Video to Prompt
Upload any video and get the structured prompt that recreates it — task, shots, camera moves, performance and dialogue.
I have read and agree to the Terms of Service. I understand that analysis results are for reference and learning only and must not infringe others’ rights in their creative content.
Results
What 向光AI extracts
Upload one video and reverse-engineer a ready-to-generate structured prompt, including:
- Event summary — every beat of the video, in order
- Image prompts — scene / character / prop, ratio auto 9:16 or 16:9
- Video prompt — per task & per shot: shot size / camera move / duration / action / performance / dialogue, ready for Jimeng / Seedance
- Auto task-splitting — videos past 15s split into clips to fit Seedance's per-clip limit
- Start keyframe — split out as a first frame depending on the target model
How it works · 3 steps
Upload a video, or paste a Douyin / Xiaohongshu / Bilibili link
Pick the reverse model / video type / target model (defaults work fine)
Get a structured prompt in ~30s — copy it straight into Jimeng / Seedance
Our Testing
- Parses Chinese-platform sources directly (Douyin / Xiaohongshu / Bilibili)
- Conversational adaptation lets you change scenes, people, props, plot, dialogue, style and pacing
- Three output formats (all-in-one reference / first-frame / first-frame+reference) — image + video prompts for Jimeng / Seedance; the structure also works for Kling and others
FAQ
Which video links are supported?+
Douyin, Xiaohongshu and Bilibili links auto-parse. YouTube / TikTok / Kuaishou aren't supported directly yet — download the video and upload the file instead.
What is the maximum video length?+
Videos up to 10 minutes are supported. Longer videos return VIDEO_TOO_LONG before analysis and do not consume video credit.
How long does it take?+
Processing time depends on the video length, file size, link download speed, and current queue. You can follow the job status, and completed results remain available in history.
Do I need to sign up?+
Yes. Sign up (email + password or Google) for free trial credit; top up time packs when it runs out — billed by video length, charged only on success.
Is my video stored?+
Your video is used only for this analysis and deleted from the server right after; only the text result is saved to your history.
How can I adapt the result?+
Continue the conversation to change the scene, people, props, plot, dialogue, visual style or camera pacing while keeping the useful structure from the reference.
Does it support vertical and landscape videos?+
Both. Image prompts auto-pick the ratio from the source — 9:16 for vertical, 16:9 for landscape — so framing isn't cropped.
Can I reverse multiple videos at once?+
Yes. Submitting several queues them in the background; Pro / Pro+ plans get priority so they don't wait at peak times.
How is this different from describing the video to ChatGPT?+
向光AI outputs in the structured format Jimeng / Seedance ingest directly (event summary / tasks / shots / image prompts), parses Chinese-platform links directly, and supports conversational adaptation — ready to use without reformatting.