Concurrent Submit | Zhenzhen Seedream V5 Lite Edit Fal

ByteDance's Seedream V5 Lite, the budget edit model, reached through fal's queue and wrapped in Zhenzhen's concurrent layer. (Yes, the category label says "Google" - that's a pack labeling slip; Seedream is ByteDance's model. Don't let it throw you.) The lightweight V5 family is a "how cheap can a good edit be" play, and this wrapper is the one to reach for when you need a big stack of edited images and want the per-image cost to stay sane.

The notable mechanical detail is the two-count structure: num_images and max_images. Total output = num_images × max_images. If you set num_images=2 and max_images=2 you're requesting four images per generation - and fal bills per image, so do that math before you hit queue. Most people leave one of them at 1.

It's an image node: shared pool (30 workers), COMFLY_IMAGE_FUTURE task output, collect with Concurrent Collect Images (ComflyConcurrent_Image_Await) into image_1…image_30.

The inputs

• prompt - the edit instruction. Multiline.
• image1 … image10 - reference images. Ten slots is a serious multi-reference edit; V5's consistency with a pile of references is one of its selling points.
• num_images / max_images - the multiplier pair above.
• image_size - auto_2K, auto_3K, auto_4K, or the square/portrait/landscape presets. The auto_ variants scale to your content - the pack's way of getting 2K/3K/4K out of an edit without forcing a square.
• enable_safety_checker - on by default.
• image_way - image_url or base64 for references.
• seed - execution seed for cache control (compatibility-only note in the tooltip).
• poll_interval / max_poll_attempts - default 1200 × 3s = 1 hour.

Install

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

Or ComfyUI Manager → search Comfyui-zhenzhen → restart, token into api_key (ai.t8star.org overseas; api.seedance.nz for the domestic discount).

Gotchas

• The multiplier bites silently. num_images=4 and max_images=4 is sixteen billed images. Every fal lane here is pre-deducted and settled later, so the first charge may already look scary before the real number lands.
• auto_4K on an edit model is not free. Auto-size means "as big as your content allows," and 4K output on a model billed per image is the expensive end. Use auto_2K for drafts.
• "Lite" is a promise about price, not a promise about everything. V5 Lite edits are great for clean swaps and consistent product work, but it's not the flagship - don't expect Pro-level fidelity on hair or fine texture.