FLUX.1
The highest quality model family. T5-XXL + CLIP-L text encoding with a flow-matching transformer.
- Developer: Black Forest Labs
- License: Apache 2.0 (Schnell), FLUX.1 Dev Non-Commercial (Dev)
- HuggingFace: black-forest-labs/FLUX.1-schnell, black-forest-labs/FLUX.1-dev
Variants
| Model | Steps | Size | Notes |
|---|---|---|---|
flux-schnell:q8 | 4 | 12 GB | Fast, general purpose |
flux-schnell:q6 | 4 | 9.8 GB | Best quality/size trade-off |
flux-schnell:bf16 | 4 | 23.8 GB | Full precision (>24 GB VRAM) |
flux-schnell:q4 | 4 | 7.5 GB | Lighter |
flux-dev:q8 | 25 | 12 GB | Full quality |
flux-dev:q6 | 25 | 9.9 GB | Best quality/size trade-off |
flux-dev:bf16 | 25 | 23.8 GB | Full precision (>24 GB VRAM) |
flux-dev:q4 | 25 | 7 GB | Full quality, less VRAM |
Fine-Tunes
| Model | Steps | Size | Style |
|---|---|---|---|
flux-krea:q8 | 25 | 12.7 GB | Aesthetic photography |
flux-krea:q6 | 25 | 9.8 GB | Aesthetic photography |
flux-krea:q4 | 25 | 7.5 GB | Aesthetic photography |
flux-krea:fp8 | 25 | 11.9 GB | Aesthetic photography |
jibmix-flux:fp8 | 25 | 11.9 GB | Photorealistic |
jibmix-flux:q5 | 25 | 8.4 GB | Photorealistic |
jibmix-flux:q4 | 25 | 6.9 GB | Photorealistic |
jibmix-flux:q3 | 25 | 5.4 GB | Photorealistic, lighter |
ultrareal-v4:q8 | 25 | 12.6 GB | Photorealistic (latest) |
ultrareal-v4:q5 | 25 | 8.0 GB | Photorealistic |
ultrareal-v4:q4 | 25 | 6.7 GB | Photorealistic, lighter |
ultrareal-v3:q8 | 25 | 12.7 GB | Photorealistic |
ultrareal-v3:q6 | 25 | 9.8 GB | Photorealistic |
ultrareal-v3:q4 | 25 | 7.5 GB | Photorealistic, lighter |
ultrareal-v2:bf16 | 25 | 23.8 GB | Full precision |
iniverse-mix:fp8 | 25 | 11.9 GB | Realistic SFW/NSFW mix |
Recommended Dimensions
| Width | Height | Aspect Ratio |
|---|---|---|
| 1024 | 1024 | 1:1 (native) |
| 1024 | 768 | 4:3 |
| 768 | 1024 | 3:4 |
| 1024 | 576 | 16:9 |
| 576 | 1024 | 9:16 |
| 768 | 768 | 1:1 |
Using non-recommended dimensions will trigger a warning. All values must be multiples of 16.
Examples
FLUX Schnell Q8 — 4 steps, seed 42:
bash
mold run flux-schnell:q8 "A majestic snow leopard perched on a Himalayan cliff at golden hour, cinematic lighting, photorealistic" --seed 42
FLUX Dev Q4 — 25 steps, seed 1337:
bash
mold run flux-dev:q4 "A cozy Japanese tea house interior with warm lantern light, steam rising from ceramic cups, watercolor style" --seed 1337
LoRA Support
FLUX models support LoRA adapters in both BF16 and GGUF quantized modes:
bash
mold run flux-dev:bf16 "a portrait" --lora style.safetensors --lora-scale 0.8
mold run flux-dev:q4 "a portrait" --lora style.safetensors --lora-scale 0.8VRAM Notes
- Full BF16 (23 GB) auto-offloads on 24 GB cards — blocks stream CPU↔GPU
- GGUF quantized (Q4/Q8) fits without offloading
- Use
--eagerto keep encoders loaded between generations (faster, more VRAM) - T5-XXL encoder auto-selects quantized variant when VRAM is tight