Image Battle

Summary for ChatGPT 4o

ChatGPT 4o distinguishes itself as a highly capable but strictly guarded professional tool. It secures a strong position on the leaderboard with an overall score of 7.62. Its standout feature is its exceptional command of text and graphic design, often scoring perfect 10s. However, users must navigate a high refusal rate (11%) triggered by strict copyright and safety policies. While it excels at photorealism in controlled portraits, it occasionally hallucinates captions or defaults to illustration styles in complex scenes.

✅ Quick Verdict: The go-to model for Typography, Logos, and Marketing Assets, but restrictive for Fan Art and Specific Architectural Reproductions.

⭐ Key Insights & Patterns

• The Typography Titan: This model is unrivaled in Text in Images and Graphic Design. Generations like the Sustainable Coffee Logo and Spring Sale Post demonstrate a pixel-perfect ability to integrate specific text into complex aesthetics without the gibberish common in other models.

• The "Safety" Wall: With 11 refusals, ChatGPT 4o is the most restrictive model in the set. It consistently blocks requests involving:

  • Children: Toddler Portrait was refused.
  • Copyrighted Styles: Direct requests for Kiki's Delivery Service or Spirited Away were blocked, though generic prompts like Totoro Style (phrased as "My Neighbor Totoro") sometimes pass if the system judges it leniently.
  • Specific Architecture: Requests for Fallingwater were denied.

• Hallucination Quirks: A unique failure mode for this model is the insertion of unprompted text elements. In the Singapore Hawker generation, the model added a hallucinated caption bar, and in the Movie Poster, it accidentally rendered the prompt instructions as the movie title.

• Style Regression in Complexity: When faced with Complex Scenes like a Bustling Market or Underwater Scene, the model often abandons photorealism for a "painterly" or "illustrative" style, likely to avoid rendering flawed human faces in crowds.

🎯 Best Use Cases

• 🖋️ Graphic Design & Branding (Score: 9.1/10) This is the model's strongest suit. It creates usable, professional assets instantly. The Mascot Design and App Icon are production-ready. Recommendation: Use for logos, UI elements, and social media graphics.

• 📸 Photorealistic Portraits (Score: 8.2/10) It excels at lighting and texture, producing stunning results like the Old Fisherman and Night Portrait. Note: Be wary of "plastic skin" smoothing in group shots like the Group Selfie.

• 🎨 Surrealism & Concepts (Score: 7.4/10) Great for clean, commercial surrealism like the Avocado Armchair. However, it struggles with highly chaotic or abstract prompts like the Galaxy Waterfall where it added unwanted elements.

• ⚠️ Areas to Avoid

  • Fan Art & Copyrighted Styles: Due to the high refusal rate in Ghibli style, it is unreliable for specific stylistic mimicry.
  • Complex Crowd Scenes: Unless you want an illustration, the model struggles to maintain photorealism in Complex Scenes.
  • Specific Architectural Plans: Requests for technical drawings like the Machiya Townhouse may trigger policy refusals.