ComfyUI Vergora Seedance 2.5
ComfyUI node for BytePlus Seedance 2.5 video generation through Vergora
ComfyUI Vergora Seedance 2.5
ComfyUI custom node for Seedance 2.5 through the Vergora API. This package
exports only VergoraSeedance25; Seedance 2.0 / Fast / Mini are installed from a
separate custom node folder so ComfyUI never mixes incompatible parameter sets.
Demo / 演示
Click the cover to open the generated Seedance 2.5 demo in the GitHub Release. 点击封面可打开由 Seedance 2.5 生成的演示视频。
Features
- Text-to-video
- First-frame and first/last-frame generation
- Up to 30 reference images, 10 videos, and 10 audios
omni_reference_task_type:auto,reference,edit,extend- MOV request parameter and latency-free MP4 fallback naming
- Seedance 2.5 Draft stage and final-from-Draft workflow
- Flat gateway fields and native BytePlus
content[] - Asynchronous polling, interruption, verified download, and ComfyUI video preview
- API key stored only in
config.iniorVERGORA_API_KEY
Model
| Vergora Model ID | Resolution | Duration | References |
| --- | --- | --- | --- |
| seedance-2.5 | 480p, 720p, 1080p | 4-30s or -1 | 30 images + 10 videos + 10 audios |
Inputs
| Input | Notes |
| --- | --- |
| prompt | Refer to assets as Image 1, Video 1, Audio 1 |
| model | Always seedance-2.5 |
| resolution | 480p, 720p, or 1080p |
| ratio | adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 |
| duration | -1, or 4-30 seconds |
| generate_audio | Enable generated audio |
| watermark | Add the service watermark |
| return_last_frame | Download the last frame when returned |
| output_format | mp4 or mov |
| omni_reference_task_type | auto, reference, edit, or extend |
| draft | Draft stage; requires 480p |
| draft_task_id | Generate the final 1080p video from a Draft task |
| first_frame / last_frame | Image tensors, or use the URL inputs |
| reference_images | ComfyUI image batch |
| reference_image_urls | One image URL/data URI per line |
| reference_video_urls | One video URL per line |
| reference_audio_urls | One audio URL per line |
| extra_json | Extra request fields, merged last |
Validation
- First-frame and first-last-frame tasks require
ratio=adaptive. editrequires a reference video,ratio=adaptive, andduration=-1.extendrequires a reference video andratio=adaptive.- Draft generation requires
480p. - Final-from-Draft requires
1080pand no prompt or media fields. - Keyframe mode and reference mode cannot be mixed.
- A last frame requires a first frame.
Current Vergora gateway compatibility
Live-tested on 2026-11-14 through POST/GET /v1/videos:
| Capability | Current status |
| --- | --- |
| Text and reference-image generation | Working |
| MOV output | Gateway stores output_format=mp4 |
| Draft | Gateway stores draft=false; final pass is rejected |
| First/last frame | Blocked by adaptive image-input token estimation |
| Edit/extend | Blocked by adaptive video-input token estimation |
The node detects these gateway mismatches. A Draft request that comes back with
draft=false is stopped instead of being saved as an ordinary 480p video.
Requested MOV output is saved using the actual returned extension with a warning.
Installation
- Put this folder in ComfyUI's
custom_nodesdirectory asComfyUI-Vergora-Seedance-2.5. - Copy
config.ini.exampletoconfig.ini. - Set
KEYto a Vergora API key. - Restart ComfyUI.
Keep the two adapters in separate folders:
custom_nodes/
ComfyUI-Vergora-Seedance-2.0/
ComfyUI-Vergora-Seedance-2.5/
Environment variables override config.ini:
VERGORA_API_KEY
VERGORA_BASE_URL
VERGORA_REQUEST_FORMAT
Request format
content[] is the default and uses the official BytePlus request shape.
REQUEST_FORMAT=auto also tries content[] first and retries once with flat
fields only when the gateway explicitly rejects the content[] schema. Other
submit errors are never retried, avoiding duplicate tasks and billing.
Development
python test_offline.py
The offline suite uses fake HTTP functions and creates no billed tasks.
License
MIT. See LICENSE.
中文说明
这是独立的 Seedance 2.5 ComfyUI 适配器,只注册 VergoraSeedance25。
2.0 / Fast / Mini 使用另一个独立目录,ComfyUI 不会在一个节点里混用两套模型参数。
2.5 支持 4-30 秒、30 图 + 10 视频 + 10 音频、MOV 参数、参考任务类型和 Draft 两阶段流程。首帧/首尾帧、Edit、Extend、Draft 的本地参数检查会在提交前完成。
当前网关实测:
- 2.5 文生和参考图:可用
- MOV:网关实际返回 MP4,节点按真实格式保存
- Draft:网关保存成
draft=false,节点会识别并停止,不再误当 Draft - 首帧/首尾帧:被 adaptive 图片 token 估算问题阻断
- Edit/Extend:被 adaptive 视频 token 估算问题阻断
安装目录请保持独立:
custom_nodes/
ComfyUI-Vergora-Seedance-2.0/
ComfyUI-Vergora-Seedance-2.5/
API Key 只从实例中的 config.ini 或 VERGORA_API_KEY 读取,不会保存到工作流。
