Zhenzhen Image G 图像生成/编辑

The "Zhenzhen" family is the interesting part of this pack: unlike the Wan, Vidu, and Nano Banana nodes that wrap well-known third-party models, Zhenzhen_Image_G2 wraps models named after the author's own brand - "Zhenzhen's affordable AI shop" (贞贞的平价AI小屋) is the storefront behind the whole reseller, and these are its house image models. In practice that means the exact upstream architecture is opaque to you. What's on the label is what you get: zhenzhen-image-g-v2-lowprice (the budget default), zhenzhen-image-g2-t2i, and zhenzhen-image-g2-i2i.

So who is this for? Someone who already has a key for this pack, wants a per-image workflow that doesn't require switching ecosystems, and cares more about a predictable price tier than about knowing precisely which model is doing the work. The honest alternative is any local open image model, which runs free forever - but if you're in this pack for its video models anyway, the Zhenzhen image nodes are a convenient way to generate stills and reference edits in the same graph, on the same bill.

How it works

Standard API-node mechanics: prompt in, local images auto-uploaded, task submitted and polled on the reseller's /v1/image/generations endpoint, result downloaded and returned as a decoded ComfyUI IMAGE. The model choice drives which inputs apply:

• zhenzhen-image-g-v2-lowprice (default) - the flexible one. Pure text-to-image, or image editing with up to 16 reference images (image1–image16). Resolution 1k/2k/4k, a size dropdown of common ratios (or custom to type your own, e.g. 5:4 or 2048x1024), and n from 1 to 10. Prompt must be 5–5000 characters.
• zhenzhen-image-g2-t2i - text-to-image only, resolution fixed at 1k, optional ratio (including adaptive).
• zhenzhen-image-g2-i2i - image editing, needs at least 1 and up to 10 reference images.

Inputs and outputs that matter

• model - the three-way switch above.
• prompt - required; 5–5000 chars for lowprice, up to 20000 for the G-2 models.
• resolution - G-2 is fixed 1k; lowprice gets 1k/2k/4k.
• ratio - G-2 only; size/custom_size/n - lowprice only.
• image1–image16 - references; G-2 editing uses the first 10, lowprice takes all 16.

Outputs: image, image_url, task_id, response.

Install

One node in T8mars/ComfyUI_Seedance:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes, pip install -r requirements.txt, restart, and add a key from api.seedance.nz/console via the Seedance API Config node or SEEDANCE_API_KEY.

Where people get burned

The main trap is the lowprice prompt floor: it rejects prompts under 5 characters, and the tooltip for G-2 is a different limit (20000), so don't assume the same rules apply when you switch models - the node hides the irrelevant fields, but the limits are still model-specific. Also, n above 1 requests multiple images from the API, but this node downloads the primary result, so the extra copies are effectively burning credits you only see in the response JSON. That's a "set n to 1 unless you have a reason" note, and one of the more subtle cost leaks in the pack. And as with every closed/reseller model here: your prompt and reference images leave the machine, and upstream moderation applies no matter what the "house model" branding suggests.