CyberNative Qwen-Image 2.1 preflight Runs offline in this tab

Free check · Qwen-Image 2.1 in ComfyUI

Check your Qwen‑Image 2.1 workflow before the first run.

Qwen-Image 2.1 needs up to four separate model files. Some mistakes run without an error: the prompt enhancer in the text-encoder slot, or the enhancer without its system prompt. Drop in your workflow and get a short list of what to fix.

Private Your workflow never leaves this page. Nothing is uploaded or stored, and the page’s security policy blocks all network requests.

Check a workflow
  • UI or API format
  • Subgraphs included
  • Every rule cites its source

Four files · where each one goes

  1. 1Prompt enhancer · optionalqwen3.5_9b_qwen_image_2.1_pe_t2i / _pe_i2iCLIPLoader → Generate Text
  2. 2Text encoderqwen3vl_8b_int8_convrot · bf16 · w4a8CLIPLoader, type qwen_image → Text Encode Qwen Image 2.1
  3. 3Diffusion modelqwen_image_2.1_int8_convrot · bf16 · .ggufLoad Diffusion Model · GGUF: Unet Loader (GGUF)
  4. 4VAEqwen_image_2.1_vae_bf16Load VAE
The PE and the text encoder both load through CLIPLoader from models/text_encoders/. Only the text encoder feeds the image encoder. The PE feeds Generate Text, and its output becomes your prompt. Swap them and ComfyUI still runs, but the images come out garbled.

Found in the official templates · package 0.11.70 · ComfyUI 0.38.0

In templates package 0.11.70, both Qwen Image 2.1 templates can run the prompt enhancer without its system prompt.

Status Fix released in templates package 0.11.73. Both templates in 0.11.73 (checked 2026-10-01) and 0.11.76 (checked 2026-10-03) contain the September 30 system-prompt fix and get no system-prompt flag here when freshly opened.

ComfyUI 0.38.0 installs templates package 0.11.70; current ComfyUI source pins 0.11.76 (checked 2026-10-03). Updating the package does not change a workflow you saved from 0.11.70: it keeps its old wiring, so check it anyway.

Image Edit in 0.11.70 has the enhancer switched on. Its system prompt node is not connected, and Generate Text’s use_default_template is off, so connecting the node alone still drops the prompt.

Text to Image in 0.11.70 connects the system prompt, but leaves Generate Text’s use_default_template off. ComfyUI 0.38.0 passes system_prompt only when that setting is on, so the prompt is dropped. The enhancer is off by default, so this starts when you switch it on.

We checked what reaches the enhancer in ComfyUI 0.38.0 on 2026-09-30: no system prompt, in either template. It still writes a prompt, but a different one. In our Text to Image test it came out 22–39% shorter.

Fix Image Edit: connect the system prompt node to system_prompt and turn use_default_template on.
Text to Image: turn use_default_template on and delete the <|im_start|>system / <|im_end|> wrapper from the text. The check below finds each step.

Check a workflow

In ComfyUI: Workflow › Export (or Export (API)). Then drop the .json file here.

1Workflow

Drop a workflow .json here or

Or paste the JSON below.

2Your files · optional

Your result appears here: a fix list, the four slots as your workflow fills them, a VRAM plan and everything that passed.

Limits

What a workflow file cannot tell us

  • Launch flags such as --use-sage-attention, which users link to blank images.
  • Your hardware. FP8 weights on Apple Silicon (MPS) are reported to give white images.
  • How ComfyUI moves weights between VRAM, RAM and disk while it runs.
  • Which version of a custom node pack you installed.
  • Image quality, garbled text inside images, or speed.

We tested it on 17 problems reported in public threads, each rebuilt from the official templates. It catches the 12 that live in the workflow or file names. It misses the 5 that come from launch flags, hardware or runtime. Real exports from ComfyUI 0.38.0 get the same result as the template files they came from.