Radar / AI media tools / Qwen-Image-2.1-PE-T2I
Qwen-Image-2.1-PE-T2I
A 9B fine-tuned LLM (Qwen3.5-VL base) acting as a dedicated prompt engineer for Qwen-Image-2.1 — takes a request in any language, returns a detailed English generation prompt plus aspect-ratio recommendation as JSON.
Why it matters
A quietly useful niche tool: attacks the prompt-quality bottleneck for non-English users building on image models, outputting pipeline-ready JSON; early adoption (6 stars, ~1.6k ModelScope downloads).
What you could build with it
An indie maker building a multilingual poster or merch generator could put this model in front of Qwen-Image-2.1, turning rough native-language requests into tuned English prompts plus aspect ratios without hiring prompt engineers.
Does it hold up?
Genuinely usable and actively adopted: the official Qwen-Image 2.1 README recommends the PE models as the default prompt path, and the community has shipped ComfyUI nodes (181 stars), GGUF/NVFP4 quants, and Apple Silicon tooling within ~2 weeks. Caveats from real usage: the shipped system_prompt.txt is mandatory (model useless without it), ~18GB VRAM for the 9B model alone, and JSON output sometimes needs repair (json_repair is used in practice).
Built with Qwen-Image-2.1-PE-T2I
- awesome-qwen-image: Qwen-Image 2.1 checkpoints, quants, prompt engines, LoRAs, and toolinggithub.com · Community curation listing PE-T2I/PE-I2I, pocket quant builds (2B, 0.8B GGUF), and tooling; documents real integration details like the JSON-after-think-tag parsing contract (127 stars, verified via GitHub API).
- qwen-image21-test-report (SKILL.md)github.com · Community test report with a production-grade PE batch pipeline: system-prompt requirement, presence_penalty 1.5 / max_new 16256 profiles, ~20-60s per case, parse_ok fallback handling — real benchmark evidence of PE usage at scale.
- benjiyaya/ComfyUI-Qwen-Image-2.1-Prompt-Enhancergithub · 181 ★ · ComfyUI custom nodes (QwenImage21_T2IPromptRewrite / EditPromptRewrite) that run the PE checkpoints inside ComfyUI with JSON parsing, wh_ratio output, and ratio_follow. Verified via GitHub API.
- janishar/qwen-image-2.1-studiogithub · 55 ★ · Local Qwen-Image-2.1 studio for Apple Silicon with --enhance flag wiring PE-T2I/PE-I2I prompt rewriting (with --think option) into the generation pipeline. Verified via GitHub API.
- xm6018924/BSAI-Qwen-Prompt-Enhancergithub · 11 ★ · ComfyUI plugin with 4-channel prompt enhancement following the official PE style, plus a templates README with a working transformers quick-start for PE-T2I. Verified via GitHub API.
Learn more
First spotted on modelscope: source.
More AI media tools
OpenMontage: open-source agentic video production systemTurns an AI coding assistant into a full video studio: 12 pipelines and 100+ tools for scripting, footage, narration…media · JEV 0.73HyperFramesHeyGen's open-source framework that turns HTML/CSS/GSAP into deterministic MP4 video via 21 agent skills and a router…media · JEV 0.69Meshy crosses $100M ARR and launches official iOS/Android appsMeshy announced Sep 30 that ARR surpassed $100M (up 100x in under two years, first AI-3D company at that milestone)…media · JEV 0.61Kandinsky 6.0 Video: open MIT-licensed audio-video generation modelsSber's team released Kandinsky 6.0 Video: Lite (3B) and Pro (29B) diffusion models generating 5-second clips with…media · JEV 0.59FLUX 3 ImageBlack Forest Labs' new multimodal image model with bounding-box layout control, up to 10 reference images, and native…media · JEV 0.57DuoMatching: few-step video generation via joint-marginal distribution matching (ByteDance)Distillation framework pairing joint distribution matching with frame-level marginal supervision from an image…media · JEV 0.51
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.