exploreBe specific with simple language
Adobe recommends at least three words and simple, direct phrasing—subject plus descriptors plus keywords—not full command sentences.
Example
Macro photo of a yellow sunflower, dew drops, soft bokeh background.
Adobe Firefly
Generative AI for images, video, audio, and design built for commercially safe creative work.
ImageVideoDesign
How it works / How to use
Generative AI for images, video, audio, and design built for commercially safe creative work. Open Text to Image on firefly.adobe.com, describe the scene in plain language (at least three words; Adobe recommends starting with five or more), generate four options, then refine with Generative Fill, remove, or expand tools. Avoid command phrases like “create an image of”—stick to subject, descriptors, and keywords.
- Sign in to Adobe Firefly and open Text to Image (or Edit image for fill/remove workflows).
- Write a prompt with subject, descriptors, mood, lighting, and style—simple phrases work; full sentences are optional.
- Click Generate to receive four image options; upload a reference photo or sketch if you need tighter style or layout control.
- Refine with Generative Fill, Generative Remove, or Generative Expand, adjusting the prompt or brush selection one change at a time.
Text-to-image generation
Open Text to Image, enter a descriptive prompt, generate four variants, then iterate by changing keywords—not by adding “generate” or “create.”
What to provide
- Subject and setting in at least three words (five or more for stronger starts)
- Lighting, angle, mood, and materials you care about
- Aspect ratio or intended use (social post, product shot, poster)
Details that improve the result
- Adobe: be specific, descriptive, and original; infuse feeling, style, and lighting
- Start simple (e.g., “raining soccer balls in the desert”), then add detail
- Upload reference images when you need a specific style or composition
Example prompt
Product shot of a ceramic mug on a wooden table, morning sunlight from the left, shallow depth of field, warm tones, minimal background, no text.
If the first output is not good
- Swap one keyword at a time: lighting, lens, or mood
- Upload a reference image and regenerate for closer style match
- Use Generative Fill on one region instead of rewriting the whole scene
Common mistakes
- Opening with “create an image of” or “make me a” (Adobe says to avoid these)
- One-word prompts with no subject or setting
- Changing ten attributes before reviewing the first four options
Generative fill (add or replace objects)
Edit image → Fill → paint the target area → describe the fill → Generate. Adjust brush size and hardness for clean edges.
What to provide
- Base image (generated in Firefly or uploaded)
- Brush selection over the area to change
- Text description of what should appear in the painted region
Details that improve the result
- Use Select less / Select more to refine the mask after brushing
- Add a reference image under References to guide look and feel
- Generative fill keeps surrounding context while changing only the selection
Example prompt
Paint over empty table space. Prompt: A child playing with toy cars. Keep the rest of the kitchen scene unchanged.
If the first output is not good
- Shrink the selection if edges bleed into the background
- Add a reference photo if the fill style drifts
- Regenerate with a shorter prompt focused on the object only
Common mistakes
- Painting too large an area when only a small object should change
- Describing the entire scene instead of the fill region
- Skipping mask refinement before generating
Generative remove
Edit image → Generative remove → brush over the object → generate a clean plate that matches surrounding texture and lighting.
What to provide
- Image with an object or distraction to remove
- Brush selection over the item to erase
- Note what must stay untouched (faces, logos, layout)
Details that improve the result
- Smaller, precise selections reduce artifacts
- Follow with Generative Fill if the background needs reconstruction
- Iterate mask edges before committing to a final export
Example prompt
Remove the stray power line in the sky. Keep the building facade and color grading exactly as shown.
If the first output is not good
- Touch up remaining artifacts with a tighter remove pass
- Use Fill to reconstruct texture if remove leaves a blurry patch
- Boost resolution only after the composition is final
Common mistakes
- Selecting half the subject you meant to keep
- Removing large regions without checking lighting continuity
- Expecting perfect text or logo cleanup without manual touch-up
Generative expand
Choose Generative expand, pick the edge to extend, describe the added environment, and generate until the new border matches lighting and perspective.
What to provide
- Source image with cropped composition
- Direction to extend (canvas edge to expand)
- Description of what should appear in the new border area
Details that improve the result
- Match prompt lighting to the original frame
- Extend one edge at a time for predictable results
- Use references if the expanded region needs a specific style
Example prompt
Expand left and right. Prompt: continuation of the foggy forest path, same desaturated green palette, no new subjects.
If the first output is not good
- Expand again with a narrower prompt if the new region invents unrelated objects
- Crop back to the best segment if only one side succeeded
- Run Generative Fill on seams that do not blend
Common mistakes
- Expanding all sides at once with conflicting descriptions
- Asking for new focal subjects that fight the original crop
- Ignoring perspective mismatch in architecture shots
Style and composition references
Upload a reference under References or composition tools, then write a prompt that says what to keep vs what to change.
What to provide
- Reference image for style, layout, or subject
- Which aspect to match (color, composition, object design)
- Your text prompt describing the new content
Details that improve the result
- Reference images guide look and feel without copying every pixel
- Small prompt changes to lighting or materials produce large visual shifts
- Browse Adobe’s prompt examples to see how specificity changes output
Example prompt
Match the soft watercolor palette from the reference. Subject: a bicycle leaning on a café wall, same paper texture and loose brush edges.
If the first output is not good
- Lower style strength if the output copies unwanted elements
- Name the one reference attribute to change (background only, palette only)
- Generate four options and pick the closest before editing locally
Common mistakes
- Uploading a reference but not stating what should differ
- Mixing conflicting style references in one prompt
- Expecting exact trademarked logos from text alone
Commercial campaign and brand visuals
Write empathetic, audience-aware prompts (Adobe recommends words like gentle, playful, powerful, uplifting), generate variants, then refine in the image editor or Creative Cloud apps.
What to provide
- Brand constraints: colors, tone, forbidden elements
- Audience and channel (email hero, paid social, print)
- Required copy spelled exactly if text must appear in-image
Details that improve the result
- Tailor mood words to what the audience cares about
- Firefly is designed for commercially usable creative workflows—verify your plan’s usage terms
- Move finished assets into Photoshop or Illustrator for final brand polish
Example prompt
Uplifting, gentle hero image for a wellness app: sunrise yoga on a rooftop garden, soft golden light, diverse group, hopeful mood, no logos, no text.
If the first output is not good
- Ask for “realistic” or “naturalistic” if results feel overly synthetic
- Adjust Visual Intensity in the side panel when available
- Export and set Content Credentials as required by your workflow
Common mistakes
- Inventing metrics, claims, or product features in the prompt
- Skipping legal review for people, trademarks, or regulated copy
- Treating the first generation as final ad-ready art without retouching
Iterative prompt refinement
Keep the core subject stable; swap descriptors for mood, lighting, camera, or materials. Adobe recommends modifying prompts and settings until the image matches your vision.
What to provide
- Last prompt and what missed (lighting, subject, style)
- One dimension to change per iteration
- Saved favorite keywords that worked before
Details that improve the result
- Longer is not always better—Adobe recommends prompts of 175 words or less in its prompt guide
- Simple phrases often outperform conversational filler
- Use the image editor after generation for pixel-level control
Example prompt
Keep: ceramic mug product shot. Change only: switch studio lighting to overcast window light; add subtle condensation on the mug.
If the first output is not good
- Maintain a personal keyword list (e.g., “soft studio lighting,” “muted earth tones”)
- Compare four-up results side by side before picking a direction
- Document prompt + settings pairs that repeatedly work for your brand
Common mistakes
- Rewriting the entire prompt when one adjective was wrong
- Adding contradictory lighting terms in the same pass
- Stopping after one generation when four options are free to compare
How to prompt
Adobe Firefly responds to clear, artistic descriptions—not commands. Use at least three words with a subject, descriptors, and keywords. Be specific and descriptive; add feeling, style, and lighting. Skip phrases like “generate” or “create.” Start with five or more words, generate four options, then iterate.
Subject: vintage camera on a bookshelf. Descriptors: warm afternoon light, dust motes, shallow depth of field. Mood: cozy, nostalgic. Constraints: no text, no watermark.
Be descriptive about light and materials
Detailed descriptions of lighting, textures, and environment get you closer to the vision than vague mood words alone.
Example
Matte black headphones on concrete, harsh side lighting, subtle scratches, studio product photography.
Be original with style and feeling
Go beyond literal subjects—infuse feeling, art direction, and lighting choices that reflect your perspective.
Example
Playful, gentle illustration of a child reading under a blanket fort, crayon texture, warm lamp glow.
Be empathetic to the audience
Adobe notes audience-tailored mood words (love, gentle, powerful, uplifting) make images more engaging and relevant.
Example
Powerful, uplifting wide shot of a team launching a sailboat at dawn, confident mood, clean horizon.
Iterate, then refine in tools
Modify prompts and settings in Firefly, then use Generative Fill, remove, expand, or Creative Cloud apps for professional finishing.
Example
After Text to Image: Generative Fill to add a prop; Photoshop for final color grade and type.
Best output tips
Subject, descriptors, keywords
Adobe’s effective-prompt guidance: stick to simple, direct language with a subject, descriptors, and keywords. Avoid command phrases that add unwanted elements.
Start simple, then elaborate
Firefly’s Text to Image page recommends a short 4–5 word phrase first, then experimenting with more descriptive keywords as you iterate.
Four variants every generation
Each Generate click produces four unique options in seconds. Compare them before changing the prompt drastically.
Reference images
Upload a reference photo or sketch when you need a specific style or layout. Regenerate after attaching references under the References control in fill/edit flows.
Generative Fill masking
Use brush size and hardness controls, then Select less or Select more to clean the mask. Describe only what belongs inside the selection.
Empathetic mood words
Words like gentle, playful, powerful, and uplifting steer tone toward audience-appropriate imagery. Match mood language to channel and brand voice.
Reduce the AI look
Adobe’s prompt guide suggests style words like realistic or naturalistic, and adjusting Visual Intensity in the side panel when results feel over-processed.
Professional finishing
Firefly is adaptable for exploration, but Adobe recommends manual refinement in the image editor or apps like Photoshop and Illustrator for production work.
Specificity changes everything
Adobe’s prompt examples show small changes to lighting, materials, or environment produce dramatically different results—name the details you care about.
Generative expand with continuity
When extending canvases, describe continuation of existing light, palette, and perspective rather than introducing unrelated focal subjects.
- Use at least three words; start Text to Image prompts with five or more when possible.
- Skip “generate,” “create,” or “make me a”—describe the scene directly.
- Generate four options before rewriting the whole prompt.
- Change one dimension per iteration: lighting, lens, mood, or materials.
- Upload reference images when style or layout must match a known look.
- Paint precise masks for Generative Fill and Remove before generating.
- Keep prompts under ~175 words—longer is not always better per Adobe’s prompt guide.
- Move finished assets into Photoshop or Illustrator for brand-critical polish.
- Verify commercial usage and Content Credentials requirements for your plan.
Try this AI
Product Details
Pricing, features, limits and latest updates
Product Details
Pricing, features, limits and latest updates
Adobe Firefly
Generative AI for images, video, audio, and design built for commercially safe creative work.
Free / Paid · Free
Pricing Plans
Firefly Free
Free
Firefly Standard
$9.99/ Monthly
Firefly Pro
$19.99/ Monthly
Firefly Pro Plus
$49.99/ Monthly
Key Features
Image Generation
Generates images in Firefly using Adobe and partner models.
Video Generation
Premium generative credits are consumed for video generation on paid Firefly plans.
API
Adobe Firefly Services is a separate developer API and is not assumed to be included in consumer Firefly plans.
Image Editing
Standard tools such as Generative Fill are unlimited on paid Firefly plans; premium models use credits.
Limits
- Image prompt minimum: 3 words. Adobe recommends at least three descriptive words and avoiding phrases like "create an image of" (Firefly prompt help).
Ideas / Prompt experiences
Share a prompt that worked for you. Username and email are shown with your submission. External links are not allowed.
Example prompt
Text-to-image product still
Prompt
Ceramic coffee mug, soft morning window light, minimal wooden desk, product photography, shallow depth of field, clean background
Short explanation
Adobe Firefly recommends at least three descriptive words and subject + descriptors rather than 'create an image of'.
Example prompt
Generative Fill edit
Prompt
[select area in image] Replace the background with a soft gradient studio backdrop. Keep the product shape, label, and lighting unchanged.
Short explanation
Generative Fill works on selections; name what must stay unchanged to avoid unwanted edits.
Share your experience
Required fields are marked. Variation, result, and explanation are optional.
