zhenzhen-image-gk-v2-edit-lowprice

Comfly_zhenzhen_image_gk_v2_edit_lowprice is the edit half of the GK V2 pair, and it's where the interesting bits live. Three reference images, a resolution choice up to 2K, a 14-way aspect-ratio dropdown (including the cinematic oddballs like 19.5:9 and 20:9), up to ten outputs per run, and a literal nsfw_check boolean. It shipped in the August domestic-storefront drop alongside its sibling tools, and it's the node you reach for when generation is fine but the image needs changing.

Cloud-run like everything in the low-price family. You connect references, describe the edit, and get edited images back as image / image_url / task_id / response.

How it works

Connect one to three reference images to image1–image3, write what you want changed in prompt, pick your output canvas, and run. The edit model does instruction-following on the references - "make the background sunset," "change the shirt to red" - the way modern edit models do.

Inputs that matter:

• prompt - the edit instruction. Be specific; "change X to Y" beats "improve it" every time.
• aspect_ratio - auto plus 14 ratios. auto follows your reference image, which is what you want for most edits.
• resolution - 1k or 2k. 2K costs more and takes longer.
• n - 1 to 10 variants per run.
• nsfw_check - off by default. Flip it on and the backend runs an explicit-content check on the request; handy if you're building a pipeline that needs to stay clean, or if borderline content keeps getting rejected and you want to know why.
• image1–image3 - the references. Most edits want one; two helps when swapping elements.

Installing and setting up

Install via ComfyUI Manager (search Comfyui-zhenzhen) or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart ComfyUI. No model downloads. Connect the Comfly Seedance2 Low Price Settings node's api_config, or set SEEDANCE_API_KEY.

Troubleshooting

Same family tripwire - no api_config, immediate "empty api_key." Specific to editing: if nothing changes between input and output, your instruction is the problem. Edit models are literal; "make it better" produces "essentially unchanged," and vague wording is the top complaint on this kind of node. Also note the reference limit is three - don't expect the multi-image composition of the G-V2 node, which takes sixteen. skip_error is available for batches.

The nsfw_check flag is worth a second thought. It's a real feature, rare in this storefront, and if you're doing content review you can wire its behavior into your workflow rather than guessing at gate failures. Just remember the gate's default is whatever the model tier ships with - the toggle is advisory, not a guarantee.