How Alien Road Uses Adobe Firefly in Client Campaigns

Three features matter most for client work. First, integration: Generative Fill and Generative Expand sit inside Photoshop, where designers already work, so generated elements land in layers that can be masked and edited. Second, Adobe’s position on training data and the IP indemnification it offers on eligible enterprise plans, which reassures legal teams. Third, Content Credentials are attached to Firefly outputs, recording that generative AI was used.

The Firefly app has also added access to some third-party models from other companies. Those models carry their own terms, so the commercial-safety arguments apply to Adobe’s own Firefly models, not automatically to every model in the interface.

Firefly is most useful in the middle of production rather than at the start or the end. Typical uses include:

• Extending a photo to new aspect ratios for social placements with Generative Expand.
• Removing distractions or adding background elements with Generative Fill.
• Producing seasonal variations of a hero image without a new shoot.
• Creating textures, patterns and vector elements in Illustrator.
• Drafting quick concept visuals for internal reviews.

It replaces neither the photographer for hero product images nor the art director’s decisions about what the campaign should look like.

Measurement looks at production efficiency and campaign results separately. On the production side, a typical engagement tracks time from brief to first approved draft, the number of format variants delivered per campaign, and the share of assets needing rework after client review. These show whether Firefly speeds production without lowering quality.

On the campaign side, generated or extended assets run alongside traditional ones where possible, and their click-through and conversion rates are compared on the same platform. If generated variants underperform, they are dropped, however quickly they were made.

Reporting to the client lists which published assets contain generated elements, the review record for each and the performance comparison. That transparency matters more to most clients than raw speed.

Firefly is not perfect, and knowing its weaknesses saves revisions:

• Complex scenes with many people or hands can still show errors.
• Exact product replication from a prompt is unreliable; use the real photo and edit around it.
• Generative credits are limited per plan, and heavy use can exhaust them before a deadline.
• Style consistency across a large set takes careful use of reference images.
• Outputs from third-party models in the app follow different terms.

Firefly is included in many Creative Cloud plans with a monthly allocation of generative credits, and extra credits can be bought. At the time of writing, Adobe adjusts these allowances from time to time, so teams should check their plan.

The bigger saving is usually in avoided reshoots and faster resizing. For example, delivering one campaign across many placements once required manual cropping and retouching for each ratio; Generative Expand reduces much of that work, while a designer still checks each output.

Clients are right to ask direct questions. Useful ones include: which assets used generative AI, which models produced them, whether Content Credentials are preserved, who reviewed product accuracy, and how disclosure is handled on each ad platform. A clear answer to each is a sign the process is under control.

Most designers learn the buttons in an afternoon. What takes longer is judgement: knowing when a generated fill will hold up at print size, when to stop iterating and when to go back to the photo. Short internal reviews help, where designers show one asset that worked and one that did not, with the prompts used.

Share a small prompt library organised by task, such as background extension, object removal and seasonal props, rather than by style. Add the brand’s do-not list to it: colours to avoid, props that clash with the product and scene types the client has rejected before. New team members then start from proven settings instead of guessing.