Character Studio
Build a reusable character in Character Studio. Generate and approve a portrait, add angles, expressions, poses and motion clips, then give it a voice.
Character Studio is the full-screen editor where you build a character asset: its portrait, its pictures and clips, its voice and its personality. You build a character once. Every image or video node you connect it to then works from the same approved pictures, so the person looks the same in every shot.
Open the studio
- Press Tab on the canvas and choose Assets › Characters › Character Asset.
- Click Studio on the node. You can also select the node and click Open Character Studio in the settings panel.
The studio opens full screen on the Profile page. Close it to go back to the canvas. The Character Asset node then shows a summary of the character.
What each page does
The sidebar groups the pages in the order you usually work: the character's sources, identity, visuals and voice.
| Group | Page | What you do there |
|---|---|---|
| Resources | References | Upload real photos of the person, one per angle. |
| Pickers | Choose structured appearance and wardrobe details. | |
| LoRA | Train a custom model on the character's pictures. Nodaro Cloud only; see Character training. | |
| Identity | Profile | Set the name, portrait, description, style and identity lock. |
| Appearance | Generate head angles, body angles and lighting variations. | |
| Visuals | Expressions | Generate the face with different emotions. |
| Poses | Generate the full body in different postures. | |
| Motions | Generate short video clips of the character moving. | |
| Emotion videos | Play the reference clips stored for each emotion. | |
| Sheet | Compose a turnaround or a full reference sheet. | |
| Board | Compose a dense identity board from the character's pictures. | |
| Character | Voice | Choose a voice and hear the character speak. |
| Personality | Write notes on temperament, speech and movement. |
A number next to a page counts its pictures or clips. A check mark shows that Voice or Personality is filled in.
Build a character, step by step
Add reference photos and pickers first
On References, upload real photos of the person. On Pickers, choose appearance attributes and wardrobe. Both are added to every portrait and picture the studio generates, so the first portrait is closer to what you want.
Fill in the profile
On Profile, name the character and describe the face, body, hair and clothing. Choose the style and the image model.
Generate and approve a portrait
Choose 1, 2 or 4 Candidates and click Generate. Click the candidate you want to approve it. The portrait is the anchor for every other picture of the character.
Build the identity anchor set
On Appearance, generate the five head angles. Every picture you generate afterwards uses them as references, so the face stays the same.
Add expressions, poses and motion
Generate the expressions, poses and clips your story needs. Then connect the Character Asset node to your image and video nodes, as described in Consistent characters.
References and pickers
References has seven slots for real photos: Face, 3/4 L, Profile L, Profile R, 3/4 R, Body and other. Click a slot to upload a photo. The six named slots hold one photo each, and a new upload replaces the old one. The other slot holds as many photos as you add.
Real photos give the models extra information, so generated pictures stay closer to the real person. Nodaro sends each photo to the generations it fits. For example, the Face photo guides expressions, and the Body photo guides poses, body angles and motions. After you change the photos, the next generation uses the new ones.
Pickers has two structured pickers. Their choices are saved with the character and added to its own portrait and picture prompts, with no connection to make.
- Appearance attributes — hair, eyes, build, age, ethnicity, skin, face and more. It is the same catalog as the Person picker.
- Wardrobe — archetype, top, bottom, outerwear, footwear, headwear, accessories, color palette, material and era.
Pickers and reference photos affect only the character's own generations. They are not sent to other nodes.
The Profile page
Generate and approve the portrait
- Choose how many Candidates to make: 1, 2 or 4. The Generate button shows the total credit cost. Each candidate costs one image on the chosen model.
- Click Generate. With one candidate, the result becomes the portrait when it is ready.
- With two or four candidates, click the one you want (Click to approve). Approving stops the candidates that are still running.
Approving costs no credits. It also writes the Canonical description: a visual description of the character, 80 to 120 words long. The studio adds it to every picture it generates, and connected image and video nodes can add it to their prompts. Open it to read or edit it, up to 4,000 characters. Click Regenerate description from current portrait to write it again from the current portrait.
Candidates you generated but did not approve stay in a Previous candidates strip for 7 days, up to 5 of them. Approve one of them later if you change your mind. Point at the approved portrait to enlarge it or copy its URL.
Identity fields
| Field | What it does |
|---|---|
| Character name | The character's name. It also becomes the name you mention: Maya Chen is @maya-chen:1. A new character starts as "Untitled character", so rename it. |
| Appearance description | The face, body type, hair and clothing, in your own words. |
| Gender | male, female or other. |
| Style | realistic, anime, 3d-pixar or illustration. Use the same style for every character in a project. |
| Base outfit | The character's usual clothes, in free text. It is combined with the Wardrobe picker. |
| Identity lock | How strongly generations keep the face: Off, Soft or Strict. Off is the default. See Identity lock. |
| Reference image URL | The address of the portrait. Paste the URL of a picture you already have to use it as the portrait. |
| Image model | The model for the portrait. Nano Banana Pro is the default. |
| Seed Prompt | A short identity description for the portrait, such as physical traits, demeanor and distinctive features. Up to 4,000 characters. |
Appearance: angles and lighting
The Appearance page has three sections. Approve a portrait first, because the angles are generated from it.
| Section | Presets | Framing |
|---|---|---|
| Head Angles | front, 3/4 left, left profile, right profile, 3/4 right | Head and shoulders |
| Body Angles | front, 3/4 left, left profile, right profile, 3/4 right, back | Full body, standing naturally with the arms at the sides |
| Lighting Variations | daylight, night, dramatic | The same character under different light |
The box at the top counts the Identity anchor set: the five head angles. Once you generate them, every picture the studio makes afterwards uses them as references. Build the set before expressions and poses, and faces stay the same across all of them.
Expressions and poses
Expressions shows the face with different emotions. Poses shows the full body in different postures.
| Page | Presets |
|---|---|
| Expressions | neutral, smile, angry, surprised, sad, talking, laughing, disgusted, fearful, smirk, crying |
| Poses | standing, walking, sitting, running, crouching, pointing, fighting stance, jumping, turning |
The bar at the bottom of each page works the same way:
- Click a preset to generate it. Each preset shows its credit cost.
- Choose the model: Nano Banana Pro (the default), Nano Banana 2, GPT Image 2, GPT Image 2.5 Flare, GPT Image 2.5 Sunburst or Seedream.
- Generate All queues every preset that is still missing. The studio asks you to confirm when four or more generations would start.
- Custom prompt at the top of the page generates any other variant from your own words.
- Real-life refs attaches real photos to one preset, for example a real laughing face for laughing.
- Import adds a picture you already have from its URL, with no generation.
Work with the cards
Point at a card to see its actions.
| Action | What it does |
|---|---|
| Regenerate same | Makes a new take of the same variant and replaces the card. |
| Add variation | Makes a new take and keeps the old card, so you can compare the two. |
| img2img refine | Changes the picture with a prompt, such as "more intense smile". Choose Replace or Add as new. |
| Rename | Click the name to edit it. The name is the tag that identifies the variant, for example in @maya:1:smile. |
| Set as default for this character node | Makes this picture the node's default. The node shows it, and connected nodes receive it instead of the portrait. Click again to unset it. |
| Add as node on canvas | Places the picture on the canvas as its own node. |
| Enlarge, copy URL, delete | View the picture full screen, copy its address, or remove it. |
Each regenerate, variation and refine costs the same as a new picture on the chosen model.
Motions
Motions holds short video clips of the character, animated from its pictures. Generate a portrait first; without one, Generate is disabled.
- Presets: walking, running, waving, sitting down, fighting stance, jumping, turning around, dancing, talking gesture.
- Models: Kling 2.6 (the default), Kling 2.5 Turbo Pro, Kling 3.0, Wan 2.6 I2V and Wan 2.7 I2V.
- Cost: a clip costs the same as an image-to-video run on the same model.
A full-body picture animates much better than a head-and-shoulders portrait. So Nodaro animates the front body angle when the character has one. When the character has no body angles yet, the studio first generates a front body angle, then animates it.
Motion cards have Regenerate same and Add variation, but no refine and no Generate All. Import adds a clip you already have from its URL.
Emotion videos plays the reference clips stored for each emotion, such as angry or happy. The page only plays clips; it does not create them.
Sheets and boards
Sheet composes a reference sheet from the character's pictures. Choose a preset:
| Preset | Panels |
|---|---|
| Studio · Main | A clean turnaround for AI video reference: the head from the front, both profiles and the back, and the body from the front, the side and the back. 7 panels. |
| Studio · Extended | A full turnaround that adds 3/4, above and below angles. 15 panels. |
Under Add boards, add bands for Expressions, Poses, Wardrobe, Detail and Palette. A sheet holds at most 24 panels, so a board is disabled when it would go past the limit. You can also change the Skin, the Aspect, Title / metadata and Panel labels.
- Click Prepare N angles, where N is the number of missing panels. Only those panels are generated, and they are saved to the character for reuse. The line above the button shows how many panels are reused and what the rest costs.
- Click Compose sheet to assemble the sheet. It is available once no panel is missing.
Open a sheet full screen, copy its URL, or click Set as node thumbnail. The Reference Sheet node builds the same kind of sheet on the canvas.
Board composites 2 to 12 of the character's pictures into one dense 4K identity board. A board is the strongest single-image anchor you can give a model.
- Click New board. Every picture the character owns appears, grouped by kind.
- Select 2 to 12 pictures. A number on each one shows its place in the board.
- Click Generate. A board costs the same as a 4K Image Collage.
A character holds up to 12 boards. Duplicate reopens the picker with the same pictures selected, and saves the result under a new name. Click a board's name to rename it. Cancel on a board that is still generating refunds its credits. In the @ mention menu, a character's boards are listed first. For boards and grids made on the canvas, read Reference boards.
Voice and personality
Voice gives the character a voice that other nodes reuse.
- Browse — choose an ElevenLabs voice from the premade voices, the public library or your own custom voices. Search and preview each one.
- Design from text — describe a voice, write a preview line of at least 100 characters, and click Audition to hear it. A designed voice is for listening only; it is not saved.
- Voice traits — describe the voice in free text on the selected voice, for example "deep, calm, British accent".
- Talk — type a line and click Speak to hear the character say it. With an approved portrait, Speak + lip-sync portrait renders a short talking clip. Speak costs the same as a Text to Speech run, and the lip-sync clip costs the same as a Lip Sync run.
Connect the character to a Text to Speech node, and the character's voice fills in that node when the workflow runs. A voice you set on the Text to Speech node itself always wins.
Personality has four text fields: Mood / Temperament, Speech Style, Movement Style and Behavioral Notes. They are saved with the character, but they are not added to prompts.
Saving and versions
- Saving is automatic. The header shows Saving… and Saved. There is no save button.
- Generations finish without you. Each picture or clip is added to the character when it is ready, even if you close the studio or refresh the page.
- Versions stay side by side. Add variation keeps every take of a variant, and Previous candidates keeps unapproved portraits for 7 days.
The Character Asset node on the canvas
The node is a compact summary card:
- The portrait, or the picture you set as the default, with the name, style and gender.
- A Role menu: the part of the character that connected nodes take by default. See Reference roles.
- An Identity lock row: Off, Soft or Strict.
- Counts for Expressions, Poses and Motions, and marks for Voice and Personality.
- Studio, which opens the studio, and Choose existing, which binds another character.
- On Nodaro Cloud, a Training… badge while a model trains and a green Trained badge when it is ready.
Point at the picture to choose its shape: 1:1, 3:4, 16:9 or 9:16. The shape also sets the aspect ratio of the character's next generations.
The node has two outputs. Character carries the full identity: the pictures, the canonical description, the role and the identity lock. Image is only the picture, for any input that accepts an image.
The settings panel adds Inject identity description in downstream prompts. When it is on, connected image and video nodes add the canonical description to their prompts. It needs a canonical description, so approve a portrait first.
Use a character you already have
Click Choose existing on the node, or Choose from Library / Gallery in the settings panel.
- My Library lists your own characters.
- Public Gallery lists characters shared by the community. Choosing one copies it into your library first, then connects the node to the copy. The tab appears on multi-user installs, such as Nodaro Cloud.
- If you already copied a gallery character, the picker asks whether to use your copy or Make a new copy.
Once a character is bound, the buttons read Replace and Replace from Library / Gallery. Replacing brings every picture, clip, voice and personality of the new character, so connected nodes use it at once.
Delete and restore a character
In the My Library tab, point at a character and click Delete from library. The character is archived, not destroyed. Nodes that already use it keep working.
An archived character can be restored with the API, the SDK or the CLI. If another active character has the same name, the restored one gets "(restored)" added to its name. On Nodaro Cloud, deleting a character also cancels a training in progress, refunds it, and deletes the trained model.
Tips
- Set pickers and references before the first portrait. They shape the portrait, and every later picture follows the portrait.
- Approve before you branch out. Pictures generated before an approved portrait are less consistent, and motions need one.
- Name the variants carefully. The names are the tags you mention, such as
@maya:1:smile. - Refine instead of rerolling. Use img2img refine with Add as new to change a detail and keep the original.
- Keep one style per project. Mixing styles across characters breaks the look of a story.
From code and agents
Everything in the studio is also available to code and AI assistants. The REST API, the SDK and the CLI create characters, generate and approve portraits, add pictures and clips, and archive and restore characters. See Characters.
The MCP tools include create_character, update_character, generate_character, approve_portrait, recaption_character, generate_character_motion, list_characters and get_character. The MCP tools cannot delete or restore a character. See the MCP tools.
Frequently asked questions
Related
Character Asset
Consistent characters
Character training
Reference boards and consistency grids
Characters
Last updated on
Consistent characters
The complete method to keep the same person, pet, product or place in every image and video, with assets, approved pictures, mentions and identity lock.
Locations
Build a reusable place in Location Studio with an approved main image, time of day, weather and angle variants, and motion clips, then reuse it in every scene.