Guide
This page documents a small local experiment: eight MP4 outputs from a Seedance 2–class fast model, paired with the prompts used for each run. Runs 1–3 repeat the same prompt text; later runs tweak a single line (transition wording, smile tone) or switch to one long “storyboard” prompt. Even when the wording matches, each generation is an independent sample—timing, camera bias, particle density, and facial performance drift. This site is third-party reference only; it does not claim an official Seedance benchmark or product guarantee.
Source basis and reading boundary
These guides are written as third-party reference summaries, not official product documentation or support content.
All three runs use the identical two-part prompt: a close-up of a schoolgirl under blue sky, then a transformation into a Greek goddess with golden magic circle, moon staff, and a wicked smile (“优雅邪笑”). The only difference between runs is the random seed. Comparing them shows how much visual variation the model produces from the same input — useful for understanding when re-rolling is worth it vs. when you need to change the prompt.
Run 4 inserts a single Chinese sentence between the subtitle and the goddess description: “女孩開始變身為另一形態” (the girl begins to transform into another form). This tests whether an explicit narrative bridge helps the model produce a smoother or more dramatic transformation sequence compared to the abrupt two-part structure of Runs 1–3.
These runs change exactly one word: “优雅邪笑” (elegant wicked smile) becomes “优雅微笑” (elegant gentle smile). The morph transition line from Run 4 is kept. Three rolls show whether this single-word tone shift consistently changes the character’s expression and overall mood in the final goddess form.
Run 8 replaces the two-part prompt with a single consolidated cinematic block containing full art direction: shot-by-shot staging, costume transformation details, magic-circle design, particle effects, wind physics, and embedded Japanese subtitle lines. This tests whether a detailed storyboard prompt gives the model more control over the sequence vs. the shorter, looser prompts in earlier runs.
First draw with the shared two-part prompt: school uniform eye close-up with Chinese subtitle line “是時候開始了”, then goddess beat with “优雅邪笑”. Establishes baseline pacing between the two beats.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅邪笑，动态特效 日语字幕：月亮賜我力量
Prompt string matches Run 1 word for word; this is a new stochastic sample. Expect shifts in particle bursts, hair motion, and how long the eye close-up holds.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅邪笑，动态特效 日语字幕：月亮賜我力量
Another independent draw with identical wording; useful to compare how “邪笑” lands across samples when you are picking a hero take.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅邪笑，动态特效 日语字幕：月亮賜我力量
Inserts “女孩開始變身為另一形態” between the subtitle beat and the goddess description so the model weights the morph transition more explicitly.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 女孩開始變身為另一形態 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅邪笑，动态特效 日语字幕：月亮賜我力量
Same structure as Run 4 but “优雅邪笑” becomes “优雅微笑”, pulling performance toward a softer, heroic smile instead of a mischievous one.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 女孩開始變身為另一形態 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅微笑，动态特效 日语字幕：月亮賜我力量
Duplicate wording to Run 5; compare how facial performance and effect density still vary between runs.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 女孩開始變身為另一形態 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅微笑，动态特效 日语字幕：月亮賜我力量
Another independent sample with the gentle-smile prompt; good for spotting which take reads clearest on a small mobile screen.
日式动漫，高中校服少女，眼神特写，藍天白雲的背景电影质感，高清流畅 日语字幕： 是時候開始了。 女孩開始變身為另一形態 日式动漫，少女变身希臘女神，金色魔法阵，月亮法杖，优雅微笑，动态特效 日语字幕：月亮賜我力量
One consolidated prompt with cinematic language, shot breakdown, Japanese lines 「さあ、始める時だ。」 and 「月よ、私に力を。」, and stronger art-direction density. Useful for comparing “many instructions in one block” vs the shorter stacked prompts above.
高品質日式動漫風，電影級分鏡感，10秒短視頻，高清細膩，流暢動畫，強烈變身張力與史詩戰爭氛圍。藍天白雲下，一名身穿日式高中校服的少女站立，微風吹動頭髮與裙擺。鏡頭以眼神特寫開始，少女由平靜轉為堅定，日語字幕：「さあ、始める時だ。」隨即地面展開巨大的金色希臘風魔法陣，帶有月亮符文、古代神話紋樣與神聖幾何圖騰，金色光柱沖天而起，金色粒子、流光軌跡、羽毛狀光片、風壓波紋環繞全身，展現華麗而有力量的變身演出。她的校服逐漸轉化為白金與月光銀色調的希臘女神戰鬥禮裙，服裝華麗神聖，帶有金色飾邊、古希臘式披肩、金屬臂環、層次飄逸裙擺與神聖紋章，整體優雅、高貴、不可侵犯。她伸手召喚一把華麗的月亮法杖，法杖頂端是新月與寶石核心，散發金銀交織的神力光芒。最後少女以希臘女神姿態站在金色魔法陣中央，優雅微笑，眼神堅定，手持月亮法杖指向前方，長髮與裙擺在強風中飛揚，月光與金光交織。日語字幕：「月よ、私に力を。」
No. Fast generative video is a sampling process. Matching text mainly aligns intent; it does not remove variation between draws. That is why teams generate multiple candidates before locking a shot.
No. This page is a third-party explanatory note with local clips. Always verify current behavior and limits on the product surface you use.
Yes. Runs 1–3 use the identical prompt and produce visibly different compositions, transformation timing, and facial expressions. Stochastic variation is significant enough that re-rolling is a valid strategy when a prompt is close to what you want.
Run 4 adds “女孩開始變身為另一形態” as a narrative bridge. Compare it with Runs 1–3 to judge whether the transformation feels smoother or more intentional. In our observation, it provides a clearer midpoint between the two states.
Changing “优雅邪笑” (wicked smile) to “优雅微笑” (gentle smile) in Runs 5–7 consistently shifts the character’s final expression from menacing to heroic/serene. Even one adjective swap can change the emotional tone of the entire clip.
Both work. Short prompts (Runs 1–7) are faster to iterate and give the model more creative freedom. The long storyboard prompt (Run 8) gives tighter control over staging, costume details, and shot flow, but is harder to debug when something goes wrong. Start short, switch to storyboard when you know exactly what you want.
Continue with the tutorial and best-practices guides for prompt structure, references, and continuity. Use the multi-clip field notes if you are stitching segments in an editor.
Guide
Best practices for Seedance 2.0: prompt formulas, reference assets, camera and motion wording, and quality checks. Based on public guides.
Open guideGuide
Write better Seedance 2.0 prompts: subject + action + camera + style formulas, @ reference tags, and practical before/after tips for text-to-video and image-to-video workflows.
Open guideGuide
Step-by-step Seedance 2.0 tutorial for beginners: text-to-video, image-to-video, prompt structure, settings, and your first generation on Dreamina. Updated April 2026.
Open guideGuide
Two anime genres, one fast model: a magical-girl transformation series (8 runs) and a first-person romance POV interaction. Categorizes prompt structure, model demands, and observable output differences.
Open guide
