Creative Space
Write MiniMax H3 Prompts with Clear Shot and Audio Plans

Contenido

Users come with a short film they want to use MiniMax H3 to generate: usually a multi-shot short film, character performance or product display where images, dialogue and sound are produced together. Agent diagnoses problems according to the writing method of MiniMax H3, rewrites the prompts, and delivers prompt cards that can be used directly. If the user is willing, he can use Flova to generate the words before and after the rewriting and compare them side by side.

Tools and responsibilities

  • Flova: Query the current parameters and reference assets requirements of MiniMax H3; optional trial generation.
  • Image or audio processing (optional): Organize reference pictures and intercept reference audio.

Method index

  • : Suitable and unsuitable tasks, recommended structure, storyboard writing method, three types of creative methods, sound and dialogue, consistency, common failures and corrections. Read steps 2–4.
  • : card format, role of uploading and reference assets, how to refer to reference assets in prompts, trial generation and comparison. Steps 4–6 Read.

Boundaries

  • dialogue copying: The dialogue confirmed by the user will be retained verbatim, without translation, polishing or rewriting.
  • Do not exaggerate the effect: Prompts can improve the success rate, but cannot guarantee the consistency of lip sync and lens cutting. Do not claim a rewrite is definitely better unless it has been validated with actual trial generation.
  • assets rights: Reference assets and reference audio must be ones that the user has the right to use. Does not imitate the look or voice of real celebrities.
  • Try generation needs to be said first: It will consume the user's Flova quota, and it can only be done after the user agrees.

Workflow

Collect requirements → Check current capabilities → Diagnose → Rewrite (confirmation point) → Trial generation comparison (optional) → Delivery

1. Collect requirements

  • The original text of the prompt that the user already has (you can also write it from scratch).
  • Purpose, core events and style of the short film.
  • One shot or multiple shots? What information does each shot carry?
  • The original text, speaker, language, emotion of each dialogue, and whether it is inside or outside the dialogue.
  • When will ambient sounds, action sound effects, and music appear?
  • What each reference asset wants to be responsible for and what it does not want to bring in.
  • Duration and aspect ratio; tried results and unsatisfactory areas.

Please refer to Section 2 of for missing information. First determine whether the task is suitable for MiniMax H3 (Section 1). For example, if strict first and last frames or seamless looping are not suitable, tell the user directly.

2. Check current ability

Use creation_generation_models to check MiniMax H3 (detail="full"): supported duration, aspect ratio, resolution, refer to the number, type and role of assets. When the query fails, it indicates that the parameter is "unverified" and continues.

3. Diagnosis

Check the writing method of comparison one by one and make a diagnosis table: problems, corresponding writing rules, and possible consequences. Highlights:

  • Structure: Refer to assets responsibilities, one sentence core idea, global settings, shot timeline, and whether the preservation constraints are complete.
  • storyboard: Whether each shot is written as a complete sentence of "scenery + subject + action + camera movement + dialogue or sound"; whether the characters, costumes, and props are recounted after the cut.
  • dialogue: Does the length match the shot? Does the dialogue that crosses the shot indicate that it will continue with the previous shot?
  • When there is no dialogue: Is it stated that the character should keep quiet, his mouth should be closed naturally, and non-speaking actions should be arranged for him. This is key to avoid accidental lip sync.
  • Sound: Whether ambient sounds, sound effects, and music are written in the order of events; whether mixing is written as an abstract percentage.
  • Reference assets: Whether the action reference is only responsible for action and rhythm, and whether the character reference is only responsible for identity.
  • shot: I wanted to take one shot to the end but wrote a cut shot; has the "surround" been separated into horizontal and panning directions?
  • Contradiction: For example, reference music requires no music, and stillness requires fast movement.

When users receive unsatisfactory results, they should first check the box in Section 7 "Common Failures and Corrections" of the writing method.

4. Rewrite (confirmation point)

  • Rewrite according to the recommended structure and output it in the format of .
  • For reference assets, write down the responsibilities and exclusions in the manner of "The first reference picture is responsible for...", in the same order as when submitted. -Write complete sentences shot by shot in multiple shots, and retell the character's anchor point each time the shot is cut.
  • Sounds are written into the timeline according to events: which dialogue is said by whom in which shot, sound effects appear at which actions, and when the music enters, weakens, and stops.
  • Attached is a change description: correspond to the problems in the diagnosis table one by one, explaining what has been changed and why.
  • Show it to the user and confirm before trying to generate or deliver it.

5. Trial generation comparison (optional)

After the user agrees, follow to use Flova to quickly generate:

  • The original prompt and the rewritten version are each generated once, and the parameters and reference assets are exactly the same.
  • Show them side by side to the user, focusing on comparison: whether the characters are consistent after cutting the scene, dialogue lip sync, whether there are any unexpected dialogues or subtitles, and the order of the sounds.
  • If there are still problems, make another round based on common failures. Only change one variable in each round, for example, only shorten the dialogue, or only adjust the sound order.

6. Delivery

  • Prompt cards (one or more), the original text of the prompt can be copied directly.
  • Diagnostic sheet and modification instructions.
  • During trial generation: changes for each round, task IDs and results, and which version is ultimately recommended.
  • Remind users: Don’t rely on prompts for strict loops and pixel-level first and last frames, leave them to post-processing.

MiniMax H3 writing method

The duration, scale, resolution and number of reference assets actually supported by the model are subject to the query results of creation_generation_models.

1. When is it appropriate to use it?

Suitable for:

  • I hope that the images, dialogue, environmental sounds and action sound effects together form a complete short film.
  • Multi-shot narrative, character performance, product display, action changes and visual deformation.
  • Also refer to the creation of characters, scenes, products, actions, camera movement or sounds.
  • The shot language and sound entry sequence need to be relatively clear.
  • Develop actions from a single image, or define characters, scenes, and start and end states from multiple images.

Not suitable for:

  • There is only sound reference but no visual information, and at the same time, it is required to strictly restore the main body of the picture.
  • Requires pixel-level locking of first and last frames or absolutely seamless looping.
  • Need to do a very precise partial replacement of the existing video while keeping other pixels completely unchanged.
  • A lot of dialogue, multiple locations and complex plot twists need to be packed into a very short clip.

2. Ask users for clear needs

Before you start writing, ask questions from users:

  • The purpose, theme, core events and overall style of the video.
  • Subject appearance, scene, props and the most important identification features.
  • Do you want one shot or multiple shots, and what information each shot carries.
  • What state does the action start from, how does it change, and where does it finally stop.
  • camera position, scene, movement direction and rhythm.
  • The original text of the dialogue, the speaker, the language, the emotion, and the on-screen or off-screen status.
  • When will ambient sounds, action sound effects and music appear?
  • What each reference asset is responsible for and what it does not want to inherit.

3. Prompt writing method

Recommended structure
[参考素材职责]
[一句话核心创意]
[全局人物、场景、风格和声音设定]
[镜头或事件时间线]
[必须保持的内容与少量关键限制]

Specific sequence suggestions:

  1. Subject and scene: Who is where, what is the current state.
  2. Core Action or Variation: What happens, and how the action is accomplished sequentially.
  3. Visual style: Realism, animation, commercial or cinematic feel, as well as light and color.
  4. shot language: scene separation, perspective, camera movement and lens cutting logic.
  5. Sound: The sequence relationship between dialogue, environmental sounds, action sound effects and music.
  6. Keep items: characters, clothing, products, props, spatial direction and other content that cannot be changed.
storyboard writing method
  • Use a complete sentence of "scenery + subject + action + camera movement + dialogue/sound" for each shot.
  • Re-specify characters, costumes, props, and target scenes when cutting to help maintain consistency across shots. -The length of dialogue must match the rhythm of the shot; short shots should not carry long dialogue.
  • When a dialogue spans between shots, make it clear that "continue talking from the last shot" and indicate that the speaker changes from on-screen to off-screen or vice versa.
  • When you want to take one shot to the end, instead of using multiple shot structures, rewrite it as a continuous action and continuous camera movement.
  • "Surround camera movement" should be split into camera lateral movement and panning directions to avoid model misunderstandings.
Example
人物参考负责女主角的脸、发型、浅蓝色汉服和银色长剑;动作参考只提供舞剑动作与节奏。

一段电影感的中式庭院剑舞,晨光柔和,全程连续运镜。
开场为大全景:她从画面左侧缓慢进入,剑尖朝下,摄影机以步行速度向右跟拍。
随后镜头自然推到中景,她拔剑并完成参考动作,脸、服装和长剑设计保持一致。
结尾为近景,剑刃停在脸侧,花瓣划过前景,她平静呼吸并看向镜头。
无对白、无音乐,只保留脚步、风声和一次短促金属声。

4. Three types of creative methods

Multimodal Reference
  • The character diagram is responsible for the identity, the scene diagram is responsible for the space, the product diagram is responsible for the structure, the video is responsible for the action and shot rhythm, and the audio is responsible for the timbre or sound style.
  • Describe the purpose of each asset item by item, don't just say "refer to these assets".
  • When referring to video migration actions, indicate whether to inherit camera movement, speed, emotion and scene; unnecessary dimensions should be explicitly excluded.
Picture driver
  • A single image can be responsible for a character, scene, composition or opening.
  • Multiple pictures should clearly define their respective responsibilities, and do not let different pictures compete for character identity or product structure at the same time.
  • The two start and end state pictures can help describe the direction of change, but they cannot guarantee pixel-level matching in the last frame.
Pure text creation
  • Complement the subject appearance, scene details, action process, style, shot and sound.
  • You can use the hierarchy of "large panorama to convey space → medium shot to carry action → close-up to emphasize details".
  • The shorter the shot, the more concise the events and dialogue must be.

5. Sound and dialogue writing

  • Dialogue retains the content confirmed by the user or storyboard verbatim, without translation, polishing, or rewriting.
  • Write down the speaker, language, emotion, speaking speed and on-screen/off-screen status for each sentence.
  • Make it clear that the characters do not speak when there is no dialogue, and arrange natural movements for the characters to reduce accidental lip sync.
  • Ambient sounds, sound effects and music are written in the order of events, for example "there is a metallic sound when the sword is drawn, and the sound disappears after the sword stops".
  • Music writes about style, instrumentation, and mood, stating from which event it begins and when it fades or ends.
  • Use relative relationships when mixing, for example, "vocals are clear up front, music is low, and rain sounds are in the distance."

6. How to improve performance and consistency

  • Characters are always described with the same face, hairstyle, clothing outline, and accessories.
  • The product always maintains its structure, material, color, logo, text and placement direction.
  • Successive transformations state which visible features of the previous form must be retained at each stage.
  • The action reference is only responsible for movements and rhythm, and the character reference is only responsible for identity to avoid the confusion of responsibilities.
  • Rename characters, costumes, props, scene positions and light directions for each cut.
  • Multiple characters can be distinguished by name, clothing, body shape, position and voice.
  • When multiple candidates are needed, only adjust one variable at a time to easily determine which of the actions, shots, or sounds has improved.
  • Important text and logos require clear visual reference, and are reserved for proofreading and retrying after generation.

7. Common failures and corrections

  • Referencing assets has no obvious role: Explain item by item what each asset is responsible for, and what content cannot be inherited.
  • Character Changes after Cuts: Fixed main identity reference, and restated face, hairstyle, clothing and accessories after cuts.
  • Meanless dialogue or lip sync: Make sure the character remains quiet, his mouth is naturally closed, and there are non-speaking actions that are continuously visible.
  • dialoguelip sync is not allowed: shorten the dialogue so that the number of words matches the rhythm of the shot, and the speaker and language are clear.
  • Music Control Unstable: Use events to indicate when the music comes in, fades out, and stops completely, instead of writing abstract percentages.
  • A shot ends but a cut occurs: Delete the shot and cut structures, and only retain continuous actions, spatial paths and camera movement.
  • Conflicting Prompt Words: Check for conflicting requirements such as reference music vs. no music, static vs. fast action, realistic vs. strong cartoon, etc.
  • Subtitles or on-screen text appear unexpectedly: Reduce long dialogues that easily trigger subtitles and review them after generation.
  • Loop is not seamless: Prompt is difficult to ensure that the last frame returns to the first frame at the pixel level; strict looping should be left to post-processing.

Prompt Cards and Flova Handoff

These inspirations deliver prompt cards, not images/videos: one per image or shot. Cards can be used directly for generation; this reference covers writing, Quick generation, and validation.

Card format

Include all fields; write “none” where absent:

FieldContent
IDE.g. P01, mapped to the requested image/source shot.
Model/capabilityE.g. Seedance 2.5 multimodal video; GPT Image 2 editing.
PurposeOne-sentence intended use.
PromptComplete copy-ready text in one code block.
ReferencesOne row each: filename, submission order, role below, responsibilities, exclusions.
ParametersDuration, aspect ratio, resolution, etc., from creation_generation_models; otherwise “unverified.”
InvariantsUnchanged people, products, text, composition.
RationaleTwo/three sentences linking choices to model-writing rules.
Acceptance checksE.g. identity, verbatim text, absence of teleporting motion.

+Write in the target language. Chinese-user requests default to Chinese unless the model guide/user calls for English. Preserve dialogue/on-screen text verbatim; never translate/polish it.

Quick generation

+Follow Flova Skill's Quick/reference guidance:

  1. Query the model through creation_generation_models, detail="full"; verify parameters, reference slots/roles.
  2. Prepare references: supported aspect ratio/padding, crops, formats.
  3. creation_upload_ticket → transfer bytes to its URL → creation_upload_complete → file_ref. It expires in roughly one hour; upload again before submission when necessary.
  4. Assign each inputs role: first_frame starts matching that image; last_frame ends on it; reference guides appearance/style/content without exact matching; driving_audio drives lips/action; base_video is edited/continued.
  5. Submit with client_request_id; retain card/reference/file_ref/task-ID mappings.

Referencing inputs in prompts

  • Use ordinal natural language, e.g. first image controls face/clothing, second controls product. Match inputs order within each media type; images/videos/audio each count from first.

Other tools

+Remove Flova-specific markers/role instructions and use the destination tool's upload mechanism and parameter controls. Explain that these prompts target a particular model; changing models may change outcomes.

Optional comparison

+Explain quota use before proceeding. Generate original and revised prompts once each with identical parameters/references; show side by side. Change one variable per round (e.g. reference roles or timeline), record edits/task IDs/results, and deliver the record. Generation is stochastic: one comparison shows a tendency; repeat important findings once or twice before concluding.

Proceso creativo

  1. Flova

    Flova

    Generate a test video

    Test the revised prompt and inspect shot continuity and dialogue.