FAL Seedream v5 Lite Text To Image (Soze)

If you want Seedream output without Seedream flagship pricing, v5 Lite is ByteDance's budget tier, and FAL Seedream v5 Lite Text To Image is the one-node wrapper for fal-ai/bytedance/seedream/v5/lite/text-to-image. Prompt in, images back as a real IMAGE batch. It's the text-to-image twin of the Lite edit node, and it carries the same pitch: the quality is a notch below the flagship, but the per-image cost makes it the sane default for anything that isn't a hero shot.

Hosted API, pay per run, nothing local. That's the whole deal - and the reason this node earns its place is that Lite's price point is where "use a frontier model per prompt" stops feeling like a splurge.

How it works

Required inputs:

• prompt - multiline.
• image_size - auto_2K, auto_3K, auto_4K plus the classic ratios. 2K is the sensible default; 3K/4K cost more and rarely matter unless a client's delivery spec says so.
• num_images (1–8) - results per run.
• max_images (1–8) - above 1 unlocks multi-image generation.
• enable_safety_checker - on by default.

Optional inputs, and here's where Lite's text-to-image differs from the edit variant: return_byteplus_urls - when on, the image_urls output gives you ByteDance-trusted URLs that expire after 24 hours (fine if you're shipping URLs around immediately; the default downloads into ComfyUI anyway, so most people never touch it). Plus sync_mode for data-URI output.

Outputs: images, image_urls, image_count, seed (fal's chosen seed - no seed input on this node, same as the Lite edit variant), and status, with live progress on the node body.

When it's worth it

The honest answer: Lite is the "good enough" tier for volume. Batch a bunch of concept images, iterate on prompts without watching the meter scream, use it as the generator feeding a local upscale pass. When a result needs to be great, that's when you reach for the flagship. Also worth knowing: because results download as real tensors, you can chain this into a local upscaler (the KB's guidance on upscaling applies - more pixels from a strong ESRGAN or SeedVR2 pass after a cheap hosted generation is a classic hybrid setup).

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].

Gotchas: blank prompt = clean skip, no image (check status); num_images/max_images confusion applies as always; and if you enabled return_byteplus_urls expecting permanent URLs, those 24-hour expiries will eventually bite anyone holding onto the link. For anything you want to keep, the images output (which downloads locally) is the one that lasts.