Runing Image Generator On Akash (akash-network#419)

AI-Image-App/Dockerfile

@@ -0,0 +1,20 @@
+FROM pytorch/pytorch:latest
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+      git && \
+    rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+COPY Pipfile Pipfile.lock .
+RUN pip install pipenv && \
+    pipenv install --system
+
+COPY scripts scripts
+COPY core core
+
+VOLUME /data
+
+ENV FLASK_APP=scripts/web.py
+
+CMD ["flask", "run", "--port", "3000", "--host", "0.0.0.0"]

AI-Image-App/README.md

@@ -0,0 +1,52 @@
+# AI-Art
+
+![AI-Art Screenshot](./Screenshot(234).png)
+
+Art generation using VQGAN + CLIP using docker containers. A simplified,
+updated, and expanded upon version of
+[Kevin Costa's work](https://github.com/kcosta42/VQGAN-CLIP-Docker). This
+project tries to make generating art as easy as possible for anyone with a GPU
+by providing a simple web UI.
+
+
+## Samples
+
+For samples check out the [AI Generated](http://1j0gql7905agf792r45ae0hl68.ingress.akashost.com/) section
+on the art page on website.
+
+
+## Using ai-art
+
+This works best if you have an NVIDIA GPU however there is a fallback CPU mode included. I've found the CPU mode to take significantly longer than even the most basic of GPUs though.
+
+
+### Quick start usage
+For running just deploy [this SDL](https://github.com/alfset/awesome-akash/blob/master/AI-Image-App/deploy.yaml) and wait for the models to finish loading, watching the process in the logs.
+
+### Logs
+
+![Checking Logs](./Screenshot(232).png)
+
+### Check GPU resources has been used
+
+![Checking GPU resources](./Screenshot(233).png)
+
+## Image sizes
+
+The larger the image the more VRAM your graphics card needs:
+
+- 6 GB of VRAM is required to generate 256x256 images.
+- 12 GB of VRAM is required to generate 512x512 images.
+- 24 GB of VRAM is required to generate 1024x1024 images.
+
+If you don't know how much VRAM your graphics card has you probably have 6 GB
+or less so stick with smaller images.
+
+That being said you can do non-square images if you want as long as you don't
+go above the number of pixels your GPU's VRAM supports, for example you could
+do ultrawide images with 6 GB of ram at "384x128" or do tall images at "128x384"
+and so on. You do not have to use numbers with a power of 2, "300x100" is also
+perfectly valid.
+
+## Demo Video
+https://github.com/alfset/awesome-akash/blob/master/AI-Image-App/ai-image-app_screen-capture.webm

AI-Image-App/deploy.yaml

@@ -0,0 +1,36 @@
+##sdl yaml
+version: '2.0'
+services:
+  gpu-app:
+    image: overshard/ai-art
+    expose:
+      - port: 3000
+        as: 80
+        to:
+          - global: true
+profiles:
+  compute:
+    gpu-app:
+      resources:
+        cpu:
+          units: 1
+        memory:
+          size: 38Gi
+        gpu:
+          units: 1
+          attributes:
+            vendor:
+              nvidia:
+        storage:
+          - size: 100Gi
+  placement:
+    westcoast:
+      pricing:
+        gpu-app:
+          denom: uakt
+          amount: 100000
+deployment:
+  gpu-app:
+    westcoast:
+      profile: gpu-app
+      count: 1

README.md

@@ -36,6 +36,7 @@ Also, follow [@akashnet\_](https://twitter.com/akashnet_) to stay in the loop wi
 - [Weaviate](weaviate)
 - [Flan-T5 XXL](flan-t5-xxl)
 - [Falcon-7B](Falcon-7B)
+- [AI-Image-App](AI-Image-App)
 
 ### Blogging