Character consistency

If you are telling a story that spans more than one shot, character consistency is the hardest problem and the most obvious when it fails. The fix is a deliberate workflow, not a secret prompt.

The reliable workflow

1. Build a character sheet first. Open Fawna Compose. Generate three to six images of your character from different angles: front, three-quarter left, profile, three-quarter right, back, close-up. Same clothing, same lighting, neutral expressions.
2. Save the best frames to Library. These become your permanent reference set. You will reuse them in every scene.
3. For each video scene, attach three refs. Pick views that match roughly what the scene needs. A scene with the character walking away benefits from a back or profile ref. A close-up needs at least one front ref.
4. Describe the character by stable attributes in every prompt. Age, build, hair, one or two visual signature items. Do not change these between scenes.

Stable description template

Write a two-line character description and reuse it word for word at the start of every scene prompt. Consistency in the text prompt is as important as consistency in the refs.

Maya is a woman in her late twenties, lean build, long dark braid,
freckles across her nose, wearing a mustard knit cardigan over a
white t-shirt and loose jeans.

Then the scene-specific prompt follows:

Maya is a woman in her late twenties, lean build, long dark braid,
freckles across her nose, wearing a mustard knit cardigan over a
white t-shirt and loose jeans. Cinematic medium shot: Maya sits on
a wooden bench in a sunlit greenhouse, trailing ivy behind her,
cradling a cup of tea. Soft side light through glass, slow dolly
forward. 50mm lens, warm palette, subtle film grain.

Picking the right model

Not every model is equally strong at consistency. The ranking, roughly, from most to least:

1. Fawna Compose (image). Best for building the initial character sheet.
2. Fawna Cinema (video). Best video-tier for multi-shot narratives with the same person.
3. Fawna Motion (video). Strong consistency, especially for stylized looks.
4. Fawna Muse (video). Excellent when driven by a 9-grid composite reference.
5. Fawna Film family. Uses first-frame image. Lock identity by generating your first frame in Compose, then feeding it in.

Wardrobe and signature items

The model latches onto distinctive clothing and accessories faster than it latches onto faces. If your character wears a specific hat, coat, or piece of jewelry in every shot, mention it by name in every prompt and show it in at least one ref. That signature acts as a consistency anchor even when the face drifts slightly.

Avoid changing wardrobe between shots in an episode. Each wardrobe change is a chance for the model to lose the identity. If you need a change, plan a transition shot where you see the character changing.

Two characters in one scene

Attach refs for both. In the prompt, refer to each character by descriptor, not by name. Names are unreliable because models do not know your "Maya" from someone else's:

• Good: "The woman with the dark braid in the mustard cardigan, and the bearded man in the yellow oilskin coat, walk along the dock."
• Weaker: "Maya and Jan walk along the dock."

Where this breaks

• Extreme angles or poses. Shots from directly above, acrobatic poses, or full silhouette often regress the identity. Favor eye-level and three-quarter framing for your hero shots.
• Heavy occlusion. Sunglasses, masks, hats pulled low. If the reference face is half-hidden the model guesses for the visible half.
• Long clips with many cuts. Within a single generation, identity stays put. Across separate generations spliced together, small drift is inevitable. Accept it, or re-roll the weakest shot.

Where to go next

• Keyframes for first and last frame control on tiers that support them.
• Scene Mode for planning multi-scene work from a script.