Zhenzhen Jimeng API

Jimeng (即梦) is ByteDance's image model, and this node is the Comfly pack's text-to-image client for it. The pitch in the README is "lowest price on the market" for a Qwen-style relay - it's a paid call per image, but the per-image cost is small enough that this is one of the more throwaway-friendly nodes in the pack. Good for Chinese-prompt image work and for when you want a ByteDance aesthetic without running anything locally.

The inputs that matter

• prompt - required. Jimeng reads Chinese prompts naturally; English works too.
• scale - the text-adherence dial, 1–10, default 2.5. Higher = sticks closer to your prompt; the README suggests cranking it when you want fidelity and leaving it low for creative drift.
• width / height - 512–2048, default 1328 square. Step of 8, so keep your numbers tidy.
• seed - -1 for random.
• use_pre_llm - prompt-expansion toggle, default off. The README's rule of thumb: turn it on for short prompts (it beefs them up), leave it off for long detailed ones (it would start rewriting your work).
• add_logo - this is a Comfly-side watermark you can stamp on, not the model's own mark: pick logo_position (corner), logo_language (中文/英文), logo_text, and logo_opacity (0–1, default 0.3). Off by default; turn it on if you're watermarking output for distribution.
• image / image_url - reference inputs. Give it an image URL directly if you have one; if you only have a local image, the node uploads it - and the README warns that reference uploads cost extra token fees based on file size. So prefer passing a URL when you've got one.

Outputs: generated_image (IMAGE), image_url, and response for the raw payload.

Install

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Or ComfyUI Manager → "Comfyui_Comfly" → restart. No models, no VRAM - you're renting Jimeng by the image. Account + credits at https://ai.comfly.chat, key in Comflyapi.json or the node's apikey.

Gotchas

The use_pre_llm toggle is the one people get wrong - the default state (off) doesn't match the README's own advice for short prompts, so actually think about it per-prompt. And the reference-image fee is easy to miss: image_url is free to use but the local image upload path silently adds a small upload charge, and uploaded URLs are transient (not stored long-term). Plan to re-upload per run or keep a stable URL around. Also, everything runs through a third-party relay - cheap is nice, but a failure mid-batch still burns the slot.