The stock QwenImage21Turbo in BF16 needs about 20GB just to load and peaks near 27GB during texttoimage generation — a 10GB card like the RTX 3080 simply runs out of VRAM The fix in one sentence: quantize the image Transformer and quantize the text encoder too, all while keeping image quality high and making it even faster
Enter the star:tonera/Qwen-Image-2.1-Turbo-Nunchaku** — an SVDQuant 4bit quantization of the official original** You can download this quantized model from Hugging Face or the ModelScope community The companion text encoder is tonera/Qwen3-VL-8B-Instruct-Nunchaku Tested on a 32GB machine: the native BF16 model OOMs — there simply is not enough VRAM to run it After quantizing the Transformer 1024×1024 resolution, 8 steps, VRAM usage drops to 2926GB with denoising at 764 it/s — images feel instant, done in about a second With the text encoder also quantized, VRAM usage falls to 1873GB at a denoising speed of 773 it/s Even then, we could only run it on the RTX XX90 series — 24GB cards With CPU offload enabled, VRAM usage is just 873GB, meaning as long as you have more than 9GB of VRAM, you can fully run this model Some may say running the GGUF format works just as well Yes, true — but this model's runtime is the Nunchaku community edition tonera/vitoom-nunchaku Measured with CPU offload on, its denoising speed still hits 611 it/s; for 8step inference it still feels like 1second image generation
Data below: measured on the same model card, an RTX 5090, quantized version vs the BF16 original:
| Metric | Key result |
|---|---|
| VRAM after load | 1983 → 1063 GiB, saving 920 GiB 464% |
| Texttoimage peak VRAM | 2644 → 1723 GiB, saving 922 GiB 348% |
| Stable texttoimage inference | 3080s → 1269s, 243×; editing 202× |
| Image quality 25image comparison | Texttoimage SSIM 0843, LPIPS 0185 — usable but not pixelperfect |
Note: the 243× is the relative speedup of the quantized version vs BF16 on the same 5090; absolute times differ on other cards is only 115–118× faster — most of the speedup comes from stepbystep inference
An LPIPS of 0185 after 4bit quantization of a large image model is already an excellent score — it means fidelity is quite high Sacrificing a bit of quality for speed and usability is perfectly acceptable
Below is a comparison of quantizing only the Transformer vs quantizing both the Transformer and the text encoder
Both images perfectly follow the prompt constraint — "a cute girl is holding a sign reads: CreatePPTXcom" — and the version with the quantized text encoder even rendered a richer background: a potted plant
We also have to mention the tonera/flux-3-image-klein-pup-nunchaku model here It was originally released by Black Forest Labs for flux3klein, designed to expand a oneline prompt into a richer, more detailed image prompt But it works for QwenImage21 as well — if you are not great at writing prompts, this gem is for you
With the same "quantized image model INT4 encoder" combo, all of the following cards can run it locally; 30/40 series use INT4, 50 series use FP4
| VRAM tier | Models precision tier | Recommendation |
|---|---|---|
| 32GB | RTX 5090 FP4 | Full 1024², no offload |
| 24GB | 3090, 3090 Ti, 4090, 4090D INT4; 5090D v2 FP4 | Runs with default settings |
| 16GB | 4080 SUPER, 4080, 4070 Ti SUPER, 4060 Ti 16GB INT4; 5080, 5070 Ti, 5060 Ti 16GB FP4 | Texttoimage works; offload on demand for editing |
| 12GB | 3080 Ti, 3060, 4070 Ti, 4070 SUPER, 4070 INT4; 5070 FP4 | Enable offload, start at lower resolution |
| 10GB | RTX 3080 INT4 | Offload lower resolution, good for prompt sketching |
Note: use FP4quantized Transformer weights on Blackwell chips; use INT4 everywhere else The text encoder always uses INT weights
In one sentence: running QwenImage21 on a 10GB card is not about more VRAM — it is three things: model quantization encoder quantization offload scheduling