Nodes/ComfyUI Seedance/并发提交|Zhenzhen Image GK v2 图像编辑(1-3 图)
ComfyUI Node

并发提交|Zhenzhen Image GK v2 图像编辑(1-3 图)

GK v2 image editing, capped at three references — the fix-the-picture batcher

By T8mars·Created 2 months ago·Updated 3 days ago· 31
并发提交|Zhenzhen Image GK v2 图像编辑(1-3 图)
  • image1
  • image2
  • image3
  • api_config
  • future
prompt
aspect_ratioauto
resolution1k
n1
nsfw_checkfalse
skip_errorfalse
seed0

Where the plain GK v2 text-to-image node is the pack's cleanest example, this one is where GK v2's editing side lives: Zhenzhen Image GK v2 图像编辑(1-3 图), in its concurrent submit form. It's a separate node from the T2I one because the API gives image editing its own request structure - up to three reference images, its own aspect/ resolution controls, and an optional content check. If your batch job is "take these three reference shots and restyle them," this is the one.

Inputs that matter:

  • prompt - the editing instruction. This is a directive ("make it sunset lighting, keep the subject"), not a caption.
  • image1image3 - reference images, 1 to 3. The node doesn't care if you connect fewer; the API's contract here is 1–3, unlike the ten-slot image nodes elsewhere in the pack.
  • aspect_ratio - auto default, or pick the documented ratios.
  • resolution - 1k default (the GK v2 edit family tops out at 2k).
  • n - 1 to 10 requested images per job; the node downloads the primary result.
  • nsfw_check - the documented content check, off by default. Flip it on if your pipeline needs the gateway's moderation verdict on every render rather than letting it pass quietly.

Output is the single future (SEEDANCE_IMAGE_FUTURE) into 并发接收图片(30 路).

How it works

Same concurrent machinery as the rest of the family: preflight validation, hand-off to the image thread pool (up to 30 workers; SEEDANCE_IMAGE_CONCURRENCY to tune), future returned. The references upload first, the request is assembled with the edit payload structure, and the receiver collects results in slot order with per-slot skip_error and its own failure_mode. Because each of your parallel slots can carry a different reference trio, this is the natural node for "restyle the same three product shots ten ways" - one submit node per variation.

Installing it

ComfyUI Manager → search ComfyUI Seedance, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
python -m pip install -r ComfyUI_Seedance/requirements.txt

Restart, look under Seedance/并发提交. requests and an api.seedance.nz key; no downloads.

Where people get burned

  • Three references, not ten. If you wire in a ten-image chain from a multi-slot workflow, this node takes the first three and ignores the rest. Use the right edit node for the job.
  • Instruction-style prompting. This is an edit model - "make X" beats "an image of X." Treating prompt like a generation caption gives you mush.
  • nsfw_check off by default. If you're running a moderation-sensitive pipeline, remember the toggle exists; the API's own check is optional here.
  • The future isn't the image. Receiver node first, Save Image second - the family rule.

Standing fine print: metered per render (three references per slot means more upload and more cost per slot than T2I), expiring links, and reference images leaving the machine to the api.seedance.nz endpoint.

CategorySeedance/并发提交

Inputs (11)

NameTypeDefaultDescription
promptSTRINGImage editing instruction. | 图像编辑指令。
aspect_ratioCOMBOautoDocumented output aspect ratio. | 文档支持的输出比例。
resolutionCOMBO1kOutput resolution. | 输出分辨率。
nINT11–10Number of requested images, 1 to 10. | 请求图片数量,1 到 10。
nsfw_checkBOOLEANfalseEnable the documented content check. | 启用文档中的内容检查。
image1optIMAGEReference image 1 of 3. | 参考图片 1,最多 3 张。
image2optIMAGEReference image 2 of 3. | 参考图片 2,最多 3 张。
image3optIMAGEReference image 3 of 3. | 参考图片 3,最多 3 张。
api_configoptSEEDANCE_CONFIGConnect Seedance API Config; otherwise SEEDANCE_API_KEY is used.
skip_erroroptBOOLEANfalseOn failure return a placeholder image instead of stopping. | 失败时输出占位图片而不中断工作流。
seedoptINT00–18446744073709550000ComfyUI cache seed. Fixed reuses the cached result while all other inputs stay unchanged; randomize/increment/decrement starts a new execution. This value is not sent to models without documented seed support. | ComfyUI 缓存种子;Fixed 在其他输入不变时复用缓存,随机、递增或递减会触发新任务。未声明支持 seed 的模型不会收到此参数。

Outputs (1)

NameTypeDescription
futureSEEDANCE_IMAGE_FUTURE