import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/flux-3-image-klein-base", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("SimpleTuner/flux-3-klein-image-LoRA-experiments")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]FLUX.3 Klein Image: A LoRA Experiment Journal
This journal follows three FLUX.3 Klein training experiments: learning Domokun at different resolutions, training a broad-data assistant LoRA, and using that assistant to finetune distilled Klein on photographic data.
The recipes build on the Qwen Image 2.1 experiments and their v3 training assistant.
Contents
- Training setup
- Domokun: three resolution controls
- A real-data-only training assistant
- Photo aesthetics with an assistant
- Progress and runtime notes
Each chapter includes its settings, checkpoint links, and previews. The complete recipes and run manifest are available alongside them.
Campaign At A Glance
| Run | Updates / target | Status |
|---|---|---|
| domokun-512 | 10,000 / 10,000 | Complete |
| domokun-1024 | 10,000 / 10,000 | Complete |
| domokun-512-1024 | 10,000 / 10,000 | Complete |
| training-assistant-real-v3 | 30,000 / 30,000 | Complete |
| photo-aesthetics-distilled-assisted-512 | 50,000 / 50,000 | Complete |
| photo-aesthetics-distilled-assisted-1024 | 23,601 / 50,000 | Training |
| photo-aesthetics-distilled-assisted-512-v2-kohaku-fa | 6,511 / 50,000 | Training |
| photo-aesthetics-distilled-assisted-1024-v2-kohaku-fa | 0 / 50,000 | Queued |
Updated 2026-10-16T22:48:21.704875+00:00.
Training uses BF16 transformers and regional compilation. Checkpoint previews and settings are linked from each chapter.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for SimpleTuner/flux-3-klein-image-LoRA-experiments
Base model
black-forest-labs/flux-3-image-klein