Creative Space

Konten

Provide: Photo, description, or scripted character; intended style/use (short film, advertising, series).

Receive: The requested character pack.

Clarify identity/styling, generate and test in Flova. Reuse references/descriptions later. Deliver assets, not a film. Scope one character, not multi-person blocking or unrestricted character brainstorming: the user supplies direction for background/personality, the agent develops/confirms it.

Tools/references

  • Flova, required: all appearance/sheet/expression/look/voice/trial generation. Repeat pose/light/background/aspect constraints in every relevant request, not separate post-processing tools.
  • Optional Teardown — retrieve and read the full “Break Down a Reference Video into an Interactive Report” inspiration through Flova MCP by name: reusable appearance/style traits from supplied references.
  • , steps 3/5: reasons, five styling layers, descriptions, period/genre, multiple looks.
  • , step 4: photo/description to views/close-up/expressions, ratios/checks.
  • , steps 6–8: contents, fixed description, downstream use, trials/drift corrections.

Workflow

Confirm identity/use → biography → appearance → views/expressions → optional looks → optional voice → optional trial → pack. The first four stages and voice choice are confirmation points.

1. Confirm identity/use

Reuse supplied answers. Ask name/identity/content type (drama/ad/MV/social/game/animation); existing photos/design/script/reference; style (realistic/cinematic/anime/3D/illustration); desired modules; aspect ratio, speaking needs, likely close-up/full-body/group-shot contexts. Once chosen, keep style consistent across the pack. Produce only requested modules. Ask at most three genuinely design-critical follow-ups; propose options for other gaps.

2. Biography — confirmation

Before drawing, confirm a one-page profile: identity/apparent age/era/region; personality and speech style (two/three sentences each); one memorable habit/object/visual trait; visual temperament/posture/expression; invariants e.g. an old left-eyebrow scar or red scarf that persist in every shot.

3. Appearance — confirmation

Use styling reference: describe the base look and reasons before a front-facing half-body image. A user photo anchors the unchanged face while clothing/hair/makeup follow design. Without a photo, offer two/three facial directions. Once approved, that face anchors every later image. Resolve facial dissatisfaction here, before sheets.

4. Views/expressions — confirmation

Follow character-sheet: front/side/back plus face close-up. Performing characters also need neutral, happy, angry, surprised, sad and one/two personality-specific expressions. Ask whether all views match each other and the approved appearance; redo mismatches.

5. Extra looks — optional

For different occasions/story stages, follow styling's multi-look method: one full-body image per outfit, changing one styling layer at a time. Repeat the same pose/light/solid background in each request; deliver individual images.

6. Voice — optional confirmation

For speaking characters, generate three timbres with the same character-appropriate line (not a generic test sentence). Let the user audition/select; record voice name and audition text.

7. Trials — optional

Following consistency-pack, generate three short shots varying shot size, angle, and light. Check the same identity; correct instability using its method and retest.

8. Pack

Deliver biography/invariants; fixed reusable description; approved appearance, three views, close-up, expressions, looks; reference inventory with filename/purpose/reusable asset identity; voice/trials if made; instructions for attaching the assets to future shots.

character-<name>/
├── source/      User photos, designs, references
├── docs/        Biography, fixed description, inventory, instructions
├── looks/       Approved appearance and outfits
├── sheet/       Three views, close-up, expressions
├── voice/       Candidates and audition lines
└── trial/       Consistency clips

Three views, faces and expressions

If you need the same person to appear in multiple shots, first make him a set of reference pictures, and then use it to generate all shots. A photo can only tell the front face of the model; the three-dimensional view determines the body proportions and angles, the close-up of the face determines the details of the facial features, and the expression set determines how the person performs.

Input

  • Final makeup photo: The one confirmed by the user in step 3 is the facial basis for all subsequent pictures.
  • Photos: Photos given by users should preferably be front-facing, evenly lit, and with a clear face. If there are no photos, use the fixed makeup photo.
  • Outfit: Use the outfit you decided on when setting your makeup. Other styles are done separately, see "Makeup Design".

Generate

Using an image model that supports reference images, using the makeup photo (and photo) as a reference, the character is labeled reference:

Three views

  • Full-length portraits of the same person from the front, side and back, standing side by side on a pure light gray background, standing in a natural posture with arms relaxed.
  • Scale: The entire three-view image is output at 16:9. Don’t create very wide strips: some video models only accept reference images with an aspect ratio within 2.5. When wider is required, split into three separate drawings.
  • Lock statement: "Keep the face, skin color, hairstyle, and body shape of the person in the reference picture exactly the same, and do not change the identity." Use verbatim for each regeneration.
  • **Light:**Uniform and soft, without strong shadows and color temperature deviations, making it easy to put into different scenes later.Facial close-up
  • A frontal close-up of the face (head to shoulders), neutral expression, evenly lit.
  • If the character has iconic facial details (scars, moles, freckles), make sure they are visible in close-ups.

Emoji Collection

  • The same person, the same composition (head to shoulder) and lighting, expressions are generated one by one: calm, happy, angry, surprised, sad, plus one or two characteristic expressions that fit the character (e.g. "restrained impatience" and "suppressed laughter").
  • One picture per expression, not in a row; each picture uses a close-up of the face as a reference, only the expression is changed.
  • Write expressions as specific facial movements (eyebrows, eyes, corners of mouth, chin), not just emotional nouns.

Check (confirmation point)

  • The three angles are the same person: the face shape, hairline, hairstyle, skin color, and body shape are consistent.
  • The sides and back are reasonable: there are no patterns or words that appear out of thin air on the back of the hairstyle or the back of the garment.
  • Unlike makeup photos: show it to the user and use it only after approval.
  • The iconic details in the close-up of the face are all there, and the positions are consistent with the three-dimensional view.
  • Each expression in the expression set is the same person, only the expression is different.
  • There are no deformities in the hands, feet, or fingers.

If it doesn’t look like it or there is a problem, regenerate it; if it still doesn’t look the same after changing it twice, go back to the makeup setting step and ask the user to change to a clearer photo or adjust the facial description.

How to use it in shot

  • When the reference picture: When generating each shot or keyframe with a character, pass in the three views (plus the required close-up of the face) as the reference picture, and write clearly in the prompt text "The first reference picture is the three-view view of the character, only responsible for the face, hairstyle, body shape and clothing."
  • Fixed description: Bring the fixed description in the character package and reuse it verbatim for all shots.
  • Speak: Keyframe is used as the first frame (first_frame), voiceover is used as the driving audio (driving_audio), and the lip sync model is used to generate it; or a video model that can generate dialogue together is used to write the original text of the dialogue into the prompt.
  • For details on how to reuse and verify, see .

Character Packs and Consistency

The character package is the deliverable of this inspiration: the character's personality, image, voice and usage rules are put together so that you can use them the next time you make a video.

What’s in the Character Pack?

ContentDescription
Character biographyWithin one page, including "things that cannot be changed"
Fixed DescriptionA 100–200 word physical description that can be reused verbatim (see below)
Final makeup photoConfirmed face, the basis for all subsequent pictures
Three views and close-ups of facesBody proportions and facial features details
Expression setOne for each expression
Each set of looksOne full-body picture for each set, numbered L01, L02...
SoundSelected tone name, audition dialogue
Reference picture listEach picture: file name, purpose, reusable assets identification (see below)
Audition filmThree verification shots and comparison frames (if done)
InstructionsHow to bring this set of assets next time

How to write fixed description

  • Write in the order of "age and gender, face shape and facial features, hairstyle and hair color, skin color and body shape, current style, and iconic details".
  • Only write about the appearance, not the emotions and plot: the emotions are responsible for the prompts of each shot.
  • Be specific about numbers and colors, and use the same word consistently (don’t write “brown hair” here and “chestnut hair” there).
  • Each shape has a separate section; all shapes for the face and body parts are the same.
  • Verbatim reuse: copy the same paragraph in each shot, do not rewrite it every time.

How to remember the reference picture list

One line for each picture: file name, type (fixed makeup, three views, face, expression, style L0x), what it is responsible for (face, body shape, clothing), what it does not inherit (background, light, pose), and assets identification.

  • assets ID: The uploaded file_ref will expire in about an hour and cannot be saved for a long time. The generated result has its own assets identifier (resource_id), which can be reused as a reference in subsequent requests and recorded in the inventory. After the uploaded original image expires, replace it with the generated result or upload it again.
  • The aspect ratio of the reference image should be within the range allowed by the target model. If it is too wide, divide it into several images or fill in the edges first.

How to use it in subsequent shots

  1. For each shot with this character, a three-dimensional view and a close-up of the face (when in close-up) are input, and if necessary, a picture of the outfit that the shot will wear.
  2. Use text to refer to the prompt: "The first reference picture is a three-dimensional view of the character, only responsible for the face, hairstyle, and body shape; the second picture is the current look, only responsible for the clothing." The order must be consistent with the order in the request.
  3. Paste the fixed description, and then write the action, scenery and light of this shot.
  4. When the shot requires the character to speak, see the method in "Three Views, Faces and Expressions".
  5. Later, use the same shot of multiple characters in other videos (this Studio does not do this step): each character has its own set of reference pictures, and describes them separately in the description. Be careful not to let the characteristics of the two people cross each other.

Audition Film: Verify Consistency

Use three shots to verify, each changing only one variable:

  1. Close-up dialogue: Frontal or side view, neutral light, character speaking a dialogue. Check whether the face looks like the makeup photo and whether the lip sync is natural.
  2. Full body movement: Panorama or medium shot, the character takes a few steps or turns around. Check the body shape, the back of clothing, the back of hairstyle.
  3. Different light: For example, warm sunset or cold light at night, middle and close shots. Check to see if skin color and facial features are still recognisable.

Comparison: Check the three shots, makeup photos, and three views one by one: facial features, hairline, skin color, body shape, and iconic details.

What to do if it drifts?

PhenomenonPractice
The face slowly changesAdd facial feature writing reference and regenerate; reuse fixed description verbatim; reduce other changes in the same shot
The color or style of clothing has changedPass in the styling drawing separately as a reference, and only write one sentence about clothing in the prompt
It doesn’t look like changing the angleAdd a reference picture of the angle (e.g. a close-up of the side face) and add it to the character package
When the light changes, the person changesBack to the uniform light, first generate the keyframe, and then add light changes to the keyframe
Face deformation when speakingSwitch to "first frame plus voiceover driver", set the character in the first frame, and the lip sync model is only responsible for the mouth
The iconic details are lostWrite it separately in the prompts and acceptance points, and use close-up reference as needed

Only change one thing at a time. After the adjustment, run the same shot again to compare and determine which part worked. The generation is random, try one or two more times for important conclusions.

Instructions for use given to users upon delivery

  • Steps for making a video next time: read the character package first, choose a look, which reference pictures to bring, and which fixed description to paste.
  • Known weak points (e.g., a large profile face is prone to distortion), as well as effective practices found in audition films.
  • When you want to add a shape, expression or voice, go back to the matching steps and continue without having to redo the confirmed parts.

makeupdesign

Makeup is not about "good-looking clothes", but allows the audience to tell at a glance who this person is, what kind of life he lives, and what state he is in at the moment. Think about the reason first, then write the description, and then generate it.

First think about "why you wear it like this"

Write a reason for each choice. If you can't give a reason, delete it or choose again. Start from these questions:

  • Identity and Occupation: What does he do? What attire is required for work?
  • Economy and Class: Are the clothes new or old, bought by yourself or given by others?
  • Era and Region: When and where?
  • Character: Does he care about dressing up? Is it careful, casual, or deliberately low-key?
  • Situation at the moment: Which plot stage is this outfit in? Just went out, or stayed up all night?

The reasons are written in the document. Users can understand the design intention and know what cannot be moved when changing the shape later.

Five levels

Each set of styles is written following these five layers and confirmed layer by layer:

LevelWhat to write
Hair style and colorLength, texture, parting, condition (neat, messy, updo, wet hair), hair color and gradient
MakeupThe texture of the base makeup, eyebrow shape, eye makeup, lip color, and blush; the male character also writes skin type, beard, and eyebrow shape
ClothingStyle, fit, fabric, color, layer (inner wear, outerwear), newness and wear
AccessoriesJewelry, glasses, hats, bags, belts, scarves, watches, and their materials and placement
Traces of statusTraces of age, scars, sweat, dust, stains, wrinkles, only write what is needed for the plot

How to write a description

  • Specific enough to draw: "Dark gray thick-knit wool cardigan with raw edges at the elbows and a washed-out white T-shirt underneath" rather than "casual and tasteful".
  • Use objects and materials, no adjectives: "Good-looking", "fashionable" and "high-end" have no information.
  • Write the color clearly: "Caramel color" and "haze blue" paired with a common comparison are more stable than writing "brown" and "blue" alone.
  • One paragraph for a set of looks, write in the order of "hairstyle, makeup, clothing, accessories, status" and reuse it verbatim later.
  • Write what you can actually see in the picture. If the inner layer is completely covered by the jacket, there is no need to write details.

Era and type

  • Modern Realism: Pay attention to seasons, climate and occasions.
  • Ancient costumes, the Republic of China, and historical themes: First ask about the Qing Dynasty or the year and region, check the information to confirm the shape, and do not piece it together based on the "ancient sense". If you are unsure, mark "to be verified" in the document and let the user decide.
  • Science Fiction, Fantasy: Write the materials and functions in the worldview first (why this dress looks like this), then the appearance, to avoid stacking cool elements.
  • 2D, animation: Simplify details, highlight outlines, color blocks and iconic elements. Hair style and color matching are more important than fabrics.

Multiple looks for one person

When the same person changes his appearance, what changes is the clothes, not the face.

  1. Use the confirmed makeup photo (or three-view front view) as a facial reference, and write "Keep this person's face, skin color, body shape, and hairline unchanged."
  2. Only record the levels you want to change, e.g. only changing clothes and accessories, leaving your hair and makeup unchanged.
  3. Only change one level at a time. Change it all over again, the face is the easiest to change.
  4. Each set generates a full-body front and a half-body, keeping the same standing posture, light and background for easy comparison.
  5. Check after generation: whether the face, hairline, skin color, body shape, and iconic details (scars, scarves, etc.) have changed; whether the description of the new shape has been implemented; whether the clothing has extra patterns or text out of thin air.
  6. Give each set of looks a name (e.g. "Daily", "Appearance in Court", "Third Day of Escape") and numbers L01, L02..., and use the same number in files and documents.

Checklist

  • There is a reason for every choice, and the shape can read the identity and situation.
  • Be specific in your description, no general adjectives.
  • The shapes of ancient costumes and historical themes have been verified, and those that are uncertain have been marked "to be verified".
  • The same person in each outfit is clearly the same person.
  • "Things that cannot be changed" are retained.

Proses kreatif

  1. Flova

    Flova

    Generate character images

    Create character looks, multiple views, expressions, and outfits.

  2. Flova

    Flova

    Generate audition clips

    Create clips of the character speaking or acting and inspect consistency across shots.