Upload folder using huggingface_hub

README.md

@@ -12,48 +12,17 @@ tags:
 base_model: "black-forest-labs/FLUX.1-dev"
 pipeline_tag: text-to-image
 instance_prompt: TOK
-widget:
-- text: A TOK composite photo of a person posing at different angles, wearing a leather black jacket and graphic t-shirt, long black flaming cargo pants, blond hair, 90s, Y2K, dark eyeshadow, black boots, white background with geometric shapes sparkles, expressive emotion, different perspectives
-  output:
-    url: multi-angle-examples/2.webp
-- text: A TOK composite photo of a person posing at different angles, wearing a crop top graphic t-shirt, red baggy pants, dark eyeshadow, red converse shoes, white background with geometric shapes grunge pattern black and white color theme 
-  output:
-    url: multi-angle-examples/1.webp
-- text: A TOK composite photo of a person posing at different angles, wearing a cropped blazer with exaggerated lapels and low-slung wide-leg trousers, chunky platform sneakers, sleek high ponytail with cyber falls, Y2K tech-wear, metallic face decals, LED-embedded collar, pure white background with floating 3D geometric shapes, confident smirk, dynamic action poses
-  output:
-    url: multi-angle-examples/3.png
-- text: A TOK composite photo a person posing at multiple angles wearing a purple and yellow puffy jacket jacket with floral flame design, New York City, illustrated elements, Y2K, 90s, hiphop aesthetic, colorful, expressive emotion, blue hair 
-  output:
-    url: multi-angle-examples/4.webp
-- text: A TOK photo collage of the same person posing at multiple angles, showcasing a vibrant evolution of 90s-Y2K fashion. The central figure wears baggy JNCO-style jeans paired with a neon green crop top, transitioning to a holographic silver mini dress with platform boots. Their hair morphs from crimped pigtails to spiky frosted tips. The background explodes with geometric shapes in bold purple and teal, interspersed with bursts of glitter and floating butterfly clips. Retro tech gadgets like a transparent landline phone and a bulky laptop hover around the edges.
-  output:
-    url: multi-angle-examples/5.webp
 ---
 
-# Create unique multi angle composite photos with Y2K and 90s vibes
+# Flux Multi Angle
 
-<Gallery />
-
-
-## Model description 
-
-Use "TOK" to trigger style, multiple angle portraits, with 90s and Y2K and throwback nostalgia.
+Trained on Replicate using:
 
-You can also run the model on Replicate.
-https://replicate.com/davisbrown/flux-multi-angle
+https://replicate.com/ostris/flux-dev-lora-trainer/train
 
 
 ## Trigger words
-
-It would help if you used `in the style of TOK` for better style preservation. It is best to place the trigger words first and describe 90s or Y2K elements in your scene like clothing or expressive elements.
-
-Or prefix your prompts with `A TOK composite photo of a person posing at different angles` for better results. Experiment with other prompting approaches to get wacky and experiment.
-
-
-# Model training
-
-Trained on Replicate using:
-https://replicate.com/ostris/flux-dev-lora-trainer/train
+You should use `TOK` to trigger the image generation.
 
 
 ## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
@@ -68,8 +37,3 @@ image = pipeline('your prompt').images[0]
 ```
 
 For more details, including weighting, merging and fusing LoRAs, check the [documentation on loading LoRAs in diffusers](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters)
-
-
-## License
-
-Please adhere to the licensing terms as described [here](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md).

config.yaml

@@ -8,11 +8,11 @@ config:
     trigger_word: TOK
     network:
       type: lora
-      linear: 16
-      linear_alpha: 16
+      linear: 32
+      linear_alpha: 32
     save:
       dtype: float16
-      save_every: 1501
+      save_every: 1001
       max_step_saves_to_keep: 1
     datasets:
     - folder_path: input_images
@@ -26,7 +26,7 @@ config:
       - 1024
     train:
       batch_size: 1
-      steps: 1500
+      steps: 1000
       gradient_accumulation_steps: 1
       train_unet: true
       train_text_encoder: false
@@ -45,7 +45,7 @@ config:
       quantize: true
     sample:
       sampler: flowmatch
-      sample_every: 1501
+      sample_every: 1001
       width: 1024
       height: 1024
       prompts: []

lora.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7115280f1323837409f1b59779ed4d51d3d6b64e3bf7b3d86e0618bb1161fccd
-size 171969416
+oid sha256:3d838a63a88aeb2a7eba5034abaf1b243abc2a93681a4fc97f1187be987b593b
+size 343805456