Qwen Image 2.1 Ultimate 12GB — Turbo 8-step 2K T2I + Edit 1-3 Images + Remove Background + Fun ControlNet Pose/Depth/Canny

☕ One coffee = one happy creator: https://buymeacoffee.com/the_frizzy1

Every Qwen Image 2.1 mode in one workflow, picked from one dropdown. Text to image at native 2K, editing with one to three of your own images, remove background straight to a PNG with the alpha kept, and Fun ControlNet on pose, depth or edges. The Turbo model is on by default, 8 steps.

What you need

✔ Custom nodes — None. Core ComfyUI only.
✔ ComfyUI version — 0.38 or newer — "Apply Fun ControlNet" is a core node since 0.38. 0.39.2 tested.
✔ VRAM — 12 GB — tested, not estimated. Peak 11.6 GB (Turbo, 2K)
✔ System RAM — 2.6 GB peak (Turbo, 2K), in a 32 GB machine
✔ Tested on — RTX 3060 12 GB · 32 GB DDR4 · ComfyUI 0.39.2 (Comfy Desktop)
✔ Licence — Workflow MIT · Qwen Image 2.1 and Turbo: Qwen Research License (non-commercial)

Models to download

Click a file name to download it. Put it in the folder shown, inside ComfyUI/models/.

Required for text to image and edit with Turbo — 3 files, 17.4 GB

diffusion_models/

File: qwen_image_2.1_turbo_int8_convrot.safetensors — 7.3 GB (Turbo on (8 steps))
Download:
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_turbo_int8_convrot.safetensors

Put it here:
ComfyUI/models/diffusion_models/

text_encoders/

File: qwen3vl_8b_int8_convrot.safetensors — 9.4 GB (all modes)
Download:
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors

Put it here:
ComfyUI/models/text_encoders/

vae/

File: qwen_image_2.1_vae_bf16.safetensors — 0.7 GB (all modes)
Download:
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors

Put it here:
ComfyUI/models/vae/

Turbo off — the base model, only if you want the 40-step path.

diffusion_models/

File: qwen_image_2.1_int8_convrot.safetensors — 7.3 GB (Turbo off (base))
Download:
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_int8_convrot.safetensors

Put it here:
ComfyUI/models/diffusion_models/

ControlNet modes — 5 more files, 7.8 GB. Only for pose, depth and edges.

model_patches/

File: qwen_image_2.1_fun_controlnet_union_int8_convrot.safetensors — 3.8 GB (ControlNet modes)
Download:
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/model_patches/qwen_image_2.1_fun_controlnet_union_int8_convrot.safetensors

Put it here:
ComfyUI/models/model_patches/

checkpoints/

File: sdpose_wholebody_fp16.safetensors — 1.9 GB (ControlNet: pose)
Download:
https://huggingface.co/Comfy-Org/SDPose/resolve/main/checkpoints/sdpose_wholebody_fp16.safetensors

Put it here:
ComfyUI/models/checkpoints/

diffusion_models/

File: rt_detr_v4-x-hgnet_fp16.safetensors — 0.1 GB (ControlNet: pose)
Download:
https://huggingface.co/Comfy-Org/SDPose/resolve/main/diffusion_models/rt_detr_v4-x-hgnet_fp16.safetensors

Put it here:
ComfyUI/models/diffusion_models/

File: lotus-depth-d-v1-1.safetensors — 1.7 GB (ControlNet: depth)
Download:
https://huggingface.co/Comfy-Org/lotus/resolve/main/lotus-depth-d-v1-1.safetensors

Put it here:
ComfyUI/models/diffusion_models/

vae/

File: vae-ft-mse-840000-ema-pruned.safetensors — 0.3 GB (ControlNet: depth)
Download:
https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.safetensors

Put it here:
ComfyUI/models/vae/

Modes

One Mode dropdown. The inputs start bypassed (purple), so text to image runs straight away.

• Text to image — prompt uses, 2K: 57 s · 1 MP: 13 s turbo, rtx 3060

• Edit (your images) — Image 1 (+ 2, 3 optional; write <Picture 1> etc. in the prompt) uses, 1 MP: 22 s turbo, rtx 3060

• Edit: remove background (PNG) — Image 1 only uses, 1 MP: 29 s (base model: 54 s) turbo, rtx 3060

• ControlNet: pose — Control image, a person — SDPose reads the pose uses, 1 MP: 24 s turbo, rtx 3060

• ControlNet: depth — Control image, any scene — Lotus makes the depth map uses, 1 MP: 27 s turbo, rtx 3060

• ControlNet: edges (canny) — Control image — lines, text, products uses, 1 MP: 23 s turbo, rtx 3060

Turbo on is the official Turbo model at 8 steps: 2K in 57 s. Turbo off is the base model at 40 steps with Comfy Kitchen attention: 2K in 4 min 07 s, and base at 25 steps is 3 min 02 s. Same prompt, same seed. The quality is your call.

What did not work

• ControlNet at native 2K broke. The pose result filled the background with scribbles and the depth result came back as a coloured depth map. At 1 MP all three modes were clean. Keep ControlNet megapixels at 1.

• 2K ControlNet also took 1 min 23 s to 1 min 30 s instead of 23 s to 27 s.

• Remove background does not always work. A portrait on Turbo and on base, and a product shot, came out cut out cleanly. My own wide room shot came back almost fully transparent 3 out of 3 times — Turbo, base, and another seed.

• In remove-background mode your prompt is ignored and Images 2 and 3 are not used.

• Keep every input you are not using bypassed. An un-bypassed input you do not need throws "Invalid image file" before the run starts; a bypassed input your mode does need stops the run at a "Needs …" node.

• Loaders go red for models you did not download, because ComfyUI checks every loader. Point the red ControlNet, SDPose and Lotus loaders at any file you have — they never load outside the ControlNet modes.

Get the workflow

• GitHub: https://github.com/Thefrizzy1/The_frizzy1-Workflows/tree/main/workflows/qwen/image-2.1-ultimate-12gb

• Hugging Face mirror: https://huggingface.co/The-frizzy1/The_frizzy1-Workflows/tree/main/workflows/qwen/image-2.1-ultimate-12gb

More videos on this stuff: https://www.youtube.com/playlist?list=PLG_da2oqTx7k