Uploaded example for stable diffusion

examples/stable-diffusion/Dockerfile

@@ -0,0 +1,21 @@
+FROM nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04
+
+RUN apt update && \
+    apt install -y --no-install-recommends git wget python3-pip && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN git clone https://github.com/srisco/stable-diffusion-tensorflow.git
+
+WORKDIR stable-diffusion-tensorflow
+
+RUN pip install -r requirements.txt && \
+    rm -rf /root/.cache/pip/* && \
+    rm -rf /tmp/*
+
+# DOWNLOAD WEIGHTS
+RUN mkdir -p /root/.keras/datasets && \
+    wget https://huggingface.co/fchollet/stable-diffusion/resolve/main/text_encoder.h5 -O /root/.keras/datasets/text_encoder.h5 && \
+    wget https://huggingface.co/fchollet/stable-diffusion/resolve/main/diffusion_model.h5 -O /root/.keras/datasets/diffusion_model.h5 && \
+    wget https://huggingface.co/fchollet/stable-diffusion/resolve/main/decoder.h5 -O /root/.keras/datasets/decoder.h5 && \
+    wget https://huggingface.co/divamgupta/stable-diffusion-tensorflow/resolve/main/encoder_newW.h5 -O /root/.keras/datasets/encoder_newW.h5

examples/stable-diffusion/README.md

@@ -0,0 +1,31 @@
+# Stable Diffusion
+
+This example is based on a Keras / Tensorflow implementation of Stable Diffusion. The following repositories were used for the creation of the image:
+
+* [srisco/stable-diffusion-tensorflow](https://github.com/srisco/stable-diffusion-tensorflow)
+* [huggingface.co/fchollet/stable-diffusion](https://huggingface.co/fchollet/stable-diffusion)
+* [huggingface.co/divamgupta/stable-diffusion-tensorflow](https://huggingface.co/divamgupta/stable-diffusion-tensorflow/)
+
+The image if pushed to a public github registry [here](ghcr.io/grycap/stable-diffusion-tf:latest) but you can see the Dockerfile that generates it [here](Dockerfile).
+
+## Deploy an OSCAR cluster
+Follow the instructions in the documentation for your desired IaaS cloud provider.
+[See Deployment](https://docs.oscar.grycap.net/)
+
+## Create the OSCAR Service
+
+The Service can be created using the OSCAR GUI by providing the [FDL](stable-diff.yaml) and the [script.sh](script.sh) file.
+
+![OSCAR GUI Creation of a service](https://oscar.grycap.net/images/blog/post-20210803-1/create_service_gui.png)
+
+## Upload the input file to the MinIO bucket
+
+Once the service is created, you can upload the input file to the bucket. The input file should be a file containing the prompt that you want to process.
+
+For example, using the following prompt:
+
+`a chicken making a 360 with a skateboard with background flames`
+
+The following image is generated:
+
+![Cool chicken](prompt.txt.png)

examples/stable-diffusion/prompt.txt

@@ -0,0 +1 @@
+a chicken making a 360 with a skateboard with background flames

examples/stable-diffusion/script.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+echo "SCRIPT: Invoked stable diffusion text to image."
+FILE_NAME=`basename "$INPUT_FILE_PATH"`
+OUTPUT_FILE="$TMP_OUTPUT_DIR/$FILE_NAME.png"
+
+prompt=`cat "$INPUT_FILE_PATH"`
+echo "SCRIPT: Converting input prompt '$INPUT_FILE_PATH' to image :)"
+python3 text2image.py --prompt="$prompt" --output=$OUTPUT_FILE

examples/stable-diffusion/stable-diff.yaml

@@ -0,0 +1,17 @@
+functions:
+  oscar:
+  - oscar-intertwin:
+      name: stable-diffusion-tf
+      memory: 16Gi
+      cpu: '4'
+      image: ghcr.io/grycap/stable-diffusion-tf:latest
+      script: script.sh
+      log_level: DEBUG
+      vo: "vo.example.eu"
+      allowed_users: []
+      input:
+      - storage_provider: minio.default
+        path: stablediff/input
+      output:
+      - storage_provider: minio.default
+        path: stablediff/output