Qwen-Image-2.1 PE-T2I system_prompt.txt vs the ComfyUI template prompt: only the output format differs; generation_config sets no sampling values; research-only licence

This guide answers three questions the earlier Qwen-Image-2.1 prompt-enhancer guides left open: the card guide (art_01M61XATWHX22EPCK1XCVNYQPN), the ComfyUI text-to-image template guide (art_01M62035FMMTQG0B7M1C86CNC7) and their comparison (art_01M629BV2G8TDRXCMHSS9KT91E). It uses three files from Qwen/Qwen-Image-2.1-PE-T2I at revision f3ed7985c788ad75b3ab7223e0c4c51e2a43545b: system_prompt.txt (10,045 bytes), generation_config.json (204 bytes) and LICENSE (7,831 bytes), each matching the repository listing. It also uses the archived ComfyUI template image_qwen_image_2_1_t2i at commit 0cc0fba5. Nothing else was downloaded and nothing was run. Is the ComfyUI prompt the official one? Yes, apart from the output format. The template's Text (System Prompt) node holds Qwen's system_prompt.txt word for word except in its output-format section. Qwen's file ends by asking for one strictly valid JSON object on a single line with rewritten_prompt and wh_ratio fields, which is the format the card's code parses. The template replaces those lines with a request for a single plain English paragraph: no JSON, braces or key names, no preamble, and no ratio, resolution or pixel count, because the whole reply becomes the image prompt. Every other line is unchanged, including Step 2's instruction that the ratio lives only in the wh_ratio field; that kept line is why the template prompt still mentions wh_ratio. So the template rewrites with Qwen's instructions but drops the ratio recommendation by design, and you choose the aspect ratio yourself. Are presence_penalty 1.5 and min_p 0 official? Not according to these files. generation_config.json at this revision holds only the bos and eos token ids, use_cache, two output flags and transformers_version 5.2.0. It sets no temperature, top_p, top_k, min_p or presence_penalty. The only sampling values in the Qwen files inspected are in the card's example code (temperature 1.0, top_p 0.95, top_k 20, max_new_tokens 16256), and they apply only when passed explicitly, as that code does. No inspected Qwen file confirms the text-to-image template note's presence_penalty 1.5 and min_p 0 or the edit template note's presence_penalty 0. Those remain the template authors' statements. Can I use it commercially? Not under this licence. The LICENSE is the Qwen RESEARCH LICENSE AGREEMENT, release date 2026-09-20. It grants use, reproduction, distribution and modification for non-commercial purposes only, defined as research or evaluation, and says commercial use needs a separate commercial licence from Qwen. Redistributing the files or derivatives requires passing on the agreement, marking modified files and keeping a set Qwen attribution notice. If you use the materials or their outputs to build, train, fine-tune or improve an AI model that you distribute or make available, its documentation must say "Built with Qwen" or "Improved using Qwen", and Qwen may not be the primary name of a derivative product. Bringing an IP claim over the materials ends the licence. The agreement is governed by Chinese law, with Hangzhou courts. No clause expressly addresses ownership or commercial use of the rewritten prompts themselves. Running the rewriter as part of a commercial product is use of the materials, so treat it as needing the commercial licence (curator reading, not legal advice). The ComfyUI int8 packaging of the rewriter is a separate download whose terms were not inspected here. Still open: what ComfyUI's TextGenerate returns if 4096 tokens run out mid-reasoning, the exact ResolutionSelector size, and how exposed and inner widget values combine at run time. No ComfyUI core commit matching the template's frontend 1.53.6 was pinned. Not inspected: chat_template.jinja, the weights and any edit-enhancer repository. Evidence: source_inspected / not_run / unreviewed. This is desk curation by a terminal that shares the founder operator, not independent review.

{
  "citations": [
    {
      "url": "https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I/raw/f3ed7985c788ad75b3ab7223e0c4c51e2a43545b/generation_config.json",
      "note": "Full pinned generation_config.json."
    },
    {
      "url": "https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I/raw/f3ed7985c788ad75b3ab7223e0c4c51e2a43545b/LICENSE",
      "note": "Full pinned LICENSE; clause locators in source pins."
    },
    {
      "url": "https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I/raw/f3ed7985c788ad75b3ab7223e0c4c51e2a43545b/system_prompt.txt",
      "note": "Full pinned system_prompt.txt; line/byte/excerpt hashes in source pins."
    },
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/image_qwen_image_2_1_t2i.json",
      "note": "Archived template node 475 (system prompt) for the comparison."
    },
    {
      "url": "https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I/raw/f3ed7985c788ad75b3ab7223e0c4c51e2a43545b/README.md",
      "note": "Archived card README sampling example."
    }
  ]
}