Seedance 1.0 vs. Veo 3: An Informal Text-to-Video Comparison

In June 2025, Matthew Berman compares ByteDance’s Seedance 1.0 with Google’s Veo 3 using a series of text-to-video prompts. He mentions a contemporary Hugging Face leaderboard, but his own exercise is a small, subjective comparison rather than a controlled benchmark or a current ranking.

The first prompt asks for a gymnast performing flips in a well-lit gym. Berman finds Seedance’s motion stiff and physically awkward, but sees fewer limb distortions than in the Veo clip and narrowly favors Seedance. His first few Seedance generations were 480p, while later ones were 1080p, so visual comparisons are not all at the same resolution.

For a nighttime street interview, Berman praises the Seedance clip’s clothing, hands, passing cars and lighting. He notices the characters appear Asian and suggests a connection to training data, but the samples alone do not establish the composition of the model’s training set.

The Veo interview includes audio, unlike the Seedance output he tested. Berman finds its speech out of sync with the speaker’s lips and notices morphing hands and background figures, so he favors Seedance visually. The audio difference makes this a comparison of the configurations and outputs he used, not a complete feature comparison.

Berman shows the Seedance interface with text-to-video and image-to-video options. He says he paid roughly $100 for 16,000 credits and that a 1080p, ten-second generation cost 800 credits in his setup; he also found it slower than Veo 3. These are his June 2025 observations, not current prices or a standardized speed test.

A head-on freight-train crash challenges physical realism. Seedance produces attractive images but puts one train behind the other rather than clearly showing the requested collision. Veo’s version has implausible, bouncing motion and an object appearing suddenly. Berman considers both poor and slightly prefers Veo as an entertaining clip.

For a Rubik’s Cube, the prompt asks hands to turn and solve the puzzle. Berman sees realistic-looking hands in Seedance but too little motion and colors changing inconsistently. Veo shows more hand movement but still does not depict a realistic solve; he gives Veo a narrow edge.

A long animated prompt follows a small robot into an abandoned toy store, where it wakes a dinosaur and other toys. Seedance produces an appealing sequence but omits many requested events and shows some morphing. Berman highlights consistent details around the moon and opening door, including how the glass changes its appearance.

Veo’s robot clip appears darker and more cinematic to Berman, with a more detailed robot and camera movement. He notices the surrounding environment changes abruptly, undermining continuity. Although he finds Seedance more consistent, he gives Veo the win for the overall scene and camera work.

A surreal restaurant prompt asks for rain only inside, with a dry sunny exterior. Seedance instead shows the couple leaving before rain falls outside. Veo shows rain inside but also outside, so neither follows the whole prompt; Berman favors Veo because it includes the interior rain.

For a Zelda-like forest scene, Seedance shows a character running and jumping but misses actions such as cutting grass and solving a puzzle, and does not clearly use the requested cel-shaded style. Berman says Veo also changes the requested style but includes game-like interface elements and an apparent puzzle, giving it a clear win on this example.

Berman asks for a horror scene in a hospital and generates two Seedance versions. He finds both frightening and visually realistic, but says movement is slow and a creature appears to emerge from a wall; the model also repeats a similar scene. He notes these samples do not show the same character appearance he observed earlier, reinforcing that a few outputs cannot establish the training set.

In the Veo horror clip, Berman prefers the running, lighting and apparent speed, although he considers the comparison close. The prompts request sound, but he notes his Seedance configuration has no audio. He treats his winner as a judgment of the specific clips shown.

An apocalyptic city prompt combines a hurricane, lightning, a derailed monorail, fighting and flooding. Berman praises the fire and water effects in Seedance, while noticing the fire vanishes quickly. Veo includes several requested elements but abruptly switches scenes; he favors Seedance for continuity.

For an Olympic-style dive, Berman initially likes Seedance’s still frame but finds the diver’s body and clothing morph during motion and the background spectators static. Veo’s dive is awkward rather than technically convincing, yet he favors it because the body remains more coherent.

A monster-truck prompt asks for drifting through a snowy city and crushing cars. Seedance’s city looks realistic to Berman, but the truck morphs and does not create the requested destruction. Veo shows more action, although a taxi changes into a newsstand; he narrowly prefers Veo.

In a futuristic space battle, Berman finds Seedance’s motion plausible in places but its detail and style more cartoon-like than requested. He prefers Veo’s explosions, gunfire, movement and sound. The comparison reflects the clips shown and includes an audio capability absent from his Seedance run.

A prompt for two UFC fighters produces a detailed Seedance image but little convincing contact or grappling. Veo’s result looks more like a video game, yet includes more punching and kicking. Berman calls it a tie: he prefers Seedance’s visual realism and Veo’s movement.

A fantastical flying leviathan over a glowing landscape produces a Seedance clip Berman finds crude, with unwanted writing and an unconvincing waterfall. He strongly prefers Veo’s detail, shadow and wing motion. These subjective judgments describe one generation per model, not the models’ entire output distributions.

A wingsuit prompt seeks fast movement down a mountain. Berman says Seedance’s person, suit and shadows look plausible, but the motion is too slow and the scenery morphs. Veo moves faster but sends the figure uphill and changes the suit, so he gives Seedance the win.

Finally, a rural Chinese mountain-village prompt asks for a woman crossing a bridge followed by a dragon dance. Seedance shows unusually large fish and an abrupt scene change; Veo’s fish look more realistic to Berman, although neither clip satisfies him fully. He narrowly favors Veo.

Across his examples, Berman gives Veo 3 a slight overall edge for more realistic motion and physics, while acknowledging Seedance’s strong visual quality and continuity in some clips. The video contains a sponsored Augment Code segment. Its informal results depend on prompts, generation settings, sample selection and the reviewer’s criteria.