FAL Seedream v5 Pro Edit (Soze)

Every Seedream tier has a reason to exist. v5 Lite is the volume option, v4.5 is the middle, and v5 Pro is the one you pick when the edit is the deliverable. FAL Seedream v5 Pro Edit wraps fal-ai/bytedance/seedream/v5/pro/edit: up to ten input images, a prompt, and edited results back in your graph as an IMAGE batch. It's the most expensive Seedream node in this pack and, for hard edits, the one that justifies it.

Hosted API, pay per run, nothing local. Same family DNA as the other Seedream nodes, but the details differ in ways that matter - most notably that Pro caps output format choices at JPEG/PNG and runs a different image-size ladder.

How it works

Required inputs:

• prompt - the edit instruction, multiline.
• image_size - Pro's ladder is auto_2K, auto_1K, then the classic square/portrait_16_9/landscape_16_9 ratios. No 3K/4K here - Pro's auto_2K is its top auto setting, so don't go hunting for a 4K preset that doesn't exist on this endpoint.
• num_images (1–6) - note the lower ceiling than the other Seedream nodes; Pro caps at six separate generations per call, not eight.
• output_format - jpeg (the default) or png. If you need alpha or lossless, png; otherwise JPEG is the model's own default and fine.
• enable_safety_checker - on by default.

Reference inputs: ten image_N slots or an image_batch (capped at ten frames, individual slots winning). sync_mode returns data URIs instead of public URLs.

Outputs: images, image_urls, image_count, seed, and status. Like the other Seedream tiers there's no seed input - you get the seed fal used, but you can't pin it going in.

When it's worth it

Pro is the tier for edits that have to hold up: hero imagery, commercial work, edits where a subtle failure means redoing the whole thing. The practical difference from Lite shows up exactly there - on hard edits, the extra spend buys fewer visible artifacts and better prompt adherence. If you're doing a run of fifty small tweaks, Lite is the financially correct choice; if you're doing one image that has to be right, this is the node.

Getting it running and gotchas

Standard pack install plus the FAL key:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

FAL_KEY in your environment, a .env in the pack root, or config.ini under [API]. Missing it reads "FAL_KEY is not configured" on the node.

The gotchas that actually bite: blank prompt = clean skip with no image (check status); upload failures show per-slot in status; and the two ceilings that catch people - num_images maxes at six on Pro (setting eight gets clamped or rejected) and output_format is JPEG-or-PNG only, so if you were reaching for webp like the GPT Image nodes offer, this endpoint simply doesn't do it. And as with Lite, don't look for a seed input - that's a v4.5 capability, not a Pro one.