Alibaba (Qwen team) · 7B parameters · Qwen Research License (non-commercial)
What GPU do you need to run Qwen-Image-2.1?Memory and cards.
Qwen-Image-2.1 needs about 16 GB of GPU memory (VRAM) at its smallest (8-bit, ComfyUI's standard files), so a 16 GB card such as an RTX 4080 runs it. The cheapest card on our list that fits, an RTX 4060 Ti 16GB, costs about $0.14 an hour.
GPU memory
Qwen-Image-2.1, by version.
Resolution, batch size, upscalers and extra LoRAs all add memory. How we work it out
| Version | Memory needed | Card size | Cards that fit | From |
|---|---|---|---|---|
| 8-bit, ComfyUI's standard files | about 16 GB | 16–20 GB | RTX 4090, RTX 5090, RTX 3090, RTX 4080, RTX 3090 Ti, RTX 5080, RTX 5070 Ti, RTX 4070 Ti SUPER, RTX 5060 Ti 16GB, RTX 4060 Ti 16GB | $0.14/hr on an RTX 4060 Ti 16GB |
Makes and edits pictures, including ones with see-through backgrounds, from up to 10 reference images. Its 8-bit image model, text reader and VAE come to about 17.3 GB (ComfyUI loads them in turn), and one user measured a peak of 13.9 GB at 1024 × 1024 on a 16 GB card. ComfyUI's template also uses a 9.5 GB prompt helper, loaded on its own. The full-precision files come to about 32 GB, all of an RTX 5090's memory. Its license allows research and testing only, not commercial use.
Run it
Start Qwen-Image-2.1 in minutes.
The rented machine downloads it straight from Hugging Face, not through your computer. On KernelMachine, Rent an Image maker with these files, below: they load into ComfyUI's model folders by themselves once it's ready. Already renting one? Pick it under Load from link on the rental's page, or run km load with each link. ComfyUI on a rented GPU
$ km load https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_int8_convrot.safetensors
$ km load https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors
$ km load 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
$ km load https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors
Files to load
Qwen-Image-2.1 in ComfyUI.
The files ComfyUI's own "Qwen Image 2.1: Text to Image" template uses, 26.8 GB in all. Each goes in its own folder, and Load from link puts it there. Then open that template from ComfyUI's template menu.
| File | Folder | Size |
|---|---|---|
| qwen_image_2.1_int8_convrot.safetensors | models/diffusion_models | 7.3 GB |
| qwen3vl_8b_int8_convrot.safetensors | models/text_encoders | 9.4 GB |
| qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors | models/text_encoders | 9.5 GB |
| qwen_image_2.1_vae_bf16.safetensors | models/vae | 676 MB |
They start loading as soon as the rental is ready, straight from Hugging Face to the machine. Already renting an Image maker? Pick Qwen-Image-2.1 under Load from link on the rental's page. A card with 16 GB or more runs them.
Common questions
Questions, answered plainly.
How much VRAM does Qwen-Image-2.1 need?
8-bit, ComfyUI's standard files: about 16 GB.
Can an RTX 4090 run Qwen-Image-2.1?
Yes. Qwen-Image-2.1 (8-bit, ComfyUI's standard files) needs about 16 GB, which fits in the RTX 4090's 24 GB.
Can I run Qwen-Image-2.1 on KernelMachine?
Yes. Rent an Image maker with these files, on this page, starts ComfyUI on the cheapest card with 16 GB or more, and the machine downloads each file from Hugging Face into ComfyUI's folders once it's ready. Already renting an Image maker? Pick Qwen-Image-2.1 under Load from link on the rental's page.
Other models
Compare with other image and video models.
Preparing an invite-only beta
Get an email when access opens.
No launch date is set. We'll email you once when KernelMachine opens. Nothing else.