Upload folder using huggingface_hub

README.md

@@ -1,59 +1,35 @@
 ---
-license: apache-2.0
-
+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
 tags:
-- text-to-image
-- template:sd-lora
 - flux
-- lora
-- flux dev
-- image-generation
 - diffusers
-- photo
+- lora
+- replicate
+base_model: "black-forest-labs/FLUX.1-dev"
 pipeline_tag: text-to-image
-emoji: 🔜
-language:
-- en
-base_model: black-forest-labs/FLUX.1-dev
-
-instance_prompt: HST autochrome photo
-
-widget:
-  - text: HST style photo of a green-eyed cat, centered title text HISTORIC COLOR DEV
-    output:
-      url: hstdev2.png
-  - text: autochrome HST style photo of a green-eyed cat, centered title text HISTORIC COLOR DEV
-    output:
-      url: hstdev5.png
-
+# widget:
+#   - text: >-
+#       prompt
+#     output:
+#       url: https://...
+instance_prompt: HST
 ---
 
-# Soonr Flux HST IV: HISTORIC COLOR 2 Dev
-A Dev version of our antique color photography LoRA for [FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-Dev). 
-See our [Schnell Version](/AlekseyCalvin/historic_color_schnell/tree/main) for a somewhat more explicit iteration of this adapter.
-Trained by A.C.T. Soon® for 6000 steps, using a very low learning rate, on one A100 via Colab Pro, using an AI Toolkit notebook by Ostris. 
-While our Schnell version of this LoRA was trained on a relatively large archive, 300 images, for the Dev variation we used a slightly smaller selection of high quality restored choice images from an expanded data set. Historic Color 3 will have yet another variant, trained on highest quality scans of original negatives.
-This data set, used for both models, consists of a selection from a remarkable and unique collection of color photographs taken during the 1900s and 1910s by Sergey Prokudin-Gorsky, who traveled and photographed widely in those years while pioneering and perfecting implementations of an early three-color-composite photography technique. 
-We urge you to explore the work of Prokudin-Gorsky for yourself, at the wonderfully organized online [archive at this link](https://prokudin-gorsky.org/), featuring many hundreds of high quality downloadable scans of composite color photo prints from the photographer's original glass plate negatives, available at this site alongside relatively recent restorations of a substantial portion of the images. The original glass-plate negatives are currently held at and administrated by the Library of Congress in Washington, DC, USA. 
-
-## Trigger words
-You should use `HST` or 'HST style' to trigger the image generation.
+# Historic_Color_Dev
 
-- base model: [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev)
+<!-- <Gallery /> -->
 
-<Gallery />
+Trained on Replicate using:
 
-## Historical Note
-Prokudin-Gorsky's color photography technique would involve three photo-exposures, either simultaneous or sequential, using specialized color-spectrum filters (basically R.B.G.: red, blue, and green), rendering the same subject/shot onto glass plates covered with light-emulsive mixture.  Prokudin-Gorsky's focus on refining the developer and filter quality, in tandem with his incessant and wide-ranging experimentation, and his persistent usage of glass plates (unwieldly and increasingly old-fashioned, but elsewise extra reliable) ultimately led him to produce a color photography oeuvre of much greater fidelity and vividness than achieved by most of his contemporaries. 
-At the same time, the peculiarities of the photographer's method, coupled with his exceptionally hands-on execution thereof, would manifest in a range of idyosyncratic color, light, and motion artifacts common across the resulting prints. Seldom marring the image as a whole, and less grave than the weaknesses of some contemporenously emerging autochrome techniques, the warm color hazes and flares framing many of Prokudin-Gorsky's prints may be seen as a kind of ephemeral signature.
-Alongside some of the more subtle chromatic, textural, and (in some measure) figural characteristics of his work, these auras have imprinted themselves into this Flux LoRA, the fourth in our series of historical adapters for Flux. 
+https://replicate.com/ostris/flux-dev-lora-trainer/train
 
-![HST style autochrome photo of a dark koala building a hut in snowy mountains](hstdev1.jpg)
 
-## Download model
-
-Weights for this model are available in Safetensors format.
-[Download](/AlekseyCalvin/historic_color_dev/tree/main) them in the Files & versions tab.
+## Trigger words
+You should use `HST` to trigger the image generation.
 
 
 ## Use it with the [🧨 diffusers library](https://github.com/huggingface/diffusers)
@@ -62,11 +38,9 @@ Weights for this model are available in Safetensors format.
 from diffusers import AutoPipelineForText2Image
 import torch
 
-pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.bfloat16).to('cuda')
-pipeline.load_lora_weights('AlekseyCalvin/historic_color_dev')
-image = pipeline('HST style photo of a cat').images[0]
-image.save("my_image.png")
+pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.float16).to('cuda')
+pipeline.load_lora_weights('alekseycalvin/historic_color_dev', weight_name='lora.safetensors')
+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)
-

config.yaml

@@ -1,87 +1,104 @@
-job: extension
+job: custom_job
 config:
-  name: Historic_Color_DEV
+  name: flux_train_replicate
   process:
-  - type: sd_trainer
-    training_folder: /content/output
+  - type: custom_sd_trainer
+    training_folder: output
     device: cuda:0
+    trigger_word: HST
     network:
       type: lora
-      linear: 16
-      linear_alpha: 16
+      linear: 128
+      linear_alpha: 128
+      network_kwargs:
+        only_if_contains:
+        - transformer.transformer_blocks.0.norm1.linear
+        - transformer.transformer_blocks.0.norm1_context.linear
+        - transformer.transformer_blocks.0.attn.to_q
+        - transformer.transformer_blocks.0.attn.to_k
+        - transformer.transformer_blocks.0.attn.to_v
+        - transformer.transformer_blocks.0.attn.add_k_proj
+        - transformer.transformer_blocks.0.attn.add_v_proj
+        - transformer.transformer_blocks.0.attn.add_q_proj
+        - transformer.transformer_blocks.0.attn.to_out.0
+        - transformer.transformer_blocks.0.attn.to_add_out
+        - transformer.transformer_blocks.0.ff.net.0.proj
+        - transformer.transformer_blocks.0.ff.net.2
+        - transformer.transformer_blocks.0.ff_context.net.0.proj
+        - transformer.transformer_blocks.0.ff_context.net.2
+        - transformer.transformer_blocks.2.norm1.linear
+        - transformer.transformer_blocks.2.norm1_context.linear
+        - transformer.transformer_blocks.2.attn.to_q
+        - transformer.transformer_blocks.2.attn.to_k
+        - transformer.transformer_blocks.2.attn.to_v
+        - transformer.transformer_blocks.2.attn.add_k_proj
+        - transformer.transformer_blocks.2.attn.add_v_proj
+        - transformer.transformer_blocks.2.attn.add_q_proj
+        - transformer.transformer_blocks.2.attn.to_out.0
+        - transformer.transformer_blocks.2.attn.to_add_out
+        - transformer.transformer_blocks.2.ff.net.0.proj
+        - transformer.transformer_blocks.2.ff.net.2
+        - transformer.transformer_blocks.2.ff_context.net.0.proj
+        - transformer.transformer_blocks.2.ff_context.net.2
+        - transformer.transformer_blocks.18.norm1.linear
+        - transformer.transformer_blocks.18.norm1_context.linear
+        - transformer.transformer_blocks.18.attn.to_q
+        - transformer.transformer_blocks.18.attn.to_k
+        - transformer.transformer_blocks.18.attn.to_v
+        - transformer.transformer_blocks.18.attn.add_k_proj
+        - transformer.transformer_blocks.18.attn.add_v_proj
+        - transformer.transformer_blocks.18.attn.add_q_proj
+        - transformer.transformer_blocks.18.attn.to_out.0
+        - transformer.transformer_blocks.18.attn.to_add_out
+        - transformer.transformer_blocks.18.ff.net.0.proj
+        - transformer.transformer_blocks.18.ff.net.2
+        - transformer.transformer_blocks.18.ff_context.net.0.proj
+        - transformer.transformer_blocks.18.ff_context.net.2
     save:
       dtype: float16
-      save_every: 1000
-      max_step_saves_to_keep: 4
+      save_every: 501
+      max_step_saves_to_keep: 1
     datasets:
-    - folder_path: /content/dataset
+    - folder_path: input_images
       caption_ext: txt
       caption_dropout_rate: 0.05
       shuffle_tokens: false
-      cache_latents_to_disk: true
+      cache_latents_to_disk: false
+      cache_latents: true
       resolution:
       - 512
       - 768
       - 1024
     train:
       batch_size: 1
-      steps: 3000
+      steps: 500
       gradient_accumulation_steps: 1
       train_unet: true
       train_text_encoder: false
-      content_or_style: style
+      content_or_style: balanced
       gradient_checkpointing: true
       noise_scheduler: flowmatch
       optimizer: adamw8bit
-      lr: 0.0001
-    skip_first_sample: true
-      linear_timesteps: true
+      lr: 0.0008
       ema_config:
         use_ema: true
         ema_decay: 0.99
       dtype: bf16
     model:
-      name_or_path: black-forest-labs/FLUX.1-dev
+      name_or_path: FLUX.1-dev
       is_flux: true
       quantize: true
     sample:
       sampler: flowmatch
-      sample_every: 1000
+      sample_every: 501
       width: 1024
       height: 1024
-      prompts:
-      - HST style photo of Lenin in 1924 surrounded by two other individuals on either
-        side of him. Film photograph, three-quarter length, from front. Lenin in the
-        center, wearing a white outfit sits in a wicker wheelchair, wearing a white
-        shirt and flat cap. His eyes look intense and slightly demented. He seems
-        convalescent and ill following a stroke. A woman in a white dress with a black
-        belt stands to the left, leaning forward with her hands on the back of the
-        wheelchair.
-      - HST style Arthur Rimbaud. He is dressed in a dark suit with a white shirt
-        and a maroon tie with small white polka dots. He is looking to his right with
-        an inquisitive expression.
-      - HST nightclub in 1921 USSR,, fish eye lens, smoke machine, lazer lights, Trotsky
-        holding a martini
-      - HST style Andrey Beliy showing off his cool new poetry books at the beach
-        in 1921 USSR, a shark is jumping out of the water to eat the books, Beliy
-        screams in surprise
-      - HST style Alexey Khvostenko and Anri Volokhonskiy and a bear collaborate building
-        a log cabin in the snow covered mountains
-      - HST style Viktor Tsoy soulfully playing the guitar, on stage in 1921 USSR,
-        singing a song, laser lights, punk rocker
-      - HST style Lawrence from Felt
-      - HST style Medium-frame photo of Robert Duncan the poet sitting in his office,
-        wearing a dark suit and tie. Bust view, facing forward. Sitting with hands
-        resting, neutral facial expression. Bookshelves filled with books in the background.
-        Wooden furniture in the surrounding space.
-      - HST style Lenin in cap and Bolshevik suit holding a sign with text, 'World
-        workers, all our eyes!'
-      - HST style Egor Letov in a leather jacket, in a desert, on a motorcycle
+      prompts: []
       neg: ''
       seed: 42
       walk_seed: true
       guidance_scale: 3.5
-      sample_steps: 20
+      sample_steps: 28
 meta:
-  name: Historic_Color_DEV
+  name: flux_train_replicate
   version: '1.0'

lora.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f805a984ab86d51ed715bf3ce92440d1a0ef8750d03415cf9cd3f70a8b72f32c
+size 117976536