Introduction
## AI Image Generation Can Enhance Old Game Graphics. AI image generation tools like Stable Diffusion, DALL-E, and Midjourney can create higher-resolution and visually richer reinterpretations of old video game graphics, though each has unique strengths and limitations. By experimenting with the intro cinematic of the 1987 game Nemesis 2 on the MSX, these tools show promise in retelling nostalgic visual stories with modern fidelity.
Comparing Stable Diffusion for Graphic Recreation
Stable Diffusion, accessed via Dream Studio, excels in replicating specific image details through iterative prompt tuning. Generating over 30 images per panel was necessary to refine the visuals, revealing the challenge of crafting effective prompts. For example, a prompt like “fighter jets flying over a red planet in space with stars in the black sky” led to varied results until style keywords were added. Stable Diffusion’s open source nature and API availability make it highly adaptable, though older versions (1.4 and 1.5) currently produce better results than newer releases. The interface balances advanced options like in-painting with usability but lacks robust generation history.
Midjourney Offers Superior Visual Quality With Less Effort
Midjourney consistently produces the most visually stunning images with minimal prompt adjustments, making it ideal for generating multiple story panels quickly. For example, Midjourney generated a realistic portrait of Dr. Venom with dramatic lighting and a green-skinned appearance using a concise prompt. It also rendered detailed spaceship schematics and starmaps effectively, although it struggles with consistent text reproduction and exact element placement. Its user interface automatically archives generated images, which is beneficial for project management. Midjourney’s version 4 model delivers high-quality images with a 3: 2 aspect ratio command, streamlining wide-format generation.
DALL E’s
DALL-E’s Outpainting Extends Image Canvases Creatively. DALL-E’s outpainting feature allows expanding the canvas beyond the original image boundary, filling surrounding space with coherent content. This is particularly useful for scenes requiring broader context or environment extension, as demonstrated by enlarging a space pilot’s portrait from Nemesis
2. Outpainting requires modifying prompts to fit each image segment, differing from typical text-to – image generation. However, DALL-E struggles with text clarity inside images, limiting its use for graphics needing precise textual elements.
Challenges in Reproducing Specific Visual Elements
Recreating iconic or complex features, such as Dr. Venom’s three eyes, remains difficult. Neither Stable Diffusion, Midjourney, nor DALL-E reliably generated this detail despite in-painting attempts. This highlights current limitations in controlling nuanced image parts. Additionally, AI tools do not yet handle accurate text rendering inside images well, requiring manual editing to add elements like star map labels.
Practical Workflow and Prompt Engineering Insights
Success with AI image generation depends heavily on prompt engineering. Searching prompt galleries like Lexica was essential to discover effective style keywords and guide model outputs. For example, combining subject description with stylistic terms like “realistic scifi spaceship, ” “dramatic lighting, ” and “vintage retro scifi” improved results. Aspect ratio commands ( – – ar 3: 2) helped produce images fitting the original cinematic panels. The process involved generating multiple candidates and selecting the closest match, with some trial and error for poses and compositions.
Summary of Tool Strengths and Limitations
Stable Diffusion’s Dream Studio is versatile and programmable with an API, good for prototyping and fine-tuning but requires prompt mastery and lacks full history tracking. Midjourney offers superior image quality and a user-friendly archive but cannot guarantee exact reproducibility of specific objects across panels. DALL-E’s outpainting extends compositions creatively but struggles with text and precise element manipulation. These AI tools collectively enable reimagining classic video game graphics in higher resolution, though users must navigate prompt complexity, partial feature reproduction, and manual edits to fully realize nostalgic visuals. As improvements continue, AI image generation is becoming a practical method for enhancing retro game art with quantitative evidence of success in multiple panel recreations.
