Adobe Firefly is Adobe’s family of generative AI models for images, vector graphics, video and text effects, built into Photoshop, Illustrator, Express and the Firefly web app. Adobe says its models are trained on licensed content such as Adobe Stock and public domain material, which makes Firefly a common choice for commercial work. The usual mistakes are not about the model; they are about how teams brief, check and file what it produces.
What makes Firefly different from other image generators?
Firefly’s main selling point is its training data approach and the commercial terms around it. Adobe offers intellectual property indemnification for qualifying enterprise customers using certain Firefly features. Outputs carry Content Credentials, the C2PA metadata that records that AI was used.
Its second strength is integration. Generative Fill and Generative Expand in Photoshop, Generative Recolor in Illustrator and text-to-template features in Adobe Express sit inside tools designers already use. The Firefly app has also added access to some third-party models; check which model you are using, because the commercial terms can differ.
Generative features use generative credits, which come with Creative Cloud plans in varying amounts. Heavy users can run out mid-project if credit use is not planned.
Mistake 1: assuming commercially safe means risk-free
Licensed training data reduces one kind of risk. It does not stop a prompt from producing something that resembles a trademark, a known character or a real person. And indemnification applies to specific plans and features, under conditions.
Agencies avoid this by keeping the same prompt rules they use for any generator: no living artists’ names, no brand names, no real people, no copyrighted characters. They also check which plan the client is on before promising indemnification, and confirm that the feature used is covered.
Mistake 2: generating the product instead of placing it
Firefly can produce a convincing perfume bottle or sneaker. It will not be your perfume bottle or sneaker. Labels, proportions and details will be invented.
The better workflow is to shoot or render the real product, cut it out, and use Generative Fill or Generative Expand only for the scene around it. The product layer stays untouched. This is standard practice for e-commerce and retail campaigns.
Mistake 3: ignoring resolution and print needs
Generated areas may have a lower effective resolution than the surrounding photo, and fine detail can look soft when printed large. Teams discover this at proof stage for billboards or packaging, when deadlines are tight.
- Check the output size and detail at 100% zoom before approval.
- Generate fills in smaller sections on high-resolution files.
- Use upscaling carefully and inspect edges and textures afterwards.
- Proof any print use at actual size, or a large section of it.
- Keep the original, unfilled file for rework.
Mistake 4: inconsistent style across a campaign
Each generation is a little different. Across twenty social assets, lighting, colour and texture drift until the campaign looks assembled from different sources.
Firefly offers style and structure reference options, where you upload an image to guide the look or composition. Agencies build a small reference set at the start of a campaign and use it for every generation. Custom models, available on some enterprise plans, go further by training on a brand’s own approved assets.
Mistake 5: stripping or ignoring Content Credentials
Content Credentials record AI use in the file’s metadata. Exporting through some tools or uploading to some platforms removes them, which means the disclosure trail disappears.
Keep credentials on the master files and add visible disclosure where realistic AI imagery could mislead, following platform rules on Meta, TikTok and YouTube. For EU audiences, the AI Act’s deepfake transparency rules add another reason to label scenes that could pass as real.
Mistake 6: no record of what was generated
Months later, a client asks whether an image in an ad was AI-generated and with which tool. Without records, nobody knows.
- Note the tool, model and feature used for each asset.
- Save the prompt and reference images.
- Record the designer and approver.
- Keep the layered file showing which areas were generated.
- Store the record with the final asset in the DAM or project folder.
How Alien Road applies Adobe Firefly
Our graphic design team uses Firefly mainly for scene extension, background work, concept boards and variations for social formats. Discovery for each campaign covers the client’s Creative Cloud plan, credit allowance, any indemnification terms and the brand’s rules on AI imagery.
Setup includes a reference image set for style consistency, prompt rules and a file naming convention that marks generated layers. Real products, logos and people are always photographed or supplied, never generated.
Every asset passes a brand check and a rights and disclosure check before delivery. Reporting covers production hours, credit use and any assets that went back for rework, so the next campaign’s plan is based on real numbers.
Mistake 7: treating Firefly as a replacement for design judgement
Firefly speeds up execution, not concept. A weak idea rendered quickly is still a weak idea. The strongest results come when a designer decides the composition, hierarchy and message first, then uses generation to fill specific gaps.
Time saved on masking, extending and retouching should go back into concept and typography. Teams that skip that step end up with more assets that all look alike.
How should you plan generative credits?
Credits run out at the worst moment, usually during final client revisions. Estimate generations per asset during the first project and multiply by the number of assets planned. Hold some credits back for revision rounds, and check usage in the Adobe account page weekly during busy periods.
Large teams should check whether credits are pooled or assigned per user on their plan. One designer exploring heavily can drain an allowance that others were counting on.
Where to start this week
Check your Creative Cloud plan to see your generative credit allowance and whether any indemnification applies. Then take one recent product photo and practise the safe workflow: keep the product on its own layer, extend the background with Generative Expand, and check the result at full size. Save the prompt, the layered file and a short note on what was generated, and use that note as the template for every future asset.
Keep reading: Graphic Design · AI Marketing