GPT Image 2 OpenAI vs Seedream 4.5 ByteDance

GPT Image 2

• + Perfectly legible text including item names, descriptions, and prices.
• + Comprehensive professional layout with categorized sections for Appetizers, Pizza, and Mains.
• + High-quality, distinct food photography for every listed item.

• − The layout is quite dense, arguably bordering on crowded rather than minimalist.

Seedream 4.5

• + Clean, minimalist aesthetic with a lot of white space.
• + Good use of vibrant primary colors for category accents.

• − Text is largely nonsensical with frequent typos and repeating placeholders.
• − Prices are unrealistic and formatted poorly.
• − Only one image per category, which doesn't showcase variety well.

GPT Image 2

• + Perfect text rendering for all requested phrases including the currency symbol.
• + Highly detailed and appetizing textures on the patty and fresh vegetables.
• + Dynamic 'exploded' view better captures the prompt's layout request.

• − The lighting on the burger feels slightly disconnected from the fiery background.

Seedream 4.5

• + Excellent integration of fire effects within the text typography.
• + Strong sense of motion through the use of realistic radial blur on flying ingredients.
• + Good atmospheric lighting that reflects on the bun and cheese.

• − Failed the 'exploded' request as most of the burger is still stacked.
• − The 'MAGIC BURGER' text has slight irregularities in letter shapes.
• − Texture on the patty is less detailed compared to Model A.

GPT Image 2

• + Perfect text accuracy and spelling for all items.
• + Highly realistic chalk texture with dusty smudges and varying stroke pressure.
• + Uniform handwriting style that feels authentic to a single person's script.

Seedream 4.5

• + Clear, legible text rendering.
• + Accurate date and item names.
• + Good use of vertical space on a portrait-oriented board.

• − Repetitive text error where 'Risotto - $24' is listed twice on two separate lines.
• − The chalk texture is very clean, almost looking like a digital chalkboard font rather than hand-drawn sticks of chalk.
• − The border of the chalkboard looks slightly less integrated with the background environment.

GPT Image 2

• + Excellent character consistency for the face, hair, and sunglasses.
• + Includes the specific scarf accessory and clothing color from Image 2.
• + Follows the exact overlapping leg pose from Image 1.

• − The scarf's positioning looks slightly unnatural across the neck in this pose.
• − Anatomy of the rear foot is slightly warped.

Seedream 4.5

• + Captures the likeness of the man in Image 2 very well.
• + Includes the graphic text/logo from the original shirt on the new pose.
• + Maintains the yellow studio background and red ottoman perfectly.

• − Fails to replicate the specific leg crossover pose from Image 1, resulting in a more generic crouch.
• − The left hand geometry is messy with extra finger-like artifacts.

GPT Image 2

• + Excellent adherence to the specific prompt instruction of having the horse on top.
• + Highly detailed textures on the spacesuit and lunar surface.
• + Clever and surreal composition that looks intentional

• − The horse's legs and the harness logic are slightly anatomically confusing.
• − The direct front-facing horse head looks slightly flat compared to the body.

Seedream 4.5

• + Vibrant, cinematic colors and nebulae backdrop.
• + Dynamic posing and good lighting on the astronaut.

• − Failed the primary prompt instruction by placing the astronaut on top/side of the horse.
• − Significant anatomical distortions where the horse's legs grow out of the astronaut's body.
• − General lack of clarity in how the two figures are joined together.

GPT Image 2

• + Perfect preservation of the original person's face, hair, and specific vitiligo patterns
• + Highly accurate reproduction of the target outfit including the scarf pattern and coat texture
• + Very clean integration around the neck and hands

• − The scarf's drape is a bit stiff/flat compared to the source

Seedream 4.5

• + Includes accessories like the sunglasses and watch from Image 2
• + Realistic fabric textures and lighting

• − Failed to preserve the base person's face and hair, creating a hybrid of both people
• − Changed the person's head pose, violating the instruction to keep the person unchanged
• − Failed to dress the person properly, leaving the chest bare under the scarf

GPT Image 2

• + Excellent texture on the capybara's fur and the jacket fabric.
• + Realistic taxi partition and interior lighting that matches a genuine NYC taxi feel.
• + Stronger depth of field providing a professional cinematic look.

• − Only one paw is clearly visible on the steering wheel, partially missing the 'both paws' requirement.
• − The passenger is quite blurry due to the shallow depth of field.

Seedream 4.5

• + Perfectly follows the instruction for 'both front paws on the steering wheel'.
• + Great composition that shows both the driver and the bored passenger clearly.
• + Accurate rendering of the 'bored' expression requested for the businesswoman.

• − The exterior perspective through the front windshield feels slightly flat and less realistic than the side windows.
• − The capybara's paws look a bit like animalistic hands rather than realistic capybara feet.

GPT Image 2

• + Excellent typography with a truly vintage gothic aesthetic
• + Exceptional detail in the border including webs and thorns
• + Strong cinematic lighting with atmospheric depth and texture

• − The dark color palette makes the text at the bottom slightly harder to read

Seedream 4.5

• + Clear and legible event details
• + Accurate fulfillment of all prompt elements
• + Vibrant central glowing jack-o-lantern

• − The thorns in the border look more like barbed wire
• − Composition feels less 'vintage' and more like a modern digital graphic
• − Typography is a bit generic compared to the requested gothic style

GPT Image 2

• + Excellent typography with correct Italian accenting and weight
• + Highly cohesive vintage woodcut illustration style
• + Professional composition with balanced ornamental framing

• − The 'EST.' text is slightly less crisp than the main title

Seedream 4.5

• + Successfully incorporates all prompt elements like the banner and steam
• + Clean, minimalist vector aesthetic
• + Accurate spelling of the name and date

• − Arched typography is poorly spaced and feels amateurish
• − The shading on the cloche is a bit simplistic compared to the vintage prompt
• − Lacks the sophisticated texture found in Model A