Upload 8 files

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+samples/HDRI[[:space:]]panoramic[[:space:]]view[[:space:]]of[[:space:]]TOK,[[:space:]]inside[[:space:]]of[[:space:]]a[[:space:]]beautiful[[:space:]]japanese[[:space:]]onsen[[:space:]]with[[:space:]]sakura[[:space:]]tree,[[:space:]]hot[[:space:]]baths.webp filter=lfs diff=lfs merge=lfs -text
+samples/HDRI[[:space:]]panoramic[[:space:]]view[[:space:]]of[[:space:]]TOK,[[:space:]]inside[[:space:]]of[[:space:]]an[[:space:]]art[[:space:]]gallery,[[:space:]]very[[:space:]]modern,[[:space:]]luxurious[[:space:]]stylish,[[:space:]]with[[:space:]]expensive[[:space:]]marble[[:space:]]and[[:space:]]copper-plated[[:space:]]stuff.webp filter=lfs diff=lfs merge=lfs -text
+samples/HDRI[[:space:]]panoramic[[:space:]]view[[:space:]]of[[:space:]]TOK,[[:space:]]visiting[[:space:]]an[[:space:]]amusement[[:space:]]park[[:space:]]about[[:space:]]harry[[:space:]]potter.webp filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,59 @@
+---
+license: other
+license_name: flux-1-dev-non-commercial-license
+license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
+language:
+- en
+library_name: diffusers
+pipeline_tag: text-to-image
+base_model: black-forest-labs/FLUX.1-dev
+tags:
+- diffusers
+- lora
+- flux
+inference:
+  parameters:
+    width: 1536
+    height: 640
+widget:
+- text: "HDRI panoramic view of TOK, visiting an amusement park about harry potter."
+  output:
+    url: "samples/HDRI panoramic view of TOK, visiting an amusement park about harry potter.webp"
+- text: "HDRI panoramic view of TOK, inside of a beautiful japanese onsen with sakura tree, hot baths"
+  output:
+    url: "samples/HDRI panoramic view of TOK, inside of a beautiful japanese onsen with sakura tree, hot baths.webp"
+- text: "HDRI panoramic view of TOK, inside of an art gallery, very modern, luxurious stylish, with expensive marble and copper-plated stuff"
+  output:
+    url: "samples/HDRI panoramic view of TOK, inside of an art gallery, very modern, luxurious stylish, with expensive marble and copper-plated stuff.webp"
+- text: "HDRI panoramic view of TOK, inside of a fighter jet cockpit, flying in the air, over an aircraft carrier, in San Francisco bay"
+  output:
+    url: "samples/HDRI panoramic view of TOK, inside of a fighter jet cockpit, flying in the air, over an aircraft carrier, in San Francisco bay.webp"
+---
+
+<Gallery />
+
+# FLUX.1-[dev] Panorama LoRA (v2)
+
+A LoRA model to generate panoramas using Flux dev.
+
+## Which image ratio to use?
+
+This model has been trained on images with a 2:1 ratio (2048x1024).
+
+So you might get good results is you use this for width and height.
+
+However panorama viewers are pretty flexible when it comes to resolution, and FLUX.1 seems to generalize well.
+
+For instance the gallery samples have been generated in 1536 × 640 (~21:9), since this is reasonably fast (16 sec on the HF Inference API).
+
+## It doesn't work for case X or Y
+
+It usually work fine for "normal" requests, but the model might have trouble creating the spherical distortion if you ask for uncommon content, locations or angles.
+
+If you give me some examples, maybe I can try to find more data to better cover uncommon panoramas.
+
+## Non-commercial use
+
+As the base model is FLUX.1-[dev] and since the data comes from Google Street View, it should be used for [non-commercial, personal or demonstration purposes only](https://github.com/black-forest-labs/flux/blob/main/model_licenses/LICENSE-FLUX1-dev).
+
+Please use it responsibly, thank you!

config.yaml

@@ -0,0 +1,60 @@
+config:
+  name: flux_train_replicate
+  process:
+  - datasets:
+    - cache_latents_to_disk: true
+      caption_dropout_rate: 0.05
+      caption_ext: filename
+      folder_path: input_images
+      resolution:
+      - 512
+      - 768
+      - 1024
+      shuffle_tokens: false
+    device: cuda:0
+    model:
+      is_flux: true
+      name_or_path: black-forest-labs/FLUX.1-dev
+      quantize: true
+    network:
+      linear: 16
+      linear_alpha: 16
+      type: lora
+    sample:
+      guidance_scale: 4
+      height: 1024
+      neg: ''
+      prompts:
+      - a sign that says 'I LOVE PROMPTS!' in the style of [trigger]
+      sample_every: 250
+      sample_steps: 20
+      sampler: flowmatch
+      seed: 42
+      walk_seed: true
+      width: 1024
+    save:
+      dtype: float16
+      max_step_saves_to_keep: 1
+      save_every: 2001
+    train:
+      batch_size: 1
+      content_or_style: balanced
+      dtype: bf16
+      ema_config:
+        ema_decay: 0.99
+        use_ema: true
+      gradient_accumulation_steps: 1
+      gradient_checkpointing: true
+      lr: 0.0004
+      noise_scheduler: flowmatch
+      optimizer: adamw8bit
+      steps: 2000
+      train_text_encoder: false
+      train_unet: true
+    training_folder: output
+    trigger_word: TOK
+    type: sd_trainer
+job: extension
+meta:
+  name: flux_train_replicate
+  version: '1.0'

flux_train_replicate.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dea506dc7b0923ee5e693449c122d8ecad700c54d78d7096a74d43c99d4a398e
+size 171969416

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09c98bb0b2280145810efaf224f4cce47d55b5fcfae123ebeac4276b0498c9d3
+size 173272836