Flux Kontext GGUF — 4GB VRAM, 2-Pass + LoRA
By: The_frizzy1 Hardware target: 4 GB VRAM (laptop) CivitAI: https://civitai.com/models/1311703/flux-kontext-gguf-workflow-lowvram-realistic-cinematic-2pass-lora YouTube: https://www.youtube.com/@the_frizzy1
🎥 Video Explainer: https://www.youtube.com/watch?v=4C0RJ01yRok
FLUX.1 [dev] Non-Commercial License — Black Forest Labs, Inc.
What This Is
Realistic, cinematic image generation using Flux Dev / Schnell / Kontext on a 4 GB VRAM laptop. Two samplers + Torch TeaCache for speed. Semi-realistic results in ~30 steps.
Model Downloads
Flux models (not VAE/CLIP):
| Model | Link |
|---|---|
| FLUX.1 Dev | https://huggingface.co/city96/FLUX.1-dev-gguf |
| FLUX.1 Schnell | https://huggingface.co/city96/FLUX.1-schnell-gguf |
| FLUX.1 Kontext | https://huggingface.co/QuantStack/FLUX.1-Kontext-dev-GGUF |
| PixelWave | https://huggingface.co/mikeyandfriends/PixelWave_FLUX.1-dev_03 |
VAE: https://huggingface.co/ffxvs/vae-flux/blob/main/ae.safetensors
CLIP:
- https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
- https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/clip_l.safetensors
Model Guide
| Model | Best for |
|---|---|
| Kontext | Quality + image editing |
| Dev | Quality generation |
| Schnell | Speed, fewer steps |
| PixelWave | Best realism |
Quantisation
| Quant | Notes |
|---|---|
| q8 | Best quality |
| q5 | Best quality/speed trade-off |
| Below q5 | Less VRAM, lower quality |
Changelog
| Version | Notes |
|---|---|
| v2.2 | Fixed settings, Torch Compile |
| v2.0 | Two-pass setup |
| v1.0 | Initial release |