Wan 2.7 Pro (BudgetPixel)

Wan 2.7 Pro is the professional edition of Alibaba's Wan image model, and its headline feature has a footgun attached. The "pro" upgrade is 4K output - but only for text-to-image. The size dropdown offers 4K, and the moment you connect a single reference image the model quietly falls back to 2K. It's written in the tooltip, but nobody reads tooltips until they've been burned, so: 4K is a pure text-to-image feature. Want 4K edits? You don't, because this node won't give them to you.

Why you'd reach for it

When you need the biggest clean render the Wan line can do. 4K text-to-image from a decent prompt is the "print it" or "put it on the hero banner" tier, and 80 credits per generation is what Alibaba charges for it against the standard model's 35. It also keeps the standard Wan feature set - thinking mode for prompt planning, and multi-reference composition from up to nine images. If you mostly work with reference images, though, the extra 45 credits over plain Wan 2.7 buys you nothing, because the reference path caps at 2K either way. That's the honest math: Pro is worth it for text-to-image at 4K, and a quiet waste for edits.

The inputs that matter

• prompt - required.
• size - 1K / 2K / 4K. The 4K tier applies to text-to-image only; with references the output falls back to 2K.
• thinking_mode - off by default; text-to-image only, ignored when reference images are connected.
• reference_images - up to 9 for editing/multi-reference composition.
• num_images - 1–4, billed per image.
• seed - -1 for random, real value to reproduce.

Output is images (IMAGE).

How it works

Same thin-client mechanism as the whole pack: IMAGE tensors to PNG, sent inline or via upload, a job created on BudgetPixel's servers, poll until done, download the result. Your local setup contributes nothing but the graph. And the shared async rule holds: interrupt and the server-side job keeps running and billing.

Install and key setup

ComfyUI Manager → BudgetPixel, or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

requests is the only dependency. API key from a paid plan via BUDGETPIXEL_API_KEY or budgetpixel.json in the pack folder; keys never serialize into workflow JSON.

Where people get burned

The 4K/2K fallback is the trap, full stop. Connect a reference and your 4K selection silently downgrades - you'll only notice when you look at the file and wonder where the resolution went. Check whether you're actually using reference images before paying for Pro. Standard gotchas also apply: force_rerun defeats ComfyUI's output caching (and costs you again), and a free account hits api_access_not_enabled until you're on a paid plan.