Best Flux Model for 8GB VRAM (8GB VRAM)
Running Flux on an 8GB GPU — Klein 4B and quantized GGUF builds
The best Flux model for 8GB VRAM in 2026 is FLUX.2 Klein 4B — its GGUF is only ~2.6GB at Q4_K_M and renders a 1024px image in 4 steps (~15–30s on an 8GB card). For the classic FLUX.1 look, run FLUX.1 dev GGUF Q4_K_S (~6.8GB) with ComfyUI’s --lowvram flag — expect ~90–150s per 20-step image. Full FP16 Flux (~24GB) will not fit 8GB.
Models that fit in 8GB VRAM
Tested reference: RTX 4060 / RTX 3060 Ti (8GB). Figures are for a 1024px image.
| Model | Size | Build | VRAM used | Speed |
|---|---|---|---|---|
| WINNERFLUX.2 Klein 4B (GGUF) Newest low-VRAM pick — 4-step, Apache 2.0, fits 8GB even at Q8_0 (~4.3GB). Outputs look different from FLUX.1. ComfyUI + Unet Loader (GGUF) | 4B | GGUF Q4_K_M | ~2.6GB | ~15–30s / image |
| FLUX.1 dev GGUF Q4_K_S The classic Flux look with the biggest LoRA ecosystem — 20 steps; needs --lowvram and the quantized T5 encoder. ComfyUI + city96 GGUF loader | 12B | GGUF Q4_K_S | ~6.8GB | ~90–150s / image |
| FLUX.1 Schnell GGUF Q4_K_S Apache-2.0 and 4-step — the fast FLUX.1-flavoured option for drafts and iteration. ComfyUI + Schnell GGUF | 12B | GGUF Q4_K_S | ~6.8GB | ~20–30s / image |
What won't fit in 8GB
- ✗Flux.1 dev FP16 (needs ~24GB) — Full precision — needs a 24GB card.
- ✗Flux.1 dev FP8 (needs ~12GB) — Needs 12GB — see the 12GB guide.
How to fit more in 8GB
- →Use the quantized GGUF T5 text encoder (t5-v1_1-xxl GGUF) — the fp16 T5 alone is ~9GB and blows an 8GB budget before the model even loads.
- →Launch ComfyUI with --lowvram for the 12B FLUX.1 builds — weights stream from system RAM (~20–30% slower, but it runs). 16–32GB of system RAM matters as much as VRAM here.
- →Stay at 1024×1024, and for the 4-step models (Klein, Schnell) keep steps at 4 with CFG/guidance 1.0 — extra steps just waste time.
Quick start
git clone https://github.com/comfyanonymous/ComfyUIComfyUI-Manager → install "ComfyUI-GGUF" (city96)Go from "it runs" to actually building
All 561 chapters — running local models, RAG, agents, fine-tuning — plus the Python Lab and every course added later.
Frequently asked questions
Can Flux run on 8GB VRAM?
Yes — two good ways. FLUX.2 Klein 4B (GGUF ~2.6GB, 4 steps, ~15–30s/image) fits easily, and FLUX.1 dev GGUF Q4_K_S (~6.8GB) fits with --lowvram at ~90–150s per 20-step 1024px image. Full FP16 Flux (~24GB) will not fit.
Klein 4B or FLUX.1 on an 8GB card?
Klein 4B is smaller, far faster (4 steps) and Apache 2.0, but its outputs differ from FLUX.1. Pick FLUX.1 dev — or the 4-step Schnell — when you need the classic look or its huge LoRA library; otherwise start with Klein.
Which Flux GGUF quant should I download for 8GB?
For FLUX.1 dev, Q4_K_S (~6.8GB) is the sweet spot and Q4_0 the compatibility fallback; drop to Q3_K_S (~5.2GB) on a 6GB card. Klein 4B is so small you can run it at Q8_0 (~4.3GB) on 8GB.
Related guides
Ready to Go Beyond Tutorials?
20 structured courses with hands-on chapters - build RAG chatbots, AI agents, and ML pipelines on your own hardware.
Was this helpful?