seedream 5.0 pro vs gpt image 2 vs mai image 2.5 vs grok imagine

seedream 5.0 pro just entered the multi-image edit arena at #2 with 1415 pts. big jump from seedream 4.5, which sat at #11 – so we put it to the test to see how it holds up

we ran everything via @aimlapi – one platform, one api, all four models, so it works like a real image-editing pipeline inside a saas

same 3 edits across all four models – colorize + composite on iconic photos:

1. ali standing over liston – colorize, swap the crowd for a nyc skyline under a red demonic sky with a whirlwind, add hp bars up top (green ali, red liston)

2. "lunch atop a skyscraper" – colorize, make the workers barefoot and bare-headed, remove the rope on the right

3. "the roaring lion" (churchill) – colorize, give him a full head of hair, make him slimmer and younger

quality per image:

ali over liston
#1 gpt
#2 seedream
#3 grok
#4 mai

lunch atop
#1 seedream
#2 gpt
#3 mai
#4 grok

churchill
#1 seedream
#2 gpt
#3 grok
#4 mai

quality overall (rank points, least = best):
#1 seedream – 4
#2 gpt – 5
#3 grok – 10
#4 mai – 11

cost (3 images):
#1 seedream – $0.176
#2 mai – $0.190
#3 gpt – $0.208
#4 grok – $0.273

speed (3 images):
#1 grok – 52s
#2 mai – 1m36s
#3 gpt – 3m02s
#4 seedream – 6m12s

observations:

• seedream excels at not touching what's already in the frame. it edits around the existing objects instead of redrawing them – that's why its outputs look the most realistic

• gpt no longer feels like the #1 on image edit. it's now weaker on both price and quality

• mai and grok seem to barely look at the source image. grok at least nailed muhammad ali, mai failed every test

savings at scale (seedream $0.0585/img vs the rest):

• per 10k images: vs grok save $325, vs gpt save $107, vs mai save $48

• per 100k images: vs grok save $3,250, vs gpt save $1,073, vs mai save $477

• per 1m images: vs grok save $32,500, vs gpt save $10,730, vs mai save $4,770

so seedream is both the cheapest and the best output here. the only tradeoff is speed – it's by far the slowest of the four