Fawna Compose

Compose is reference-driven image generation. Upload photos of a character, an object, or a style plate and the model preserves that identity while obeying a new prompt. This is the tool for character consistency and product-accurate imagery.

Tier label

Compose

Price

From 10 credits per image

Aspect ratios

1:1, 16:9, 9:16

Outputs per generation

1, 2, or 4

Character refs

Up to 9

Style ref

1 plate

Keyframes

Not applicable (still image)

Negative prompt

Not supported

Magic Prompt

Off

When to pick Compose

• You need the same character to appear across many images.
• You need a specific product to look exactly right in every shot.
• You want to transfer the look of a reference plate while composing a new scene.
• You are building a character sheet for later use in video tiers (Muse, Film).

Building a character sheet

The most valuable Compose workflow is creating a reusable character sheet.

1. Upload 3 to 5 photos of the character as character refs. Different angles, same wardrobe, even lighting.
2. Generate a front-on portrait first: "A photo of the character facing camera, shoulders square, neutral expression, soft studio lighting, plain grey backdrop."
3. Generate a three-quarter left view: "A photo of the character, three-quarter left view, same wardrobe and lighting."
4. Continue: profile, three-quarter right, back, close-up.
5. Save the best of each to Library. You now have a full character sheet.
6. Optional: compose these into a 9-grid for Muse.

Putting the character in scenes

Once you have a sheet, each new scene reuses the strongest 3 refs plus a consistent text description.

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. A photo of Maya sitting on the edge of a rooftop at
dusk, city lights below, leaning forward with her elbows on her
knees. Soft magic-hour backlight, 50mm lens, shallow depth of
field, warm natural palette, subtle film grain.

Style plate

Compose accepts one optional style plate in addition to character refs. Drop a film still, a painting, or a photograph with the look you want. The model transfers palette, lighting quality, and film texture from the plate while following the prompt's subject and scene.

Strengths

• Best-in-lineup character consistency.
• Accepts 9 refs, enough for full angular coverage.
• Style plate input for look transfer.
• Clean output that composites into video tiers (first-frame for Film, 9-grid for Muse).

Where it struggles

• No negative prompt. Steer with the positive prompt alone.
• No Magic Prompt expansion. Prompts must be fully hand-written.
• Compose can over-preserve. If you want a big visual change (different outfit, different age), write that change explicitly and reduce the ref count to 1-2.

Tips

• Consistent text description. Write your character description once and reuse verbatim in every prompt.
• 3 refs is the sweet spot. More than 5 can over-constrain the model. 1-2 gives more scene flexibility at the cost of identity.
• Mention the character by descriptor inside the scene: "A photo of Maya, the woman with the dark braid, standing at...". Repetition helps the model bind the refs to the subject.
• Save finalists to Library. Compose results become refs for the next generation.

Where to go next

• Character consistency for the end-to-end workflow.
• Fawna Muse for animating Compose output with 9-grid input.