GPT-Image-2 (BudgetPixel)

If you've ever asked an image model to render text and gotten back alphabet soup, you know why GPT-Image-2 exists. OpenAI's image model is the current king of in-image typography - signage, posters, product labels, UI mockups, anything where a word has to come out correctly spelled. That single capability is why people pay a premium for it through hosted nodes like this one instead of complaining that there's no local equivalent. There isn't one that matches it.

The premium is real, and it's the most complex billing in the whole pack. It's per image, resolution × quality: at medium quality that's 60/120/180 credits for 1K/2K/4K; low is half that (30/60/90); high is the wallet-opener at 250/500/750. A 4K high-quality run is nearly a month of some plans. This node is where the Cost Estimate sibling earns its keep.

How it works

Standard BudgetPixel image-node flow: POST /images/gpt-image-2 with your prompt, poll /images/{id}, download the IMAGE batch. The capability set is where it gets interesting: text-to-image, plus editing with up to nine reference images - the most of any node in this pack. Wire a big batch in and it becomes a composition engine with typography superpowers. The nine-slot ceiling is worth remembering when your other model nodes cap at three or four.

The inputs that matter

• prompt - required, multiline. For text work, put the words you need in quotes and describe the style; GPT-Image-2 is the one model where the exact string in your prompt actually shows up on the canvas.
• resolution - 1K / 2K / 4K. The biggest driver of both quality and price.
• quality - low / medium / high. The other half of the bill. Medium is the sensible default; high exists for hero shots where the price is the point.
• reference_images - up to 9 for editing and composition.
• aspect_ratio and num_images (1–4) round it out, plus seed.

Output is images (IMAGE batch).

Install

Manager (search "BudgetPixel") or:

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

Restart, set BUDGETPIXEL_API_KEY or budgetpixel.json ({"api_key": "bpx_live_..."}) in the pack folder. No weights, requests only.

Where people get burned

The billing is the whole game here. The node help text spells out the resolution × quality matrix, and it's not decoration - a careless default of high/4K will vaporize credits on a run you thought was cheap. Run the Cost Estimate node first with {"quality": "high", "resolution": "4K"} and watch the number before you commit. Second, the moderation reality: this is OpenAI's model, so OpenAI's content policy rides along - the closed-source-models doc covers how DALL-E's censorship lineage wobbled over time, and GPT-Image-2 inherits a real filter. Third, don't send typography jobs to cheaper models expecting GPT results; the entire value proposition is the text. If you're generating layout-heavy graphics, this is the node; if you're generating a portrait, it's the wrong premium to pay.