Firefly Composition vs Style Reference: Which Image Controls What?
Updated November 5, 2026
Direct answer: Use a composition reference to control the scene’s structure—including outline and depth—and a style reference to control its visual style and aesthetic. In the jar example, the real jar photo should define product identity, the plain arrangement sketch should define left/right positions, and the mood image should define color and lighting. Keep those jobs separate. Exact product identity is not guaranteed, and “product reference” describes a role here, not a universal Firefly button; available controls depend on the selected model and surface.
Assign each image one editorial job
The useful distinction is not simply “photo versus sketch.” It is identity, structure, or treatment. Before making a change, decide which question each image is allowed to answer.
| Input and role | What to borrow | What not to borrow | How to judge it |
|---|---|---|---|
| Real jar photo: factual identity source | The jar’s silhouette, proportions, and visible label details | Background lighting or composition, unless separately requested | Compare visible identity details directly with the source |
| Arrangement sketch: composition reference | Left/right positions, relative spacing, outline, and depth | Placeholder shapes as substitutes for the real jar or label | Review the layout independently of color and product details |
| Mood image: style reference | Color palette, lighting direction, and overall aesthetic | Product shape, label information, or object placement as facts | Review whether the requested visual treatment is present |
Apply the roles to one concrete scene
Imagine an ecommerce still life with a jar positioned on the left, an upright product card on the right, and a tabletop background. A plain line sketch establishes that arrangement. A mood image shows a muted blue palette with warm, directional illumination. The factual source photograph shows what the actual jar looks like.
Use the real jar photo as the factual identity source: preserve its silhouette, proportions, and visible label. Use the arrangement sketch only for structure: place the jar on the left and the card on the right. Use the mood image only for color and light: apply a muted blue background with a warm glow. Do not take product details or object positions from the mood image.
This wording tells the editor what each image contributes without asking every input to control everything. It also reduces ambiguity around the mood image: if it happens to contain a jar, that jar is not evidence of the real product’s identity.
Change one control at a time
First, confirm which reference controls are available in the selected model and surface. If both composition and style references can be used together, assign them separately. If they cannot, test them in stages rather than using one input as a substitute for the other.
- Test composition: Hold the jar source, style reference, and wording constant. Change only the composition input. Check whether the jar stays left, the card stays right, and the relative spacing, outline, and depth match the arrangement.
- Test style: Hold the jar source, composition input, and wording constant. Change only the style reference or style wording. Compare color, light, and overall aesthetic.
- Test the instruction: If the references look right but their assigned roles do not, revise one instruction at a time. For example, clarify that the sketch controls position while the mood image controls illumination.
Save each version with a short note identifying the single change. That makes it easier to identify which input caused a difference instead of guessing from the final image alone.
Judge layout and product label in separate passes
Layout pass: Temporarily ignore label detail and color fidelity. Check left/right placement, relative spacing, object scale, outline, and depth. These are composition concerns.
Identity and label pass: Return to the factual source photo and compare the jar’s silhouette, proportions, visible label, and label placement. A plausible-looking label is not necessarily the correct label. Because exact product identity is not guaranteed, final approval should depend on this direct visual comparison rather than on the general impression that the result “looks like the product.”
The style request and factual source answer different questions. “Muted blue palette and warm directional light” describes the desired treatment; the source photograph describes the product’s visible identity. Compare the Firefly result with those two requirements, not with another vendor’s output or an external ranking.
Use a short review routine
- Confirm which controls the selected model and surface expose.
- Label every input as identity, composition, or style before prompting.
- Make one role-specific change and save that version.
- Review composition first, using only the source photo and arrangement sketch as criteria.
- Review identity and label separately against the real jar photo.
- Adjust only the input responsible for the remaining problem.
Continue with the next task
- Product photography and editing resources
- Fashion and model workflow resources
- Choose a local repair method
- Control an editing revision
Sources and scope
Sources checked November 5, 2026. Rewarx is the publisher. New covers are MiniMax-generated concept illustrations, not outputs of the reviewed tool, measured specimens, or proof of product accuracy. Use the labeled body diagrams for numerical examples. Worked examples and diagrams are labeled editorial illustrations; no merchant conversion result or hands-on provider benchmark is claimed.