Qwen-Image 2.1 Turbo - 8-Step Text to Image ComfyUI workflow (16 GB-friendly)
From: https://freakyjolly.com/best-qwen-image-2-1-turbo-comfyui-workflow-16gb/

FILE
  Qwen_Image_2.1_Turbo_T2I_16GB.json - ComfyUI UI/graph format. Drag it onto the ComfyUI canvas.
  Based on the official Comfy-Org template "Qwen Image 2.1 Turbo: Text to Image"
  (workflow_templates: templates/image_qwen_image_2_1_turbo_t2i.json). Changes vs the template:
    - Resolution Selector set to 16:9, 1.0 MP, multiple 32 -> 1376x768 (template: 1:1, 1 MP, multiple 8)
    - fixed seed 20261025 and a photo prompt with one short sign text
    - output prefix FreakyJolly/Qwen_Image_2.1_Turbo_16GB
    - an extra note with 16 GB tips
  The sampler graph (8-step ManualSigmas, euler, CFG 1, Qwen Image 2.1 Cache, optional prompt enhancer)
  is unchanged and lives inside the "Text to Image (Qwen Image 2.1 Turbo, 16 GB)" subgraph node.

CUSTOM NODES
  None. Every node is a ComfyUI core node. Update ComfyUI (v0.38 or newer; the node list was checked
  against a ComfyUI 0.38.1 install): TextEncodeQwenImage21, QwenImage21Cache, ResolutionSelector,
  TextGenerate, ComfySwitchNode, ManualSigmas, SamplerCustomAdvanced, SaveImageAdvanced, PreviewAny.
  The workflow uses a subgraph, so use a current ComfyUI frontend.

MODELS (weights NOT included; Qwen Research License = non-commercial use only)
  All from https://huggingface.co/Comfy-Org/Qwen-Image-2.1

  ComfyUI/models/diffusion_models/
    qwen_image_2.1_turbo_int8_convrot.safetensors   7.26 GB  (default in this workflow)
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_turbo_int8_convrot.safetensors
    (optional) qwen_image_2.1_turbo_bf16.safetensors 14.23 GB
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_turbo_bf16.safetensors

  ComfyUI/models/text_encoders/
    qwen3vl_8b_int8_convrot.safetensors   9.35 GB  (default)
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors
    (optional, smaller) qwen3vl_8b_w4a8.safetensors  6.31 GB
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_w4a8.safetensors
    Prompt enhancer (only needed when refine_prompt is ON):
    qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors  9.47 GB  (default)
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors
    (optional, smaller) qwen3.5_9b_qwen_image_2.1_pe_t2i_w4a8.safetensors  6.72 GB
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3.5_9b_qwen_image_2.1_pe_t2i_w4a8.safetensors

  ComfyUI/models/vae/
    qwen_image_2.1_vae_bf16.safetensors   0.68 GB
      https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors

  Note: the ComfyUI loader drop-downs must show these exact file names. If you pick the w4a8 files,
  select them inside the subgraph node (clip_name / PE_model widgets).

KEEP THESE SETTINGS (Turbo model card + official template)
  8 steps through ManualSigmas: 1.0, 0.978453, 0.95418, 0.926626, 0.89508, 0.845148, 0.704534, 0.414568, 0
  euler sampler, CFG 1 (negative prompt is unused at CFG 1)

SIZES (Resolution Selector, multiple 32)
  16:9 1.0 MP -> 1376x768 (default) | 16:9 2.0 MP -> 1920x1088 | 16:9 4.0 MP -> 2720x1536 | 1:1 4.0 MP -> 2048x2048

TRANSPARENT PNG
  Prompt: This is an RGBA image with transparency. <your description>. The image has alpha channel and the background is transparent.
  Keep the output format PNG.

16 GB TIPS
  - If you run out of VRAM, set cache_device to cpu on the subgraph node, or switch to the w4a8 text encoder.
  - Leave refine_prompt OFF until the base run works; when ON, also turn thinking_mode ON (template note).

Output: PNG in ComfyUI/output/FreakyJolly/.
