AI Image Generation: Expanding Paintings with DALL-E 2

Table of Contents

• Introducing DALL-E 2 AI Image Generation
• Generating Additional Sections of the Painting
• Reviewing and Selecting the Best AI Outputs
• Potential Enhancements Using Photoshop
• Saving and Exporting the Final Image
• Conclusion and Next Steps

Introducing DALL-E 2 AI Image Generation

DALL-E 2 is an artificial intelligence system from OpenAI that can generate images from text descriptions. It builds on top of DALL-E, an earlier version, and is named after the surrealist artist Salvador Dali. DALL-E 2 has a feature called 'outpainting' which can take an existing image and expand or continue it by generating additional sections.

For example, I had an existing oil painting that was a landscape scene. By uploading it to DALL-E 2 and selecting the outpainting feature, the AI analyzed my painting and then generated extensions of the landscape to the north, south, east and west. It tries to match colors, styles, and content so that the generated sections blend seamlessly with the original.

The results can be surprisingly realistic and detailed. However, some fine-tuning is usually required, either by iterating with different text prompts or by making adjustments in Photoshop afterwards. But it can save artists a tremendous amount of time compared to having to paint an entire large, complex scene by hand from scratch.

Understanding Outpainting

With outpainting, you upload an existing image file such as a JPEG. DALL-E 2 analyzes it, detects patterns, colors, styles etc. and tries to continue them in a sensible way in the expanded regions. You position a frame that overlaps your original image slightly. This frame signals where DALL-E 2 should generate the new content. So if you drag the frame north, it will generate new landscape extending upwards. The slight overlap gives it some existing data right next to the new area, helping it transition smoothly.

Using DALL-E 2 on an Existing Painting

I uploaded a JPEG file of one of my recent oil paintings - a landscape scene for a video game background or 'overworld map'. With the outpainting frame, I extended the image to the left. Based on colors and content from the original painting visible in the overlap region, DALL-E 2 generated a continuation of the landscape, with a cloudy sky, trees, mountains and an urban city area. It created 4 variations, with different art styles for the city section, allowing me to pick my favorite. I ended up really impressed with how it matched the color palette and 'feel' of my original painting content on the right side of the frame.

Generating Additional Sections of the Painting

After expanding the left side of my painting with the city landscape, I wanted to also extend the right side of the image. Here there was a pathway leading into some woods, suggesting another distinct area to explore.

So I positioned the outpainting frame on the right side with a slight overlap, and supplied the text prompt: "A massive, rolling forest as far as the eye can see. There should be a magical quality to it." This makes clear my desire to have an enchanted, vibrant wooded landscape stretch far into the distance.

DALL-E 2 did an admirable job generating variou forest scenes, and once again I chose the second variation output. The color palette, lighting, and overall mystical mood aligned very closely with my original painting's style.

Reviewing and Selecting the Best AI Outputs

Each time I used the outpainting feature, DALL-E 2 provided four different image variation outputs. I noticed that often the second option was strongest in accurately continuing patterns, colors and content from my uploaded source painting.

It seems like the first output is more experimental, while subsequent iterations better integrate what the AI detected from analyzing the original input image. In any case, having four options allowed me to quickly compare and find the one whose new generated areas blended most seamlessly with my existing artwork.

Potential Enhancements Using Photoshop

While the AI did an impressive job of outpainting my landscape and matching the original stylistic elements, there is still ample room for improvement. As an artist and game developer, I can take DALL-E 2's outputs into Photoshop for refinement.

For example, I may adjust colors that came out too bright or garish. Or erase certain distracting elements it added like towers and antennas. I could also paint over seams between the original and outpainted areas. This combination of AI generation and human artistic polish helps realize my creative vision efficiently.

Saving and Exporting the Final Image

Once satisfied with the final image, expanded and enhanced thanks to DALL-E 2's outpainting capabilities, I saved the high resolution JPEG file with improvements from Photoshop. At over 30 inches wide, this epic landscape artwork has so much detail that zooming in almost feels like exploring a miniature open world.

Thanks to the AI kickstart, I have an intricate background scene full of magical forests, twilight skies, and mysterious towering cities. It would have taken an immense amount of time to manually paint. Now I can incorporate it into my video game concept as an immersive, bird's eye environment for players to adventure through.

Conclusion and Next Steps

In conclusion, experimenting with DALL-E 2's outpainting opened my eyes to the huge potential for AI assistance in content creation workflows. Rather than fearing it will replace artists and creators, I see how AI image generation can amplify our skills while saving tremendous amounts of time and effort.

My next step is to keep iterating - I may expand this cyberpunk magical landscape even further, or try outpainting portraits and character art. Endless creative possibilities have emerged from this technology demonstration. I look forward to seeing how AI image generation matures in the years ahead!

FAQ

Q: What is DALL-E 2 used for?
A: DALL-E 2 is an AI system from OpenAI that can generate realistic images and art from text descriptions. It can expand and continue existing images through a technique called outpainting.

Q: How does outpainting work?
A: Outpainting involves providing DALL-E 2 with a section of an existing image, then prompting it to generate the areas around that section, effectively expanding the image.

Q: What kind of touch-ups can be done?
A: The AI-generated sections may not perfectly match the original artwork, so using Photoshop to smooth out edges, adjust colors, add contrast, etc. can greatly improve the final expanded image.

Q: What file formats can the images be exported to?
A: The DALL-E 2 interface allows exporting generated images in standard formats like JPG, PNG, SVG that can then be used in other programs.