FAL Seedream v4.5 Text To Image (Soze)

ComfyUI's local generation is where most people live, and it's good. But there are moments - a style your local model can't touch, a prompt it keeps fumbling, a client who needs the "frontier model" look - where you want ByteDance's Seedream v4.5 in the loop. FAL Seedream v4.5 Text To Image wraps fal-ai/bytedance/seedream/v4.5/text-to-image in one node: prompt in, images back as a real IMAGE batch. It's the text-to-image twin of the pack's Seedream edit node.

The economics are the whole story: hosted API, pay per image, nothing runs locally. If that's not what you want, keep reading about upscalers instead. If it is, this is a clean, honest wrapper.

How it works

Required inputs are the same core as the edit variant, minus the reference images:

• prompt - multiline, the whole job.
• image_size - auto_2K and auto_4K are the interesting options (auto infers from nothing here, so they're size presets; square, portrait_16_9, etc. give you the classic ratios).
• num_images (1–8) - results per run.
• max_images (1–8) - above 1 unlocks multi-image generation.
• enable_safety_checker - on by default.
• seed - 0 = fal picks; otherwise you pin it, and the node reports the seed it used.

The only optional input is sync_mode - when true, fal returns the images as data URIs rather than public URLs, which keeps everything self-contained.

Outputs: images (the IMAGE batch), image_urls (newline-joined), image_count, seed (the actual seed used - log it if you ever want to reproduce a run), and status. Live progress streams on the node body as the API call runs.

When it's worth it

This is the node for the "I need Seedream-quality output and I don't have the hardware" situation, or for dropping a frontier model into an otherwise-local workflow for a specific step. The multi-image mode (max_images > 1) is genuinely handy for picking a favorite from variants in one paid call rather than paying for separate runs. For everyday generation your local checkpoint is free and likely fine - this node is a specialist's tool, not a default.

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.

Gotchas: a blank prompt skips the run (clean skip, but no image - check status if you got nothing), and the num_images/max_images confusion from the edit node applies here too. If you set both high you're buying a lot of images in one call; if you only set num_images and leave max_images at 1, some API versions treat that as single-image mode. Start small, verify, then scale.