SUMM

Seed Dance 1.0 by ByteDance ranks #1 on Hugging Face leaderboards, beating VO3 by 50 ELO points
The video tests Seed Dance against V3 with various prompts and scenarios to evaluate video generation quality

First prompt: a female gymnast performing a floor routine
Seed Dance shows some stiffness but avoids major limb deformities, common in AI gymnastic videos
V3 displays more pronounced limb deformities and unrealistic motion
Slight win for Seed Dance, mostly due to fewer anatomical errors and more realistic movement

Realistic street interview scene generated by both models
Seed Dance’s output features Asian characters, high visual quality, and flawless animation; indistinguishable from real video for most viewers
V3’s version has audio but suffers from poor lip sync, background character morphing, and unnatural hand movements
Seed Dance easily wins this scenario

Two generation modes: image-to-video and text-to-video
1080p, 10-second video at top settings costs 800 credits (~$100 for 16,000 credits)
Seed Dance is notably slower than V3 in generation speed

Prompt: two freight trains crashing head-on
Both models fail to convincingly simulate the crash; physics and object interactions are unrealistic
Neither model produces satisfactory results; slight preference for V3 due to more engaging (though still flawed) scene

Prompt: close-up of hands solving a Rubik’s Cube
Seed Dance creates realistic-looking hands but lacks realistic arm and cube movement; color positions not accurate
V3 shows more movement and slight improvement in realism, granting it a minor win for this prompt

Prompt describes a robot and dinosaur in a toy store; emphasis on expressive animation and lighting
Seed Dance delivers highly realistic door effects, lighting, and moon reflection, but omits some prompt specifics
V3 produces more detailed, cinematic scenes but with inconsistency and abrupt scene changes
Consistency and physical realism win for Seed Dance; camera movement and cinematic style win for V3

Prompt: couple dines in a restaurant; it rains only inside, then they exit into sunshine
Seed Dance fails to depict indoor rain; rain appears outside after couple exits
V3 achieves indoor rain but is inconsistent after the couple leaves
V3 wins solely for including indoor rain

Prompt: Zelda-style cell-shaded video game scene, including action and puzzle solving
Seed Dance creates a video game-like character, not cell-shaded, and omits much of the required action
V3 includes HUD elements, puzzle-solving action, and better game-like styling; clear win for V3

Prompt: hyperrealistic horror in a hospital hallway with a faceless monster
Seed Dance produces terrifying visuals, consistent scene structure, but slow, unrealistic motion and flawed monster physics
V3 offers better lighting, more realistic movement, and superior horror atmosphere
V3 wins by a margin in realism and movement

Prompt: apocalyptic mega city under hurricane; chaos, looting, exosuits, and destruction
Seed Dance showcases strong fire and water effects, dramatic building destruction, and impressive visuals
V3 delivers striking imagery but switches scenes, hurting consistency
Seed Dance wins for maintaining a cohesive, detailed sequence

Gymnastic/Diving: Both models struggle, with Seed Dance experiencing extreme morphing and static backgrounds; V3 is better in body physics but still flawed
Monster Truck: Seed Dance offers detailed city but lacks interaction; V3 shows more chaos and destruction, making it the better choice
Space Soldier/Sci-Fi: Seed Dance’s output is cartoonish; V3 presents higher-quality action scenes, effects, and sound, providing a clear win
UFC Fight: Seed Dance excels in realism but lacks motion accuracy; V3 offers better fight mechanics but less photorealism; outcome is considered a tie

Prompt: bioluminescent Leviathan over alien landscape
Seed Dance output appears simplistic, lacking physical plausibility and visual quality
V3 easily outperforms by delivering detailed, dynamic animation
Fast wingsuit flight challenge: Seed Dance looks good but lacks speed, with grainy background details
V3 demonstrates more speed but suffers from morphing and continuity flaws
Seed Dance wins on the wingsuit challenge for better coherent imagery

Prompt: serene Chinese mountain village, cultural elements, dragon dance
Seed Dance generates visually strong scenes but with unrealistic fish and scene transitions
V3 displays more realism, especially with natural elements; both models are only moderately convincing
V3 narrowly wins for this cultural scenario

Reviewer gives a slight overall edge to V3 due to better physics realism across more prompts
Both models excel in different scenarios: Seed Dance in consistency and photorealism, V3 in dynamic action and contextual details
Closing sponsorship and YouTube channel reminders

New #1 AI Video Model is HERE (Beats VEO 3)