Seedream 5.0 Pro · Text to Image

The odd widget here is enable_thinking, and it's on by default. Before generating, the model runs a pass that refines your prompt - filling in the composition and lighting details you didn't specify. Which makes this the most forgiving text-to-image node in the pack: a sloppy prompt still comes back with something deliberate. It's also the slowest, and if you're a person who writes precise prompts, the pass is latency you're paying for nothing.

About 5.4 credits (~$0.054) an image. Pro sits at the top of ByteDance's Seedream 5.0 line; Flash is cheaper and near-equally good at a glance, Lite is fixed-size, and Pro is the one for detail and legible text.

How it works

PopNow's node for bytedance/seedream-5.0-pro/text-to-image, shaped directly from the model's input schema. On queue, the node quotes the run (jobs/quote) so the price is fixed before anything is held, submits with an idempotency key, polls, downloads and decodes into a ComfyUI IMAGE batch. No local sampling, no weights, no VRAM - a closed model behind an API call, with the standard terms: per-image cost, your prompt leaving the machine, and the vendor's content filter applying.

The inputs

• prompt - subject, composition, style, lighting. With enable_thinking on you can be conversational; with it off, write like a spec.
• aspect_ratio - 1:1 default, plus 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 4:5, 5:4, 21:9, 9:21, 2:1, 1:2, 3:1, 1:3. Nearly any delivery shape without a crop.
• resolution - 1k or 2k. Two tiers on Pro rather than Flash's three, at the same per-image price, so use 2k.
• output_format - jpeg (default) or png. PNG if it's going into a local pipeline afterwards.
• enable_thinking - the switch worth experimenting with first. Run the same prompt both ways once and you'll know which mode you are.

Outputs

images - a batch - plus task id for the API console logs or support. No alpha on this node; transparency lives on Pro Edit, which has the background switch.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Or Manager → search PopNow → Install. Then key it: popnow.ai/api/console/keys → Settings > PopNow, or POPNOW_API_KEY in the environment, and restart a second time so your account's catalog rebuilds the node list. There's nothing to pip install - the pack's dependency list is empty and it uses only what ComfyUI ships with.

One housekeeping note from the README worth repeating: if you used the old single-file popnow_api.py, delete it. Its PopNow Generate node is the same node type with the same inputs, so workflows survive the switch.

Where it goes wrong

• PopNow 40201 - not enough credits. Per-image pricing means a batch loop eats a balance quietly; do the multiplication before a 40-image run.
• PopNow 400 naming a field - a ratio or format that isn't on the list, or an empty prompt.
• PopNow 40004 - the model can't serve that combination; change the one the message names.
• PopNow 40310 - unverified email. Blocks runs and reads like a mystery.
• content_rejected / generation_failed - model-side refusal or failure, refunded. Reword and go again.
• An error ending in request_id ... - give that id to support; it identifies the request exactly.
• Missing node after a model appears on popnow.ai - restart ComfyUI; the catalog is fetched at startup.

Where it fits

Honestly? If you're generating drafts, use Flash - it's cheaper and the gap at thumbnail scale is small. Pro earns its extra credit when the image is the deliverable: type, product detail, faces at 2K, anything where you'd otherwise be doing a second pass. And enable_thinking is a genuine differentiator against the local stack, where your prompt goes in exactly as typed. Whether you want a model quietly improving your wording is a taste question - but at least here it's a switch you control.