Seedance 2.0 FEELS like old Sora but BETTER. Fight Scenes Are Finally GOOD!

Bite Dance’s “Seance 2.0” is being pitched as a major leap in text-to-video quality—especially for animated action—while also standing out for being less restrictive than competing models. Early reactions focus on fight choreography that stays readable frame-to-frame, more convincing physical effects (including smoke and water-like behavior), and audio that often lands with a “cinema” feel. Viewers also describe the model as closer to Sora-style motion quality than to purely cinematic “look-at-this-shot” generators, with the added promise that clips can be built for longer workflows.

A key theme is coherence: action sequences don’t just look dynamic; they maintain continuity across the duration of the clip. Animated fight scenes—particularly anime-style work—are repeatedly singled out as a standout use case. One example blends live-action environments with 2D anime characters, where the character’s entrance, camera rotations, and 3D-realistic body presence are said to remain consistent enough to feel intentional rather than randomly assembled. Even when artifacts appear—like occasional finger problems or impossible movements—users describe the hallucinations as increasingly “sussable,” suggesting the model is getting better at staying within the rules of a scene.

The transcript also emphasizes multimodality and control options. The platform is described as allowing references for images, backgrounds, videos, and audio, plus “first and last frame” controls for steering motion. Aspect ratio choices range from classic 4:3 to widescreen 21:9 and vertical formats like 9x6, with “smart aspect ratio” and “smart length” settings that let the system decide resolution and duration. The creator’s hands-on test uses text-only prompting (because uploads weren’t available at the moment), targeting a 15-second Atlantis criminal-hiding movie scene, and notes that generation speed can be fast—often under a minute—though it can also stall during longer batches.

Community outputs are used as proof points. Discord-made clips are described as “uncensored” and are compared favorably against Sora 2 and “Sora 2 Pro,” with recurring praise for audio realism and animation adherence. Examples include Rick and Morty-style scenes, sports-like ball interactions, anime-meets-real-world action, and even mashups resembling GTA 6 rendered with Pixar-like aesthetics. Still, limitations remain: a separate test is cited where a duck fails to follow a maze and instead passes through walls, showing that the model can produce visually plausible motion without reliably obeying constraints.

Overall, the transcript frames Seance 2.0 as a model that feels “fresh” and broadly capable—particularly for animation and action—while acknowledging that it’s not yet dependable at strict rule-following. The closing sentiment is that competition among video generators should accelerate progress, with hopes for open, high-quality models in the coming years, while the immediate focus stays on experimentation and sharing results.