Meet Ruby, the presenter in the video above: red hair, freckles, green hoodie, a bright fast voice. Every one of her ten clips was generated separately, on different prompts, and she looks and sounds the same in all of them. That is what an AutoFlow character is for. This guide shows how to make one in the Queue Manager and use it in every queue.
Why a character, not a description
Describing a person again in each prompt gives you a slightly different person each time: the model reads "a woman with red wavy hair" afresh for every clip. A Flow character is saved once — a portrait, a name and a voice — and attached to the clip itself, so the face comes from the portrait and the voice from the character. AutoFlow makes your characters once, in a Flow project called AutoFlow Characters, and every queue borrows them from there; nothing has to be copied into each new project.
Make a character in the Characters tab
Open the Queue Manager's Characters tab and add a character. You need three things:
- A name — the word you will write in prompts. Keep it unique: "Ruby", not "the girl".
- A face, from a photo or from words:
- From a photo — upload a clear, front-facing, well-lit picture. This is free.
- From a description — age, face, hair, clothes, setting. Flow draws one portrait with the image model you choose (Nano Banana 2 Lite is the cheapest; 2.1 and Pro are sharper).
- A voice — see below.
Optionally add a personality ("fast-talking, optimistic, explains AI like a best friend") and choose whether the description is also written into every prompt that names the character. Making the character takes about a minute; its status turns to ready.
A tip from making Ruby: a small or blurry photo makes a blurry character. Our first portrait was a 160-pixel thumbnail, so we drew her from her description instead — one image, and a sharp face for every clip after.
Give it a voice
Pick one of Flow's 30 voices, each described in the list — Leda is "female, youthful, mid-high pitch", Sadaltager "male, knowledgeable, mid pitch". To make it your own, describe the performance: "bright, fast-paced, highly enthusiastic, natural conversational cadence, speaks in English". AutoFlow turns that into a custom voice in Flow and gives it to the character. Saying the language in the performance helps keep the accent you want.
Use it: just write the name
In any video queue using Ingredients (references), write the character's name in the prompt and AutoFlow attaches the character for that clip:
Ruby sits in her LED-lit bedroom studio, leans toward the camera and says:
"What if you could just tell an AI the video you want?" Medium close-up.Put dialogue in quotes after the person who says it. With two characters in a shot, name both and make clear who speaks which line — ambiguous dialogue is the usual reason two characters swap voices.
Scenes that continue
For a scene that flows across clips, put the prompts that continue inside square brackets. Each one starts from the last frame of the clip before it, added as a reference, so position, light and framing carry over — and the character still comes from its saved face and voice.
Characters you already made in Flow
Made characters by hand in Flow? Open that project in your Flow tab and import them into the Characters tab. They keep the face and voice you gave them in Flow, and every queue can use them by name.
Let the agent cast your video
With the AutoFlow agent connected, your AI assistant can do all of this from a sentence: "make a presenter called Ruby, red wavy hair, green hoodie, voice Leda, bright and fast", then "queue my script with Ruby presenting". It lists the characters and voices, makes the character (telling you first that the portrait costs one image), and writes her name into every prompt.