Check my GPU →

FLUX.2 [klein] 4B VRAM requirements

The smallest FLUX.2: 4B parameters, Apache-2.0 licence, a few steps per image. Fits almost any modern card.

Released 2026-01Licence: Apache-2.0Steps: 4Text encoder: Qwen3 4B (5.6 GB as FP8)
TypeImage
Parameters4B
Fits entirely from4 GB
8-bit or better from6 GB

01Measured on my RTX 5060 Ti 16 GB

FP8 · 1024×1024FASTEST
2.9s / image
flux-2-klein-4b-fp8.safetensors
Peak VRAM12.2 GB
Run range2.8–2.9 s
16-bit · 1024×1024
3.9s / image
flux-2-klein-4b.safetensors
Peak VRAM14.1 GB
Run range3.9–4.0 s
Q8_0 · 1024×1024
4.3s / image
flux-2-klein-4b-Q8_0.gguf
Peak VRAM12.4 GB
Run range4.3–4.3 s

02The files, and how much VRAM each needs

FileSizeNeededMin. VRAMQualitySource
16-bit7.8 GB9.6 GB10 GBthe original weightsblack-forest-labs/FLUX.2-klein-4B →
Q8_04.3 GB6.1 GB7 GBpractically identical to the originalunsloth/FLUX.2-klein-4B-GGUF →
FP84.1 GB5.9 GB6 GBpractically identical to the originalblack-forest-labs/FLUX.2-klein-4b-fp8 →
Q6_K3.4 GB5.2 GB6 GBvery close to the originalunsloth/FLUX.2-klein-4B-GGUF →
Q5_K_M3.1 GB4.9 GB5 GBclose; small differences in fine detailunsloth/FLUX.2-klein-4B-GGUF →
Q4_K_M2.6 GB4.4 GB5 GBgood; some loss in fine detail and textunsloth/FLUX.2-klein-4B-GGUF →
Q3_K_M2.1 GB3.9 GB4 GBnoticeable loss of detailunsloth/FLUX.2-klein-4B-GGUF →
Q2_K1.8 GB3.6 GB4 GBheavy loss; a last resortunsloth/FLUX.2-klein-4B-GGUF →

“Needed” = file + 1 GB working memory + 0.8 GB system reserve.

Free ComfyUI workflow, tested on my RTX 5060 Ti: FP8 workflow (.json). Drag the file into ComfyUI. All workflows →

03By amount of VRAM

6GBRuns wellFP88GBRuns wellQ8_010GBRuns well16-bit12GBRuns well16-bit16GBRuns well16-bit20GBRuns well16-bit24GBRuns well16-bit32GBRuns well16-bit

04Best GPU for FLUX.2 klein 4B

The cheapest cards (by launch price) that run it well, and every card sorted by memory: best GPU for FLUX.2 klein 4B → Planning bigger images or longer clips? Open the calculator →

05By graphics card

GPUVRAMVerdictBest fileNeeded
Desktop graphics cards
RTX 2060 6 GB6 GBRuns wellFP85.9 GB
RTX 3050 6 GB6 GBRuns wellFP85.9 GB
RTX 2070 Super 8 GB8 GBRuns wellQ8_06.1 GB
RTX 2080 Super 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3050 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3060 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3060 Ti 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3070 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3070 Ti 8 GB8 GBRuns wellQ8_06.1 GB
RTX 4060 8 GB8 GBRuns wellFP85.9 GB
RTX 4060 Ti 8 GB8 GBRuns wellFP85.9 GB
RTX 5050 8 GB8 GBRuns wellFP85.9 GB
RTX 5060 8 GB8 GBRuns wellFP85.9 GB
RTX 5060 Ti 8 GB8 GBRuns wellFP85.9 GB
RX 7600 8 GB8 GBRuns wellQ8_06.1 GB
RX 9050 8 GB8 GBRuns wellFP85.9 GB
RX 9060 XT 8 GB8 GBRuns wellFP85.9 GB
Arc B570 10 GB10 GBRuns well16-bit9.6 GB
RTX 3080 10 GB10 GBRuns well16-bit9.6 GB
RTX 2080 Ti 11 GB11 GBRuns well16-bit9.6 GB
Arc B580 12 GB12 GBRuns well16-bit9.6 GB
RTX 2060 12 GB12 GBRuns well16-bit9.6 GB
RTX 3060 12 GB12 GBRuns well16-bit9.6 GB
RTX 3080 12 GB12 GBRuns well16-bit9.6 GB
RTX 3080 Ti 12 GB12 GBRuns well16-bit9.6 GB
RTX 4070 12 GB12 GBRuns well16-bit9.6 GB
RTX 4070 Super 12 GB12 GBRuns well16-bit9.6 GB
RTX 4070 Ti 12 GB12 GBRuns well16-bit9.6 GB
RTX 5070 12 GB12 GBRuns well16-bit9.6 GB
RX 7700 XT 12 GB12 GBRuns well16-bit9.6 GB
RX 9070 GRE 12 GB12 GBRuns well16-bit9.6 GB
Arc A770 16 GB16 GBRuns well16-bit9.6 GB
RTX 4060 Ti 16 GB16 GBRuns well16-bit9.6 GB
RTX 4070 Ti Super 16 GB16 GBRuns well16-bit9.6 GB
RTX 4080 16 GB16 GBRuns well16-bit9.6 GB
RTX 4080 Super 16 GB16 GBRuns well16-bit9.6 GB
RTX 5060 Ti 16 GB16 GBRuns well16-bit9.6 GB
RTX 5070 Ti 16 GB16 GBRuns well16-bit9.6 GB
RTX 5080 16 GB16 GBRuns well16-bit9.6 GB
RX 7600 XT 16 GB16 GBRuns well16-bit9.6 GB
RX 7800 XT 16 GB16 GBRuns well16-bit9.6 GB
RX 7900 GRE 16 GB16 GBRuns well16-bit9.6 GB
RX 9060 XT 16 GB16 GBRuns well16-bit9.6 GB
RX 9070 16 GB16 GBRuns well16-bit9.6 GB
RX 9070 XT 16 GB16 GBRuns well16-bit9.6 GB
RX 7900 XT 20 GB20 GBRuns well16-bit9.6 GB
Arc Pro B60 24 GB24 GBRuns well16-bit9.6 GB
RTX 3090 24 GB24 GBRuns well16-bit9.6 GB
RTX 3090 Ti 24 GB24 GBRuns well16-bit9.6 GB
RTX 4090 24 GB24 GBRuns well16-bit9.6 GB
RX 7900 XTX 24 GB24 GBRuns well16-bit9.6 GB
Arc Pro B70 32 GB32 GBRuns well16-bit9.6 GB
RTX 5090 32 GB32 GBRuns well16-bit9.6 GB
Laptop GPUs
RTX 3050 Laptop 4 GB4 GBTightQ3_K_M3.9 GB
RTX 3050 Ti Laptop 4 GB4 GBTightQ3_K_M3.9 GB
RTX 2060 Laptop 6 GB6 GBRuns wellFP85.9 GB
RTX 3050 Laptop 6 GB6 GBRuns wellFP85.9 GB
RTX 3060 Laptop 6 GB6 GBRuns wellFP85.9 GB
RTX 4050 Laptop 6 GB6 GBRuns wellFP85.9 GB
RTX 2070 Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 2070 Super Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 2080 Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 2080 Super Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3070 Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3070 Ti Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 3080 Laptop 8 GB8 GBRuns wellQ8_06.1 GB
RTX 4060 Laptop 8 GB8 GBRuns wellFP85.9 GB
RTX 4070 Laptop 8 GB8 GBRuns wellFP85.9 GB
RTX 5050 Laptop 8 GB8 GBRuns wellFP85.9 GB
RTX 5060 Laptop 8 GB8 GBRuns wellFP85.9 GB
RTX 5070 Laptop 8 GB8 GBRuns wellFP85.9 GB
RX 7600M 8 GB8 GBRuns wellQ8_06.1 GB
RX 7600M XT 8 GB8 GBRuns wellQ8_06.1 GB
RX 7600S 8 GB8 GBRuns wellQ8_06.1 GB
RX 7700S 8 GB8 GBRuns wellQ8_06.1 GB
RTX 4080 Laptop 12 GB12 GBRuns well16-bit9.6 GB
RTX 5070 Laptop 12 GB12 GBRuns well16-bit9.6 GB
RTX 5070 Ti Laptop 12 GB12 GBRuns well16-bit9.6 GB
RX 7800M 12 GB12 GBRuns well16-bit9.6 GB
RTX 3080 Laptop 16 GB16 GBRuns well16-bit9.6 GB
RTX 3080 Ti Laptop 16 GB16 GBRuns well16-bit9.6 GB
RTX 4090 Laptop 16 GB16 GBRuns well16-bit9.6 GB
RTX 5080 Laptop 16 GB16 GBRuns well16-bit9.6 GB
RX 7900M 16 GB16 GBRuns well16-bit9.6 GB
RTX 5090 Laptop 24 GB24 GBRuns well16-bit9.6 GB
Unified memory
Radeon 8060S (Strix Halo) 96 GB96 GBRuns well16-bit9.6 GB
Radeon 8065S (Gorgon Halo) 160 GB160 GBRuns well16-bit9.6 GB
Apple Silicon Macs (by memory)
Mac 16 GB12.7 GBRuns well16-bit9.6 GB
Mac 18 GB14.4 GBRuns well16-bit9.6 GB
Mac 24 GB19.6 GBRuns well16-bit9.6 GB
Mac 32 GB26.8 GBRuns well16-bit9.6 GB
Mac 36 GB30.2 GBRuns well16-bit9.6 GB
Mac 48 GB40.2 GBRuns well16-bit9.6 GB
Mac 64 GB55.7 GBRuns well16-bit9.6 GB
Mac 96 GB85 GBRuns well16-bit9.6 GB
Mac 128 GB115.4 GBRuns well16-bit9.6 GB
Mac 192 GB175.4 GBRuns well16-bit9.6 GB
Mac 256 GB236.9 GBRuns well16-bit9.6 GB
Mac 512 GB498.1 GBRuns well16-bit9.6 GB

06Text encoder, VAE and other files

FileFolderSizeWhen
Qwen3 4B FP8
qwen_3_4b_fp8_mixed.safetensors
models/text_encoders5.6 GBtext encoder · smaller, recommendedDownload →
Qwen3 4B BF16
qwen_3_4b.safetensors
models/text_encoders8.0 GBtext encoder · alternativeDownload →
FLUX.2 VAE
flux2-vae.safetensors
models/vae0.3 GBrequiredDownload →

The files the official ComfyUI workflows load next to the model. Sizes read from Hugging Face (2026-10-01).

07Where the files go in ComfyUI

FileFolderLoader node
Diffusion model (.safetensors: 16-bit, FP8, INT8)ComfyUI/models/diffusion_modelsLoad Diffusion Model
GGUF file (.gguf)ComfyUI/models/unetUnet Loader (GGUF) - from the ComfyUI-GGUF node pack
Text encoderComfyUI/models/text_encodersLoad CLIP / DualCLIPLoader (or the GGUF versions)
VAEComfyUI/models/vaeLoad VAE

08Measured and reported results

LabelGPUSetupResultPeak VRAMDateSource
measuredRTX 5060 Ti 16 GBflux-2-klein-4b-fp8.safetensors · 1024x1024 · 4 steps
3 timed runs after 1 warm-up (6.66 s incl. loading). Text encoder Qwen3 4B FP8. ComfyUI 0.25.0 portable, PyTorch 2.12 cu130, 32 GB RAM. 1024x1024, 4 steps, euler, Flux2Scheduler, C
2.85 s / image12.2 GB2026-10-01
measuredRTX 5060 Ti 16 GBflux-2-klein-4b.safetensors · 1024x1024 · 4 steps
3 timed runs after 1 warm-up (6.94 s incl. loading). Text encoder Qwen3 4B FP8. ComfyUI 0.25.0 portable, PyTorch 2.12 cu130, 32 GB RAM. 1024x1024, 4 steps, euler, Flux2Scheduler, C
3.93 s / image14.1 GB2026-10-01
measuredRTX 5060 Ti 16 GBflux-2-klein-4b-Q8_0.gguf · 1024x1024 · 4 steps
3 timed runs after 1 warm-up (8.95 s incl. loading). Text encoder Qwen3 4B FP8. ComfyUI 0.25.0 portable, PyTorch 2.12 cu130, 32 GB RAM. 1024x1024, 4 steps, euler, Flux2Scheduler, C
4.34 s / image12.4 GB2026-10-01
reportedRTX 4070 Super 12 GBflux-2-klein-4b (distilled) + qwen_3_4b
“5 秒でした。”
ComfyUI 4B T2I template (distilled, 4 steps default); resolution not stated; base 4B T2I 37 s incl. model load; distilled edit 14-17 s
5 s / image—2026-01-20note.com →

Other people's numbers are copied as published, with a link. Send yours.

09Training a LoRA for FLUX.2 klein 4B

TrainerVRAMTypeSettings and quoteSource
SimpleTuner8 GBstated minimumint8 base (~12 GB in bf16); 32GB+ system RAM; single 16GB GPU recommended
“klein-4b: ~12GB: ~8GB: 32GB+”
github.com →

reported From each trainer's own docs, 2026-09-25. Why the figures differ →

10Every file tracked for FLUX.2 klein 4B

FileTypeSizeRepo
flux-2-klein-4b.safetensorsBF167.75 GBblack-forest-labs/FLUX.2-klein-4B →
flux-2-klein-4b-fp8.safetensorsFP84.07 GBblack-forest-labs/FLUX.2-klein-4b-fp8 →
flux-2-klein-4b-BF16.ggufBF167.75 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-F16.ggufF167.75 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q8_0.ggufQ8_04.30 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q6_K.ggufQ6_K3.41 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q5_1.ggufQ5_13.15 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q5_K_M.ggufQ5_K_M3.07 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q5_K_S.ggufQ5_K_S3.05 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q5_0.ggufQ5_02.92 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q4_1.ggufQ4_12.69 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q4_K_M.ggufQ4_K_M2.60 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q4_K_S.ggufQ4_K_S2.58 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q4_0.ggufQ4_02.46 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q3_K_M.ggufQ3_K_M2.12 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q3_K_S.ggufQ3_K_S2.10 GBunsloth/FLUX.2-klein-4B-GGUF →
flux-2-klein-4b-Q2_K.ggufQ2_K1.83 GBunsloth/FLUX.2-klein-4B-GGUF →

EXTRA (2026). Apache-2.0 licensed small FLUX.2. Uses Qwen3-4B text encoder (same file family as Z-Image; see qwen3-4b entry).