MiniMax I2V-01-Live: how to write prompts the model actually understands
MiniMax · Updated:
MiniMax I2V-01-Live is a specialized Image-to-Video model for animating 2D illustrations: anime, manga, digital portraits, concept art. Up to 720P, 6 seconds. The model preserves the original art style and line work, optimized for subtle Live2D-style motion. The prompt describes motion only; 10-30 words is optimal.
What I2V-01-Live does
This is a separate specialized MiniMax model, not a mode of Hailuo 2.3 or 02. It's built specifically for animating 2D illustrations — anime, manga, digital portraits, book illustrations, drawn characters. The headline feature is preservation of the original image's lines, colors, and art style. The model doesn't «redraw» the illustration and doesn't drift into realism.
It works best with subtle Live2D-style motion: blinking, breathing, slight head turn, flowing hair, fabric ripple. Ideal for VTuber content, animating manga panels, concept art, and book illustrations. Up to 720P, duration up to 6 seconds.
- I2V only — no reference, no generation
- Specialization: anime, manga, digital portrait, concept art
- Art-style and line preservation from the original
- Live2D-style animation (subtle motion)
- 720P, up to 6 seconds
Prompt structure
Formula: [Motion description] + [Direction/type of movement] + [Atmospheric changes]. Optimal 10-30 words — far shorter than for Hailuo 2.3.
Example: «Hair gently flowing, eyes blinking slowly, soft light shifting on her face, subtle smile.» The prompt does not describe the illustration's contents — the model sees it. Only what moves and how is described.
A short prompt is the norm for I2V-01-Live, not a shortcoming. Long descriptions with complex action work poorly: the model is tuned for Live2D aesthetics where minimal motion creates an «alive» effect without breaking the source style.
Subtle motion is the model's feature
Live2D-style animation is the subtle smooth motion characteristic of VTuber content. What works well: blinking (blinks slowly), breathing (chest rises and falls gently), slight head turn, flowing hair (hair gently flowing), fabric ripple (fabric rippling softly), expression shift (subtle smile appears).
What works poorly: running, jumping, flips, fight scenes, complex physical action. For that you need Hailuo 2.3, not I2V-01-Live. Trying to generate dynamic action on this model breaks the illustration's lines, introduces edge artifacts at motion boundaries, and loses the art style.
Art-style preservation
The model's headline promise — it doesn't «repaint» the illustration into photorealism. An anime portrait with cel-shading goes in, cel-shading comes out. A black-and-white manga panel stays monochrome. Watercolor with soft transitions keeps its transitions.
This distinguishes I2V-01-Live from general video models: Hailuo 2.3, Runway, or Sora tend to «improve» an illustration toward a more realistic style. I2V-01-Live doesn't — it's specifically trained to hold the input image's art style. So style is NOT set in the prompt — it's set by the illustration. Don't write «in anime style» if the illustration is already anime.
Common mistakes
1. Describing the illustration's contents
«Anime girl with purple hair in a school uniform sitting by the window, blinks» — the model sees the illustration; describing its contents is pointless and wastes tokens. Write motion only: «slowly blinks, hair gently flowing.» This shortens the prompt to 10-30 words and focuses the model on animation.
2. Action that's too dynamic
Running, jumping, flips, strikes, fight scenes don't fit 2D Live2D-style animation. The model is tuned for subtle motion. On dynamic action the illustration's lines break, edge artifacts appear, the art style is lost. For dynamic action use Hailuo 2.3 — it handles motion better, even stylized.
3. Describing art style in the prompt
«In anime style», «with cel-shading», «manga aesthetic» — style is already set by the input illustration. A style call in the prompt is either ignored or conflicts with what the model parsed from the image. For I2V-01-Live, style is a property of the reference, not a prompt parameter.
4. Using it on photorealistic images
I2V-01-Live is specialized in 2D illustrations — anime, manga, digital portrait, concept art. With a realistic photo as input the model will try to process it like an illustration: the result is odd, with loss of photographic detail. For photos use Hailuo 2.3 or Veo, not Live Illustrations.
5. Prompt that's too long
For I2V animation of 2D illustrations, 10-30 words is optimal. Long prompts with many motions cause overload and artifacts — the model tries to animate everything at once and breaks the lines. One or two primary motions + an atmospheric layer = the ideal prompt.
Before / after examples
Example 1
Before
make the anime girl alive
After
Hair gently flowing in the breeze, eyes blinking slowly, a faint smile appears at the corner of her mouth, soft light shifting on her cheeks.
Neither appearance nor style is described (both are on the illustration). Only subtle motion: hair, blinks, expression, light shift. Length is in the 10-30 word target range.
Example 2
Before
bring my character to life
After
The character's chest rises and falls with slow breathing, cloak gently rippling at the hem, eyes blink once. Faint magical sparkles drift around her shoulders, light atmospheric shimmer.
Main chain of subtle motions (breath, cloak, blink) plus an atmospheric layer (sparkles, shimmer). This counterpoint conveys dynamics without breaking the illustration.
Example 3
Before
make the manga hero move
After
Wind tousles his hair from left to right, eyes narrow slightly into a focused expression, the edge of his cape lifts and falls. Light shadow shifting on his face.
Hair direction is specified (left to right), micro-expression (eyes narrow), subtle clothing motion (cape lifts). The model holds the monochrome/manga style from the source without any style call in the prompt.
