Characters
Character entities -- properties, media, voice configuration, and profile.
A character is an entity that represents a person, creature, or figure in your story. Characters are created at the project level and reused across all episodes.
The cast has a Character level of its own in the sidebar, and All characters is its first row -- that is this page. Under it sit the pages of one character: Properties, Media, Voice and Profile.
To add one, open All characters, click New character, and type a name -- a short ID is filled in for you from the name, and you can keep it. You stay on the list, because creating is usually a batch act: add your whole cast in one go, then cast them into the episode using the In this episode column. Click a character's row when you are ready to open it -- it opens on Media, and Properties in the sidebar is where you fill in description and look. Whichever character you have open, the breadcrumb at the top of the page names them, and the sidebar's Character pages stay on that one until you pick another from All characters. The Characters page lists every character in the project, with star, rename, and delete actions.
Cards, list and looks
One control at the right of the controls row chooses how the page is drawn, and it has three settings: Cards, List, and Looks. Whichever you pick is remembered for next time, and Characters, Locations and Props each remember their own.
Cards is what you get to begin with -- a grid of pictures, which is how you recognise a character long before you read a name. List is the table. Looks shows the same assets with their looks nested underneath.
A card shows the character's picture with their name underneath, and under the name the In this episode control: choose Major or Minor to cast the character into the open episode. The List's In this episode column is the same setting, so a change in one shows in the other. With no episode open the control is greyed out. You never choose that picture: it is the starred image from that character's Media gallery, or the newest image in it when you have not starred one. Star a different image and every card follows at once. A character with no images shows a plain card.
Cards carry the same star and "..." menu as the rows, in the same places as a media gallery card: star on the left, menu on the right.
Media
The character's Media Gallery holds generated images and visual references. Star the images you are happy with. The one picture that IS the character is its reference image (below); where there is none, the starred image is used as the reference input when running operations that include this character.
Mark the one image that IS this character with Set as reference in the item's ... menu -- the Assistant uses it whenever a request needs this character's likeness. See the reference image. You can also choose or clear it on the Properties page: the Reference image row under Appearance shows the picture with an x in its corner to clear it, and when none is chosen, tapping the empty slot opens the Media Browser on this character's gallery. Tapping the picture opens it full screen.
From the gallery you can generate character imagery. The operations you will use most on a character are listed under Operations below.
Poses
A pose is a name on a media item -- "Standing, hands behind back", "Arms crossed, leaning on the desk". You name a pose on the item itself: open its menu, choose Edit Metadata, and fill in the Pose field, then press Save (or Cancel). As you type, the poses this gallery already has are listed, so you can pick one to keep the spelling consistent (and keep editing it). The X in the field clears it, and saving an empty field removes the pose from the item.
The name belongs to the item, so two items can have the same pose name -- the name goes with the item if it is moved. Every look remembers its own preferred item for each pose name (a look is usually a different outfit, so another look's clips would show the wrong clothes); that is the item a Shot and the Assistant use. Name an item with a pose the look has no preferred item for and it becomes the preferred one; if the look already has one you are asked whether to keep the current one or use this one. When a pose name has more than one item, the preferred one shows a green tick and the others have Make preferred pose in their menu. Delete the preferred item and the newest remaining one takes over. Poses work the same in every gallery -- an experiment, a scene or a Shot can name its own, to try things before choosing the winner -- and Edit Metadata also has a Camera angle field for camera angles.
A pose name shows wherever tags show: on the grid card, on the details row and on the Info card, cut with "..." when it does not fit. The filter button beside the sort controls has Poses and Camera Angles; a tick means the filter is on, and the gallery then shows only items that carry that kind of name (both on shows either).
When the AI plans a scene's shots, it chooses a pose for each character from the poses of the look that character wears in that shot -- only the look's preferred ones, never another look's. The pick is kept on the shot as the pose its start frame uses; a pose you already chose is never replaced (a different suggestion waits beside it with Accept and Reject). When none of the poses fits, the AI asks for a new one instead, with the reason: the shot shows it as a proposal with Make it. Pressing it opens this character's look with the turn-around recipe below loaded and the Pose filled in, and the guide bar walks you through it one step at a time -- nothing is generated until you press Run step 1. When the pose picture is ready the bar shows it with what to look for and two choices: Looks right, next step or Try again (which makes a new take). Only then does it offer Run step 2, and when the spin clip is ready you finish with Use as the pose, or you can cancel. While a step renders the bar points you at another request to do meanwhile, and a notice tells you when the result is ready. Run all is not offered during this.
To make a turn-around sheet, load the Pose a character without props, then spin it around recipe from the cookbook. It has one input, Pose: type the pose, or pick one the gallery already has from the list (the X clears it). The pose step poses the character from the look's reference picture (already in its first image slot -- set the reference picture first, or choose one yourself). A pose that names an object, such as sitting on a chair, is always drawn with that object, so the next step, Remove everything but the character, takes it away: everything around the character is painted plain grey, so a chair or table goes without the character moving. Something held in the hands is named in that step's Objects To Remove -- filled in from the Pose for you ("holding a coffee cup" gives "coffee cup") -- and is taken out of the hands too, with only the gaps it left in the body filled in. The spin is then told the hands hold nothing, so the object does not come back. The spin step then turns that picture into a clip in which the camera orbits the character, using the look's reference picture too, so the back and sides match it. To keep the chair, table or prop in the turn-around, load Pose a character with props, then spin it around instead: the same recipe without the erase step. The words around the pose are built into the recipe, so there is no prompt to write, and the recipe sets the clip's length (5 seconds, which the orbit is timed for) and both pictures' sizes itself -- a line under each step says so, and those inputs are not shown. When you run the recipe yourself, the spin clip is named with the Pose by itself when it arrives in the gallery (in a Shot's gallery too -- the recipe can run from there; in a look that already prefers another item for that pose, a message says so and Make preferred pose switches). When the guide bar walks you through it, the clip is named only when you press Use as the pose, and Ordinary Animator also works out in which frames the character faces the camera, left, away and right (desktop only), so a Shot can later ask for "facing left". To pose a character from an exact skeleton instead, see Pose the character exactly.
Operations
Operations are picked from Choose an operation in the character's Media Gallery. These are the ones you reach for most on a character; the gallery offers the full set.
Properties from an image
If you already have a picture of this character, add it to the gallery and pick Image to Text (Gemini) in the operation picker. A vision model describes it, and the description lands in the gallery as a text item; Extract Properties from Text then turns it into proposed values for the character's Appearance properties, which appear at the top of the Properties page for you to accept or reject one by one. See working backwards from an image you already have.
Pose the character exactly
To put this character in a specific pose, give the pose as a pose skeleton -- a stick figure in the standard pose colours -- rather than as a photo of someone else. A photo carries that person's face, body and clothes, and the model blends them into your character; a skeleton carries only the pose.
- Get a skeleton. Pick Extract Pose (DWPose) on a photo of a person in the pose you want, or make one with 3D Pose Placement or the webcam pose recorder.
- Pick Pose Transfer (Qwen Image Edit 2511). Put the character's picture in Character Image and the skeleton in Pose Image -- there is no prompt to write.
The result keeps the character's face, build, clothing and art style, in exactly the skeleton's pose -- even a pose that body would struggle to hold. For a looser pose that the model adapts to the character, use References to Image with the pose photo instead.
Pose Transfer (Qwen Image 2.1) does the same several times faster, but its model is licensed for non-commercial use only -- use the Qwen Image Edit 2511 version for anything commercial.
Load as Operation
Load as Operation on any result rebuilds the operation that made it -- same prompt, same settings, same images fed in -- so a look you like can be run again for another character. It restores that one operation, not the whole history behind the item. See do that again.
Properties
Character properties are grouped into cards, in this order down the page:
| Card | What it holds |
|---|---|
| Core properties | Name, description and notes |
| Story properties | Story role, backstory, personality traits, external goals, internal needs, fears and vulnerabilities, quirks and mannerisms, key relationships, and character arc |
| Music | Music preferences by mood |
| Appearance | Visual summary, age range, gender presentation, body type, height class, height, width, face features, skin tone, hair style and colour, eye colour, distinguishing features, wardrobe style, signature colours and accessories, and identity preserve rules |
On a look only Core properties and Appearance are shown: a look is a visual variant of the same character, so it does not carry its own story or music. Appearance starts from the base's values and you override only what differs.
Height and Width are real lengths, used to size the character's stand-in when you block a shot. Type them in metres (1.68) or feet and inches (5'6.3"); either is understood, also when pasted, and the page shows both. Width is front to back (the body's depth). Leave them empty and the stand-in's size comes from the Height class and Body type words instead.
Voice
The Character Voice page is where you configure how this character sounds. Down its left side is an Engines rail, split into Text to Speech and Voice Changers. A character can have several voice engines configured at once -- a green tick marks each one that is set up -- and you mark one as the draft voice from its menu (a bold green tick shows which), the voice the one-press dialogue generation uses for every line.
Under Text to Speech:
| Engine | Notes |
|---|---|
| Google Chirp | Google Cloud Text-to-Speech; deterministic, so re-runs reuse cached audio |
| Google Gemini | Expressive voices that act the line's direction -- see below |
| ElevenLabs | High-quality cloud synthesis, many voices; requires an API key |
| Typecast | Alternative cloud engine; also requires an API key |
| ComfyUI | Run a ComfyUI speech-generation workflow (e.g. Qwen, F5) as the voice |
Google Gemini is the one engine that performs a direction rather than just reading text. Its 30
voices are each offered in Australian, British, Indian and US accents -- pick the accent on the row,
the same way you pick the voice. When a
line in the screenplay carries a parenthetical -- (nervously), (barely holding it together) --
that text is handed to the voice as delivery instruction, in your own words. There is no fixed list
of emotions to choose from; whatever the writer wrote is what the voice is asked to do. Because it
re-performs a line differently each time, generating the same line again gives you a fresh take to
choose between rather than the previous one. The other engines ignore parentheticals.
Under Voice Changers there is one entry, ComfyUI: pick a ComfyUI voice-changer workflow and it turns a recorded take on a dialogue line into this character's voice. You can attach Voice References as the target timbre. A Test Voice Changer panel lets you try it before committing.
See Audio & Voice Pipeline for the one-press dialogue flow, and Building Dialogue from Operations for the underlying model.
Profile
The Character Profile page provides a rich document workspace for reference material -- written character bible, notes, images, links. It's freeform and not structured as form fields.
Looks
A character can have looks -- alternate versions that share Story properties but have independent Appearance properties. Use looks for:
- Costume changes (day outfit vs evening outfit)
- State variations (normal vs injured)
- Age looks (young vs old)
Looks are created from the Properties page and appear as child entries in the asset list. See Looks for how the inheritance works.
Identity LoRAs
An identity LoRA is this character trained into a small add-on file, so a model can draw them the same way every time. Reference images get you a long way; a LoRA is what you reach for when a face has to survive a close-up, a different angle and a change of lighting.
You do not make one in Ordinary Animator -- you train it elsewhere, or someone trains it for you -- and then you upload it into the look's gallery, where it sits beside that look's images as one item. Locations and props work exactly the same way, on their own galleries.
Uploading one
From the look's Media gallery, choose Upload identity LoRA... and fill in:
| Field | What it is |
|---|---|
| Model | The model the LoRA was trained for. This is the important one -- see below. |
| Name | What the LoRA is called in the gallery. |
| The file (or files) | The .safetensors file. Some models are built in two pieces, and those ask for both. |
| Trigger word | The word the LoRA was trained to respond to. Leave it blank if yours has none. |
| Strength | How hard it is applied, from 0 to 1.5. Start at 1.0 and come down if the look starts to overwhelm the picture. |
| In use | Tick this to make it the one that gets applied. |
A LoRA belongs to ONE model. It was trained against that model's exact internals, so a LoRA made for Wan I2V 2.2 does nothing useful on Flux Klein -- not a worse likeness, no likeness. A character who appears in both a video model and a still one therefore needs two, trained separately. That is why the Model field is not a detail: it is what decides when the LoRA applies.
Two files for some models. A few models (Wan 2.2's larger ones) are built as two halves that always load together, so their LoRAs come as a matching pair. The upload asks for both, and a pair with only one half uploaded is not applied at all -- half a likeness is worse than none.
In use
A look can hold as many identity LoRAs as you like -- an older version worth keeping, a retrain, one for each model. Only the one marked In use is applied, and you can have one in use per model: marking a new Wan LoRA as in use clears the previous Wan one and leaves your Flux one alone. The in-use item carries a green tick in the corner of its card, the same mark the rest of the app uses for "the one of a set that gets used". Everything else about it -- Set in use, Edit Metadata (where the strength is changed), Delete -- is in the card's ... menu.
Uploading a LoRA changes nothing on its own. Nothing is applied until you mark it in use.
What happens on a run
When you run an operation on a shot, Ordinary Animator looks at everyone in it -- the characters, the location, the props -- and collects the identity LoRA each of their looks has in use for the model that run is about to use. Each one that applies shows as a chip on the run row, and its trigger word goes to the front of the prompt, which you can see and edit before the run starts.
Removing a chip drops that identity for that run only. Nothing about the look changes.
Afterwards the result's details list the identity LoRAs that were applied, so you can tell months later which likeness made a given picture -- and Run again puts the same ones back.
Why one might not apply
The run row always says which of these it is:
- "This workflow takes no identity LoRAs." That workflow has nowhere to put one. It is not a fault -- most workflows are built without the slots. Pick a different workflow for this operation if the likeness matters here.
- "No identity LoRA for <Model>." The workflow could take one, but nobody in the shot has one in use for that particular model. Either upload one for this model, or run a model you already have one for.
- "2 identities could not be applied: Mabel, The Workshop (3 slots)." More of the cast have LoRAs than the workflow has room for. The ones that did apply are listed as chips, the ones that did not are named here, and nothing is dropped without saying so. Characters are fitted first, then the location, then the props.
Which workflows have room for identity LoRAs is a property of the workflow, and the visual style you are using decides which workflow runs.