Skip to content
Ani VideoRUEN
Guide

Magic prompt

A neural net looks at your picture and rewrites a short description into a full prompt for the video model: it breaks the motion into phases and describes camera and sound. You choose how much it may imagine.

New 2.1 features are in beta. Check availability and the final price in the order panel.

Contents
  1. 01. What it is
  2. 02. Before and after
  3. 03. MP imagination levels
  4. 04. Which model MP writes for
  5. 05. The /mp command: a prompt without a clip
  6. 06. What it costs
  7. 07. Where to see the result
01

What it is

A neural net looks at your picture and rewrites a short description into a full prompt for the video model: it breaks the motion into phases and describes camera and sound. You choose how much it may imagine.

BEFOREone line in your languageNeural netlooks at the pictureintegrated_multimodal_descriptionoverall_soundscapenon_diegetic_musicthree fields, markers, timecodesTurn it on: if you write one line and skip shot planning.Turn it off: if you wrote a detailed English prompt yourself.
BEFOREone line in your languageNeural netlooks at the pictureintegrated_multimodal_descriptionoverall_soundscapenon_diegetic_musicthree fields, markers, timecodesTurn it on: if you write one line and skipshot planning.Turn it off: if you wrote a detailed Englishprompt yourself.

It assembles a ready prompt with three fields, shot markers and timecodes, plus speech with its language specified.

02

Before and after

Before: what a person writes

Before
a girl takes a sip from a can, then says «Вкусно!»; the camera stays still

After: what goes into the model

After
integrated_multimodal_description: [Shot 1] 2D-animated, a medium shot of the anime girl with long blonde hair and a black bow. She raises the can to her lips and takes a quick sip, her throat moving as she swallows. [Shot 2] At 00:02.500, the camera cuts to a close-up of her face as she lowers the can and her eyes widen. The girl with a bright, energetic young female voice (S1) says: <d>[Russian] Вкусно!</d> Her free hand stays at her side and her hair settles on her shoulders, the background stays exactly as it is in the first frame. The camera stays completely still, no zoom. Smooth anime animation, natural drinking motion, clear lip movement.

overall_soundscape: A quiet indoor room tone continues throughout. The can lid clicks and she swallows audibly.

non_diegetic_music: N/A
03

MP imagination levels

OffUses your own text without MP. Describe the actions, camera and sound yourself.
LowFollows your request closely and adds only necessary transitions between actions.
MediumFills out the scene: preparation and reactions, objects and people in the frame and one connected event, based on the poses and objects in your sources.
HighWorks like a director: a setup, a turn and a strong finish, new events even in a short clip, two or three beats from 6 seconds and an active camera. The main action is preserved.

At every level, your sources, explicit restrictions and intended outcome come first. MP cannot be disabled without a description because it must invent the request. The price does not change with the level.

Video references work differently

MP uses Spark with low imagination here to follow the clip. The button only toggles MP on or off; your choices for other modes are kept.

04

Which model MP writes for

Video models and MP models are different choices. Flow supports Spark, Pulse or Wave; Core and video references use Spark. The MP name appears under the finished clip even when MP is off.

🌊 Navi FlowThe official format: three fields, shot markers with timecodes, speech with its language specified. This is the kind of prompt shown above.
⚛️ Navi CoreA coherent paragraph in direct words, by the rules of the chosen scenario. Sound and whatever must not appear in the frame go on separate lines.

Each Navi Core scenario has its own rules. For Glass Kiss MP writes briefly: the character kisses the screen, and only what you asked for is added. For Dance it asks for dance music with a beat by default; “no music” leaves the sounds of movement. For NSFW the tempo is unhurried until you ask for faster or harder.

If you don’t want any sound, write “no sound”.

You don’t need to write the prompt for Navi Core yourself, but the general rules are the same as in the Prompts guide.

05

The /mp command: a prompt without a clip

/mp works like an order, but the result is only a ready prompt. Attach a picture (and a clip, if you like), write your request and get text you can edit and paste into a regular order.

FormatThe format button switches what to write for: /video, /ref or Navi Core (one image, no clip or audio).
Names of sourcesThe neural net knows your files as “image 1”, “video 1”, “audio 1”: use those names in your request.
Price30 tokens per prompt; the first two a day are half price, and with a plan some prompts a day are free. Exact numbers are on the Pricing page.

A ready Navi Core prompt from /mp, pasted into an order as a whole, goes to the model as it is: Magic prompt is off for it.

/mp analyses up to 10 seconds of a clip. The 20-second limit applies to video-reference orders, not this command.

06

What it costs

Cost

+12% on top of the clip; +8% with Rough, +6% with Sketch, free with Study. For Navi Core, Magic prompt is already part of the clip price.

07

Where to see the result

Full prompt under a clip shows the text used. The imagination level appears in the panel and result settings. The MP button opens settings for regular clips and directly toggles the assistant for video references.