zhenzhen-seedream-v5-pro

This is the big-res version of the pack's Seedream V5 Pro access. Where the low-price Comfly_sd2_seedream_v5_pro_lowprice node goes through the domestic shop, Comfly_seedream_v5_pro calls the overseas zhenzhen shop's OpenAI DALL-E-compatible /v1/images/generations endpoint - which means it behaves like an official image API: sizes up to 4096×4096, url or b64_json response format, and up to ten reference images for editing. It's the "I need a big, clean Seedream render and I want it to act like the official API" option.

Mechanically it's a direct HTTP call, not the fal queue: your prompt (and optional reference images, shipped as base64 data URIs) go to the shop, and the image comes back either as a URL or base64, which the node decodes into a normal ComfyUI IMAGE tensor. Credentials go in the apikey field (note the spelling - the openai-format nodes use apikey, not api_key), which gets saved into the pack's config so you don't have to re-enter it every run.

The inputs that matter

• prompt - required.
• model - defaults to seedream-v5-pro; it's a free-text field, so you can try alternate model IDs the shop exposes.
• size - 1024x1024 up to 4096x4096, plus custom (then set custom_size, e.g. 2048x1536). Bigger is dramatically slower and pricier.
• response_format - url (default) or b64_json. The README's advice: if you're passing a lot of reference images, prefer url to avoid base64 bloat.
• output_format - png or jpeg.
• image1–image10 and image_urls - reference images, either as tensors or as a newline/comma-separated list of URLs. The URL list is the cheap way to edit with images you already have hosted.

Outputs: image (tensor), image_url, and response. There's no task_id here because this endpoint is synchronous - no async task to look up.

Installing it

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

or ComfyUI Manager (search "Comfyui-zhenzhen"), restart. Paste your overseas zhenzhen token (https://ai.t8star.org) into apikey. No models, no VRAM.

Common issues

• "API key not provided" - you typed into a different node or the config save failed; the key is stored per-pack, so a fresh apikey input overrides and re-saves it.
• 4K hangs or fails - big renders over the overseas endpoint are the slowest job in this pack. Prefer url response format and give it time; 4K Seedream is genuinely minutes.
• Reference editing flaky - too many base64 references can hit payload limits. Move to image_urls and url response format.

If you need Seedream V5 Pro at production resolution and you're outside the domestic shop's reach, this is the node - it's the official-format path, which also makes it the easiest to reason about when something goes wrong.