Zhenzhen Image G v2.5 LowPrice 生成/编辑

There is no model file. Nothing downloads, nothing loads into VRAM, and your GPU can be a potato. You type a prompt, maybe wire in some reference images, hit queue, and a server somewhere makes the picture and hands it back as a normal ComfyUI IMAGE. Worth saying up front, because half the confusion around these packs is people hunting for a checkpoint that doesn't exist.

This is the cheapest rung of the Zhenzhen Image G v2.5 line in ComfyUI_Seedance, added in v0.15.0: text-to-image, plus reference-image editing with up to 15 images at 1K, 2K or 4K. It always returns exactly one image.

Why you'd reach for it

Reference-image editing, basically. Locally that's Qwen-Image-Edit or Flux Kontext's job - feed it a picture, ask for a change, keep the subject - and it's the dominant edit workflow of 2026. It also takes a GPU and a working node stack. This trades both for a per-call bill.

Fifteen reference slots is the pitch. Most local edit pipelines are effectively one or two images at a time; here you can hand over a set of angles and ask for something coherent. Just don't expect magic: reference conditioning is strong on faces and looser on outfit and body, which is exactly why the community's consistency work drifted toward LoRA training. You're buying the zero-setup version, not the most reliable one.

What this pack actually is

Be honest before you spend: this is a third-party paid gateway pack, not ComfyUI's official Partner Nodes. The README's front page is a shop - 贞贞的平价AI小铺, "Zhenzhen's discount AI stall" - with one door for domestic Chinese models, one for overseas, affiliate links included. The model ID this node sends (zhenzhen-image-g-v2.5-lowprice) is the gateway's own label, so there's no model card to go read. Third-party Seedance access is exactly the grey zone people warn about, up to and including hosts that quietly serve Kling and call it Seedance.

So your prompt, and every reference image you connect, get uploaded to that service. That's the mechanism, not a bug. It's also the node category that once shipped credential-stealing malware, so give a fresh pack a read before you hand it a key.

Inputs that matter

Four required fields, and only two of them will ever surprise you:

• prompt - required, 1 to 5000 characters. The node rejects a blank or oversized prompt before it spends a cent.
• resolution - 1k, 2k or 4k. Defaults to 1k, which is what you want while you're iterating.
• size - the documented aspect ratio, default 16:9. Sixteen options: auto, the usual suspects (1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 5:4, 4:5) and the wide/tall extremes (2:1, 1:2, 1:3, 3:1, 21:9, 9:21). Free-form WxH is not a thing here - that's the Official node.
• nsfw_check - off by default, enabling the documented pre-generation content check. An extra gate you can switch on, not a filter you can switch off.

Optional: image1 … image15 for editing (connect none and you get text-to-image), api_config to reuse the pack's Seedance API Config node, skip_error for a placeholder image instead of a dead workflow, and seed, which only controls ComfyUI's cache - it never reaches the model.

Two mechanism quirks are worth knowing. Slots go up in numeric order, and if you connect image3 and image7 with nothing between them, the node logs a warning and compacts them to positions 1 and 2 - so order your references deliberately. And every reference gets re-encoded to PNG and uploaded before submission, which is why a 15-image edit feels slow to start: the API rate-limits uploads and the node backs off and retries.

Outputs

Four, and you'll normally use two. image is the result as a standard ComfyUI tensor - wire it to Save Image or Preview Image. image_url is the temporary direct link from the API, task_id is the remote job ID for support tickets, and response is the full JSON. Result links expire, so don't build a workflow that only stores URLs.

Install

ComfyUI Manager: search ComfyUI Seedance or seedance, install, restart. Registry publisher is t8star, node ID seedance, so the CLI route works too:

comfy node install seedance

Manual:

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

That requirements file is one line: requests>=2.28. No torch bump, no ffmpeg requirement, no model downloads. Use the Python that actually runs ComfyUI (the portable build ships its own). Restart, and the node shows up under the Seedance category.

Then give it a key, from api.seedance.nz. Three options: add a Seedance API Config node and wire its api_config output in (cleanest for a shared workflow), set SEEDANCE_API_KEY before launching ComfyUI, or drop a config/.env next to the pack with SEEDANCE_API_KEY=.... Never bake a key into a workflow you'll share.

When it goes wrong

• "prompt is required for Image G v2.5 LowPrice" at queue time - you left it blank, or it's whitespace. The node validates before submitting.
• "Missing API key" - no config node, no env var, and no .env. Pick one.
• Node missing from the menu - wrong folder, or you never restarted. It must live at ComfyUI/custom_nodes/ComfyUI_Seedance.
• Edit requests ignore half your references - check the console for the slot-compaction warning.
• A refusal where a face should be - closed hosts filter at the source, and there are no weights to patch. You can't abliterate this, and a reseller's looser moderation still ends at the vendor's model.
• SSL errors on a Windows portable build - python -m pip install -U requests certifi in ComfyUI's Python first, then look at SEEDANCE_CA_BUNDLE.

One last thing, since it's the whole point of the "LowPrice" name: nothing here is metered by ComfyUI. You're on someone else's price list, so check it before you build a 30-image batch on top of it.