--- license: other base_model: "black-forest-labs/FLUX.1-dev" tags: - flux - flux-diffusers - text-to-image - diffusers - simpletuner - safe-for-work - lora - template:sd-lora - standard inference: true widget: - text: 'unconditional (blank prompt)' parameters: negative_prompt: 'blurry, cropped, ugly' output: url: ./assets/image_0_0.png - text: 'A girl in light blue sits at the bar counter, holding an ice-cold wine glass and drinking alone on top of the Eiffel Tower, with a night view outside the window.. It features a close-up shot of her sitting by herself. She has long hair, wears glasses, faces away from the camera, and is wearing white shoes, black pants, a gray jacket, and a green scarf. with bright colors and a Paris night background featuring the Eiffel Tower. The composition is elegant, with the woman sitting on a high stool' parameters: negative_prompt: 'blurry, cropped, ugly' output: url: ./assets/image_1_0.png --- # jazzy-st-2211 This is a standard PEFT LoRA derived from [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev). The main validation prompt used during training was: ``` A girl in light blue sits at the bar counter, holding an ice-cold wine glass and drinking alone on top of the Eiffel Tower, with a night view outside the window.. It features a close-up shot of her sitting by herself. She has long hair, wears glasses, faces away from the camera, and is wearing white shoes, black pants, a gray jacket, and a green scarf. with bright colors and a Paris night background featuring the Eiffel Tower. The composition is elegant, with the woman sitting on a high stool ``` ## Validation settings - CFG: `3.0` - CFG Rescale: `0.0` - Steps: `20` - Sampler: `FlowMatchEulerDiscreteScheduler` - Seed: `42` - Resolution: `1080x1920` - Skip-layer guidance: Note: The validation settings are not necessarily the same as the [training settings](#training-settings). You can find some example images in the following gallery: The text encoder **was not** trained. You may reuse the base model text encoder for inference. ## Training settings - Training epochs: 6 - Training steps: 5000 - Learning rate: 0.0004 - Learning rate schedule: polynomial - Warmup steps: 100 - Max grad norm: 2.0 - Effective batch size: 1 - Micro-batch size: 1 - Gradient accumulation steps: 1 - Number of GPUs: 1 - Gradient checkpointing: True - Prediction type: flow-matching (extra parameters=['shift=3', 'flux_guidance_mode=constant', 'flux_guidance_value=1.0', 'flow_matching_loss=compatible', 'flux_lora_target=all']) - Optimizer: adamw_bf16 - Trainable parameter precision: Pure BF16 - Caption dropout probability: 10.0% - LoRA Rank: 32 - LoRA Alpha: 32.0 - LoRA Dropout: 0.1 - LoRA initialisation style: default ## Datasets ### jazzy-512 - Repeats: 10 - Total number of images: 23 - Total number of aspect buckets: 1 - Resolution: 0.262144 megapixels - Cropped: False - Crop style: None - Crop aspect: None - Used for regularisation data: No ### jazzy-768 - Repeats: 10 - Total number of images: 23 - Total number of aspect buckets: 1 - Resolution: 0.589824 megapixels - Cropped: False - Crop style: None - Crop aspect: None - Used for regularisation data: No ### jazzy-1024 - Repeats: 10 - Total number of images: 23 - Total number of aspect buckets: 1 - Resolution: 1.048576 megapixels - Cropped: False - Crop style: None - Crop aspect: None - Used for regularisation data: No ## Inference ```python import torch from diffusers import DiffusionPipeline model_id = 'black-forest-labs/FLUX.1-dev' adapter_id = 'linhqyy/jazzy-st-2211' pipeline = DiffusionPipeline.from_pretrained(model_id), torch_dtype=torch.bfloat16) # loading directly in bf16 pipeline.load_lora_weights(adapter_id) prompt = "A girl in light blue sits at the bar counter, holding an ice-cold wine glass and drinking alone on top of the Eiffel Tower, with a night view outside the window.. It features a close-up shot of her sitting by herself. She has long hair, wears glasses, faces away from the camera, and is wearing white shoes, black pants, a gray jacket, and a green scarf. with bright colors and a Paris night background featuring the Eiffel Tower. The composition is elegant, with the woman sitting on a high stool" ## Optional: quantise the model to save on vram. ## Note: The model was not quantised during training, so it is not necessary to quantise it during inference time. #from optimum.quanto import quantize, freeze, qint8 #quantize(pipeline.transformer, weights=qint8) #freeze(pipeline.transformer) pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level image = pipeline( prompt=prompt, num_inference_steps=20, generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42), width=1080, height=1920, guidance_scale=3.0, ).images[0] image.save("output.png", format="PNG") ```