,
175 3,
176 0,
177 "AUDIO"
178 ]
179 ],
180 "groups": [],
181 "config": [],
182 "extra": {
183 "ds": {
184 "scale": 0.5404449616503781,
185 "offset": [
186 1786.4337984914434,
187 1032.5114596356975
188 ]
189 },
190 "frontendVersion": "1.45.20"
191 },
192 "version": 0.4
193}