FLUX.2 [klein] 9B on 16 GB of VRAM
Download Q8_0 (10.0 GB). With the model's working memory it needs about 12.3 GB, leaving 3.7 GB spare on 16 GB. Quality: practically identical to the original.
On RTX 40 and RTX 50 GPUs take the FP8 file instead of Q8_0: same class, faster there. On older NVIDIA, AMD and Intel GPUs Q8_0 is the better 8-bit file.
01What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | flux-2-klein-9b-Q8_0.gguf Q8_0 · unsloth/FLUX.2-klein-9B-GGUF | models/unet | 10.0 GB | Download → |
| Text encoder | qwen_3_8b_fp8mixed.safetensors Qwen3 8B FP8 · Comfy-Org/flux2-klein-9B | models/text_encoders | 8.7 GB | Download → |
| VAE | flux2-vae.safetensors FLUX.2 VAE · Comfy-Org/flux2-dev | models/vae | 0.3 GB | Download → |
| Total download · keep about the same free on disk | 19.0 GB | |||
Alternatives: Qwen3 8B BF16 (16.4 GB). The 16-bit text encoder is a little more faithful; the smaller one is chosen here because it loads faster and needs less RAM. Sizes read from Hugging Face (2026-09-25). “Download” links start the file directly; the file name links are the same files the official ComfyUI workflows use.
System RAM: 32 GB or more. ComfyUI keeps the model file, the text encoder and the VAE in system RAM and moves them to the GPU as needed. With this set of files that is about 19.0 GB, plus roughly 6 GB for Windows, ComfyUI and a browser: 25.0 GB in total. With less RAM it still runs, but Windows starts swapping to disk and loading gets very slow. calculated
02Every file, on 16 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · black-forest-labs | 18.2 GB | 20.5 GB | 4.5 GB too big | the original weights | Hugging Face → |
| Q8_0 ← GGUF · unsloth | 10.0 GB | 12.3 GB | fits · 3.7 GB spare | practically identical to the original | Hugging Face → |
| FP8 SAFETENSORS · black-forest-labs | 9.4 GB | 11.7 GB | fits · 4.3 GB spare | practically identical to the original | Hugging Face → |
| Q6_K GGUF · unsloth | 7.9 GB | 10.2 GB | fits · 5.8 GB spare | very close to the original | Hugging Face → |
| Q5_K_M GGUF · unsloth | 7.0 GB | 9.3 GB | fits · 6.7 GB spare | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · unsloth | 5.9 GB | 8.2 GB | fits · 7.8 GB spare | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M GGUF · unsloth | 4.8 GB | 7.1 GB | fits · 8.9 GB spare | noticeable loss of detail | Hugging Face → |
| Q2_K GGUF · unsloth | 4.0 GB | 6.3 GB | fits · 9.7 GB spare | heavy loss; a last resort | Hugging Face → |
Sizes from the Hugging Face file listing, read 2026-09-25. “Needed” = file + 1.5 GB working memory for this model + 0.8 GB kept free for the system.
03The text encoder
Qwen3 8B: 16.4 GB as 16-bit, 8.7 GB as FP8. ComfyUI encodes the prompt first and can push the encoder out of VRAM before sampling, so it does not have to fit together with the model. On 16 GB the FP8 encoder fits on its own, so prompt encoding stays fast.