Adobe Firefly stopped being a Midjourney alternative a while back and turned into something closer to a production pipeline. As of August 2026, the app runs on Firefly Image Model 4 and the premium 4 Ultra tier, with pricing trackers already showing credit costs for a next-tier model rolling into the same interface. Firefly Image 3, the version most tutorials online still reference, is being pulled from Photoshop’s model picker and fully retired from the service this month. If you followed an older guide, half of what it told you no longer applies.
This tutorial walks through the current Adobe Firefly workflow end to end: account setup, the credit system that replaced flat subscriptions, prompt structure, Generative Fill, Generative Expand, Style and Structure Reference, Firefly Boards, Photoshop integration, video generation, and the developer API. By the end you’ll have generated, refined, and exported a complete marketing asset using nothing but the current Firefly toolset, and you’ll know exactly where the credit meter runs out.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Is Adobe Firefly in 2026?
Adobe Firefly is Adobe’s generative AI suite for images, video, audio, and vector graphics, built into the standalone Firefly web app and threaded through Photoshop, Premiere Pro, Illustrator, and Adobe Express. The pitch that has carried it through 2026 hasn’t changed: Firefly models are trained on Adobe Stock, licensed content, and public domain material, which is why Adobe markets it as the “commercially safe” option next to Midjourney, Flux, or Nano Banana Pro.
What has changed is the model lineup and how you pay for it. Firefly Image Model 4 and 4 Ultra now handle generation up to 2K resolution with materially better control over camera angle, format, and style adherence than Firefly 3 offered. Firefly Image 3 support ended in Photoshop’s 27.6 release on April 28, 2026, and Adobe is removing it from the service entirely across all plans this August. If you’re picking a model in the dropdown today, Image 4 (or the Ultra variant) is the baseline, not the upgrade.
Firefly also isn’t a single-model tool anymore. Higher Firefly plans now bundle unlimited generations on select partner models for the first year, including Nano Banana, Flux.2 Pro, and Kling 2.5 Turbo for video, alongside Adobe’s own Image 4 and 4 Ultra. That’s a meaningful shift: Firefly is positioning itself as an orchestration layer over the entire generative image and video landscape, not just Adobe’s in-house model.
The practical effect of that shift is that the “which AI image generator should I use” question increasingly gets answered inside a single subscription rather than across five separate tabs. A designer on a Firefly Pro Plus plan can pull a Nano Banana generation for one asset and an Image 4 Ultra render for another, all billed from the same credit pool, without ever leaving the Firefly interface or juggling separate logins. That consolidation is a bigger deal for production teams than any single model upgrade, since it collapses vendor management into one dashboard.
It also changes how you should think about “which model is best.” The old framing — pick one tool and live with its tradeoffs — doesn’t really hold anymore inside Firefly. The model picker is the tradeoff decision now, made per generation rather than per subscription. That’s the mental model to carry into the rest of this tutorial: you’re not just learning Firefly’s own Image 4 engine, you’re learning an interface that routes between several engines depending on what a given shot needs.
Prerequisites: What You Need Before You Start
Firefly runs entirely in the browser for the core workflow, so hardware requirements are light. Here’s what to have ready before Step 1:
- Adobe account — free to create at adobe.com; no credit card required for the free tier
- Browser — Chrome 120+, Edge 120+, Safari 17+, or Firefox 120+ (Chromium-based browsers render the canvas tools most reliably)
- Internet connection — Firefly generation happens server-side; a stable connection matters more than local GPU power
- Photoshop 27.5 or later — only needed for Step 10 (Boards integration and in-app Generative Fill); Photoshop 27.6+ is required since it drops Firefly Image 3 from the model picker
- Payment method — required only if you move past the 25 free monthly credits; Adobe accepts major cards and PayPal in most regions
- For the API section — a free Adobe Developer Console account and basic familiarity with REST requests (curl or Python)
Budget roughly 90 minutes for the full walkthrough if you’re doing every step, including the API portion. If you only need the web app basics, you can finish in about 30 minutes.
Adobe Firefly Pricing and Credits in 2026
Firefly moved to a generative-credit model rather than unlimited-use subscriptions, and the credit cost differs by which model you call. Here’s the current tier breakdown for U.S. pricing:
| Plan | Monthly Price | Generative Credits | Best For |
|---|---|---|---|
| Free | $0 | 25 credits/month | Testing prompts, watermarked output |
| Firefly Standard | $9.99 | 2,000 credits/month | Solo creators, light regular use |
| Firefly Pro | $19.99 | 4,000 credits/month | Freelancers, heavy solo workflows |
| Firefly Pro Plus | $49.99 ($34.97 first year, promo ends Aug 26, 2026) | 10,000 credits/month | Small teams, agencies |
| Firefly Premium | $199.99 ($139.91 first year, promo ends Aug 26, 2026) | 50,000 credits/month + unlimited video model access (first year) | Studios and high-volume production |
Credits don’t roll over month to month, which is the single most common complaint in Firefly reviews — unused credits simply expire at the billing cycle reset. Adobe Express Premium runs a parallel track: it bundles 4,000 generative credits per month plus unlimited access to standard features like Generative Fill in Photoshop, which is worth checking if you’re already on a Creative Cloud All Apps plan at $59.99/month, since that tier includes Firefly Pro-level credits by default.
The credit cost per generation depends on which model you call, not on resolution alone:
| Operation | Model | Credit Cost |
|---|---|---|
| Standard image generation | Firefly Image Model 4 | 10 credits per image |
| Premium image generation | Firefly Image 4 Ultra | 20 credits per image |
| Text-to-video (5-second clip) | Firefly Video Model | ~10 credits per clip |
| Video translation/dubbing | Firefly Video Model | ~300 credits per minute (approx., based on add-on pack ratios) |
Run the math before you commit to a plan: on Firefly Standard’s 2,000 credits, that’s 200 standard-quality images or 100 Ultra-quality images per month. Teams that burn through credits fast should look at the add-on packs (roughly $29.99 for 7,000 extra credits, $49.99 for 10,000) rather than jumping straight to Premium.
Step 1: Create Your Adobe Account and Pick a Plan
Go to the Adobe Firefly product page and sign up with an email address or an existing Adobe ID. The free plan activates automatically with 25 generative credits — enough to test the interface but not enough to iterate seriously on a real project.
For this tutorial, Firefly Standard ($9.99/month) is enough to complete every step including the final project without hitting a credit wall. If you already pay for Creative Cloud All Apps, skip a standalone Firefly subscription entirely — check your existing plan’s generative credit allowance first under Account > Plan details.
Step 2: Open the Firefly Web App and Learn the Layout
Navigate to the Firefly web app dashboard. The home screen splits into modules: Text to Image, Generative Fill, Generative Expand, Text Effects, Generate Video, and Boards (your saved collections). Each module opens its own workspace with a prompt bar, model selector, and a settings panel on the right for aspect ratio, style, and reference images.
Spend two minutes clicking through the model selector dropdown before generating anything. You’ll see Firefly Image 4, Image 4 Ultra, and — depending on your plan — partner models like Nano Banana or Flux.2 Pro grouped separately. Each one has a different credit cost, shown directly next to its name.
Step 3: Generate Your First Image with Text to Image
Click Text to Image, select Firefly Image Model 4 from the model dropdown, and type a prompt into the bar. Set your aspect ratio (square, landscape 16:9 or 3:2, portrait 9:16 or 2:3 are the standard presets) and click Generate. Firefly returns four variations per generation by default, each consuming credits from your pool.
Try this starter prompt to see the model’s default behavior:
A minimalist product photo of a ceramic coffee mug on a
concrete surface, soft studio lighting from the top left,
shallow depth of field, neutral gray background, commercial
product photography styleNotice the structure: subject, setting, lighting, and style, each as a distinct clause. That’s not accidental — it’s the prompt formula the rest of this tutorial builds on.
Step 4: Master the Four-Part Prompt Structure
Firefly’s model responds more predictably to structured prompts than to freeform description. The pattern that consistently produces usable first-pass results breaks into four parts:
- Subject — what’s actually in frame, described specifically (not “a person,” but “a woman in her 30s wearing a navy blazer”)
- Setting/composition — where it is and how it’s framed (studio backdrop, outdoor street, close-up vs. wide shot)
- Lighting — direction and quality (golden hour side light, harsh overhead fluorescent, soft diffused studio light)
- Style — the rendering approach (photorealistic, editorial photography, 3D render, watercolor illustration)
Skipping the style clause is the most common reason first-time users get inconsistent results — Firefly defaults to a fairly generic photographic look, and it will drift between generations without an explicit style anchor. Keep every prompt under roughly 75 words; Firefly’s model tends to ignore or blend late clauses in overly long prompts rather than weighting them evenly.
Step 5: Apply Style Presets and Style Reference
Open the Styles panel next to the prompt bar. Adobe ships preset style tags (photorealistic, 3D, comic, watercolor, cinematic) that you can stack with sliders controlling style strength. For one-off images, presets are faster than writing out full style descriptions in the prompt.
For brand-consistent work — the scenario most tutorials skip — use Style Reference instead. Upload an existing image (your own artwork, a past campaign asset, a brand mood board image) and Firefly extracts its color palette, texture, and general visual treatment, applying that style to new prompts. This is the workflow to use if you need twenty product shots that all look like they came from the same photo shoot.
Steps to use Style Reference:
- Click the reference image icon in the settings panel
- Upload a style source image (JPEG or PNG, under 40MB)
- Set style strength — start around 60% so the prompt still has room to define new subject matter
- Write your subject prompt and generate
Step 6: Use Structure Reference to Lock Composition
Structure Reference works alongside Style Reference but controls a different variable: layout, camera angle, and object placement rather than visual style. Upload a reference image, enable Structure Reference, and Firefly will generate new content that follows the same composition — same camera angle, same relative positioning of elements — while applying whatever new subject or style you describe in the prompt.
This matters for production work where you need consistent framing across a series: think a product catalog where every item needs to sit in the same three-quarter angle, or a UI mockup set that needs identical layout across different color themes. Generate one composition you like, then reuse it as the structure reference for every subsequent variation instead of re-describing framing in every prompt.
Step 7: Remove or Replace Objects with Generative Fill
Generative Fill is Firefly’s inpainting tool, available both in the standalone web app and inside Photoshop. Here’s the web app version:
- Open Generative Fill and upload the source image (or pull one from a Board)
- Use the selection brush to mask the area you want changed — an object to remove, a background to swap, empty space to fill
- Type a prompt describing what should appear in the masked area (leave it blank to let Firefly fill contextually, which works well for simple object removal)
- Click Generate — Firefly returns multiple fill variations, each isolated to the masked region
- Accept a variation or regenerate with an adjusted prompt
Each Generative Fill pass draws from the same credit pool as standard generation. For clean removals (stray objects, background clutter), leave the prompt empty — an over-specified prompt on a removal task tends to produce a fill that looks stylistically disconnected from the rest of the photo.
Step 8: Extend Canvas with Generative Expand
Generative Expand (outpainting) fills in new canvas area around an existing image — useful for reformatting a square product shot into a widescreen banner without recomposing from scratch. Upload your source image, drag the canvas boundary outward in the direction you need, and Firefly generates content that extends naturally from the original edges.
The catch: Generative Expand works best on images with simple, low-detail backgrounds. Expanding a busy scene with lots of distinct objects near the edge tends to produce visible seams or repeated elements. If the expand result looks off, try expanding in smaller increments (25-30% of the original canvas at a time) rather than doubling the canvas in one pass.
Step 9: Organize Output with Firefly Boards
Firefly Boards function as cloud-based mood boards for your generations — a place to pin, group, and compare outputs across sessions instead of losing good variations in your generation history. Save any result to a Board directly from the results panel.
Boards integrate with Photoshop 27.5 and later: you can send work-in-progress directly from Photoshop into a Board, and Boards now surface inside Photoshop’s Variations panel. For a real project, the practical workflow is to generate broadly (15-20 variations across a few prompt directions), curate the strongest 5-8 into a Board, then move those into Photoshop for final refinement.
Step 10: Integrate Firefly with Photoshop
Inside Photoshop 27.6 or later, Generative Fill and Text to Image live directly in the Contextual Task Bar. Make a selection, and Photoshop surfaces a Generative Fill prompt field right where you’re working — no need to jump between the web app and your editing session.
Remember that Photoshop 27.6 dropped Firefly Image 3 from the model picker as of April 28, 2026. If you’re working from an older Photoshop install or a saved workflow preset that still references Image 3, update it now — those calls will fail once the model is fully retired from the service this August. Check Help > About Photoshop to confirm you’re on 27.6 or newer before relying on in-app generation for production work.
Step 11: Generate Video with the Firefly Video Model
Click Generate Video from the Firefly dashboard. Text-to-video clips run 5 seconds by default and cost approximately 10 credits each on the native Firefly Video Model. Higher Firefly plans also unlock partner video models, including Kling 2.5 Turbo, Runway, and Luma, for unlimited generations during the first year on qualifying tiers.
Video prompts follow the same four-part structure as image prompts, with one addition: describe camera movement explicitly (slow push-in, static shot, pan left) since Firefly’s video model treats motion as a distinct parameter from composition. For product marketing use cases, static or slow-push shots render more reliably than complex camera movements — save the ambitious camera work for models specifically tuned for it.
Bonus: Text Effects and Vector Generation
Two Firefly modules worth knowing about even though they don’t fit neatly into the image/video split: Text Effects and Generative Recolor. Text Effects applies AI-generated textures, patterns, or materials directly to typography — type a headline, describe a texture (“molten gold,” “cracked ice,” “woven denim”), and Firefly renders the letterforms in that material while keeping them legible. It’s a fast way to produce a styled title card or social header without opening Illustrator.
Generative Recolor works on vector artwork: upload an SVG or open one in Illustrator, describe a new color scheme in plain language (“autumn palette,” “high-contrast neon”), and Firefly remaps the existing shapes to that palette without altering the underlying vector paths. Both tools draw from the same generative credit pool as image generation, though at a lower cost per use since they’re modifying existing assets rather than generating from scratch.
Step 12: Automate Generation with the Firefly API
For batch workflows — generating dozens of product variants, or plugging Firefly into an internal content pipeline — use the Firefly Services API rather than the web app. Register a project at the Adobe Developer Console to get your client credentials.
First, authenticate and get an access token:
curl -X POST https://ims-na1.adobelogin.com/ims/token/v3 \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "grant_type=client_credentials" \
-d "client_id=YOUR_CLIENT_ID" \
-d "client_secret=YOUR_CLIENT_SECRET" \
-d "scope=openid,AdobeID,firefly_api,ff_apis"Then call the image generation endpoint with your prompt and model ID:
curl -X POST https://firefly-api.adobe.io/v3/images/generate \
-H "Authorization: Bearer YOUR_ACCESS_TOKEN" \
-H "x-api-key: YOUR_CLIENT_ID" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A ceramic coffee mug on a concrete surface, soft studio lighting, commercial product photography",
"model": "image-model-4",
"size": { "width": 2048, "height": 2048 },
"numVariations": 4
}'A Python wrapper for batch runs looks like this:
import requests
def generate_firefly_image(prompt, access_token, client_id, model="image-model-4"):
url = "https://firefly-api.adobe.io/v3/images/generate"
headers = {
"Authorization": f"Bearer {access_token}",
"x-api-key": client_id,
"Content-Type": "application/json"
}
payload = {
"prompt": prompt,
"model": model,
"size": {"width": 2048, "height": 2048},
"numVariations": 4
}
response = requests.post(url, headers=headers, json=payload)
response.raise_for_status()
return response.json()["outputs"]
prompts = [
"A ceramic coffee mug on concrete, studio lighting, product photography",
"A leather laptop bag on a wooden desk, natural window light, editorial style",
]
for p in prompts:
result = generate_firefly_image(p, ACCESS_TOKEN, CLIENT_ID)
print(f"Generated {len(result)} variations for: {p}")If your pipeline runs on Node instead of Python, the equivalent request looks like this:
const generateFireflyImage = async (prompt, accessToken, clientId) => {
const response = await fetch("https://firefly-api.adobe.io/v3/images/generate", {
method: "POST",
headers: {
"Authorization": `Bearer ${accessToken}`,
"x-api-key": clientId,
"Content-Type": "application/json",
},
body: JSON.stringify({
prompt,
model: "image-model-4",
size: { width: 2048, height: 2048 },
numVariations: 4,
}),
});
if (!response.ok) {
throw new Error(`Firefly API error: ${response.status}`);
}
const data = await response.json();
return data.outputs;
};
generateFireflyImage(
"A leather laptop bag on a wooden desk, natural window light, editorial style",
process.env.FIREFLY_ACCESS_TOKEN,
process.env.FIREFLY_CLIENT_ID
).then((outputs) => console.log(`Got ${outputs.length} variations`));API calls draw from the same generative credit pool as the web app — a standard call against Image Model 4 costs 10 credits, and the Ultra model costs 20, regardless of whether you triggered it through the UI or a script. Consult the developer portal for current endpoint versions and authentication scopes before deploying this in production, since Adobe updates API paths more frequently than the web UI.
Complete Working Project: A Product Marketing Campaign in Firefly
Here’s how the full workflow comes together for a real deliverable: a five-image product marketing set plus one 5-second video teaser, built entirely in Firefly.
- Generate a base composition. Use Text to Image with Firefly Image 4 to create your hero product shot. Generate 4 variations, pick the strongest composition.
- Lock the structure. Save the chosen image as a Structure Reference so every subsequent shot shares the same camera angle and framing.
- Generate variants. Run 4-5 prompts against the Structure Reference, swapping product color, background context, or lighting mood for each. This keeps the whole set visually cohesive.
- Clean up with Generative Fill. Remove any unwanted artifacts, stray reflections, or background elements from each variant.
- Reformat with Generative Expand. Extend your best square shot into a 16:9 banner and a 9:16 story format without recomposing.
- Curate in a Board. Pin your final five images to a Board for client review and version tracking.
- Add a video teaser. Feed your hero image into Generate Video with a slow push-in camera prompt for a 5-second social teaser.
- Export and verify Content Credentials. Confirm the Content Credentials metadata is intact on export so the campaign’s AI provenance is documented.
Total credit cost for this project on Firefly Image 4: roughly 90-110 credits (9-11 generations at 10 credits each), well within a single month’s Standard plan allowance, with room left for revisions.
Common Pitfalls When Using Adobe Firefly
- Skipping the style clause in prompts. Without an explicit style anchor, output drifts between generations, making it hard to build a cohesive set.
- Burning credits on Image 4 Ultra by default. Ultra costs double standard generation. Reserve it for final hero assets, not exploratory drafts.
- Assuming credits roll over. They don’t. Unused credits expire at each billing cycle — plan monthly usage rather than banking for a big project later.
- Over-specifying Generative Fill prompts. For simple object removal, an empty or minimal prompt usually blends better than a detailed description.
- Expanding canvas in one large jump. Generative Expand handles small extensions cleanly; doubling canvas size in one pass often produces visible seams.
- Relying on old Firefly Image 3 workflows. Saved presets or tutorials referencing Image 3 will break once the model is fully retired this August — rebuild them against Image 4.
- Ignoring Content Credentials on export. Some export settings strip provenance metadata, which matters for organizations with disclosure policies on AI-generated content.
Troubleshooting Adobe Firefly
- Generation fails with “out of credits.” Check Account > Plan details for your remaining balance; either wait for the monthly reset or purchase an add-on credit pack.
- Style Reference isn’t applying strongly enough. Increase the style strength slider past 60%, or verify your reference image has a clearly identifiable visual style (busy, low-contrast references transfer poorly).
- Generative Fill results look mismatched to the rest of the photo. Shrink your selection mask closer to the actual edit area — overly large masks give the model too much freedom and it loses the original lighting context.
- Structure Reference isn’t preserving composition. Confirm the toggle is actually enabled in the settings panel; it’s easy to upload a reference image without activating the reference mode itself.
- Photoshop can’t find Image 3 in the model picker. That’s expected — Image 3 was removed starting with Photoshop 27.6. Update saved actions or presets to reference Image 4.
- API calls return authentication errors. Access tokens from the IMS endpoint expire; refresh your token before each batch run rather than caching it long-term.
- Video generation queue is slow during peak hours. Video rendering is more compute-intensive than image generation and can queue during high-traffic periods; retry off-peak or check Adobe’s status page.
- Boards aren’t syncing between the web app and Photoshop. Confirm you’re signed into the same Adobe ID in both, and that Photoshop is on 27.5 or later — Boards integration doesn’t exist in older versions.
- Free tier images come out watermarked. That’s by design on the free plan; watermarking is removed starting at the Standard paid tier.
- Generative Expand repeats elements near the new edge. Reduce the expansion percentage per pass and run it iteratively instead of expanding the full target size at once.
Advanced Tips for Production Workflows
Once the basics are solid, a few habits separate one-off hobby use from a repeatable production pipeline. First, standardize your prompt template across a team — a shared four-part structure with placeholders for subject, setting, lighting, and style keeps output consistent even when different people are generating assets for the same campaign.
Second, treat the API as your default for anything beyond five or six images. The web app is fine for exploration, but batch scripting through the Firefly Services API removes the manual click-generate-download loop entirely and makes it trivial to regenerate an entire set when a client requests a single tweak to the prompt template.
Third, if you’re on a team plan, pool credits deliberately. Enterprise and team tiers allocate generative credits per user starting around 250 credits/user/month, and it’s common for some team members to burn through their allocation while others barely touch theirs. Centralized credit pooling avoids that imbalance if your admin console supports it.
Fourth, use Structure Reference more aggressively than most tutorials suggest. It’s the single fastest way to turn one good composition into a consistent set of ten, and it cuts prompt iteration time dramatically compared to re-describing framing from scratch on every generation.
Fifth, build a personal or team prompt library instead of starting from a blank bar every session. Save the four-part prompts that produced strong results, tag them by use case (product shot, hero banner, social teaser), and reuse the structure with swapped-out subject and setting clauses. This is the difference between a five-minute generation session and a forty-five-minute one when a deadline is close.
Adobe Firefly for Teams and Enterprises
Enterprise and team plans change a few things beyond just the credit total. Generative credits can be pooled and allocated per user rather than assigned in a flat monthly amount per seat, with allocations starting around 250 credits per user per month on team tiers and scaling from there based on contract size. That pooling matters in practice: on individual plans, unused credits simply expire, but centralized pools let an admin console shift unused allocation from light users to heavy ones within the same billing cycle.
Enterprise deployments also get deeper access to Style and Structure Reference for brand governance — locking a house style across a whole creative team so that campaign assets stay visually consistent regardless of who’s generating them. Combined with Boards for documenting creative decision paths, this gives design leads a way to review AI-assisted work the same way they’d review a Photoshop layer history: with a visible trail of what was tried and what got approved.
On the compliance side, enterprise customers get the fullest version of Adobe’s indemnification coverage and the most explicit documentation on how Content Credentials metadata should be handled internally. If you’re evaluating Firefly for a regulated industry — finance, healthcare, government contracting — that’s the plan tier to request a sales conversation about rather than trying to piece together coverage details from the standard consumer terms of service.
Adobe Firefly vs. Midjourney vs. GPT Image 2 vs. Nano Banana Pro
Firefly isn’t competing to be the single best image model on raw quality — it’s competing on legal safety and ecosystem depth. Here’s how it stacks up against the other tools people are actively comparing it to in mid-2026:
| Tool | Starting Price | Free Tier | Standout Strength |
|---|---|---|---|
| Adobe Firefly | $9.99/mo | 25 credits/mo, watermarked | Commercial safety, Photoshop/Express integration, Content Credentials |
| Midjourney | ~$10/mo (200 images) | None | Artistic quality and stylized aesthetics |
| GPT Image 2 (via ChatGPT/API) | ~$0.04/image via API | Limited via ChatGPT | Prompt accuracy and ease of use |
| Google Nano Banana 2/Pro (Gemini) | Free | Yes, no card required | Best genuinely free option for everyday use |
If your priority is raw artistic output, Midjourney still tends to win head-to-head aesthetic comparisons. If you need zero-cost everyday generation, Nano Banana via Gemini is hard to beat. Firefly’s case is narrower but sharper: if your images need to survive legal review, plug directly into a Photoshop pipeline, or carry documented AI provenance, it’s the tool built for that job specifically — and paying the credit premium buys you that safety net.
Worth a specific mention: if in-image text is your main requirement — packaging mockups, posters, UI comps with real copy — none of the four tools above are the strongest option. Ideogram still leads that particular category, with accuracy on rendered text well ahead of Midjourney and roughly comparable image-generation pricing. Firefly’s Text Effects module covers stylized headline treatments well, but for dense body copy or small-print legibility inside a generated scene, it’s not the tool built for that job. Pick based on what a given asset actually needs rather than defaulting to one generator for every task.
Commercial Use, Copyright, and Content Credentials
Firefly’s core commercial pitch rests on training data provenance: Adobe trains its models on licensed Adobe Stock content and public domain material rather than scraping the open web, which is the basis for its “commercially safe” positioning against models trained on less transparent datasets. Adobe extends indemnification protections to paid users under its Creative Cloud and Firefly terms of service — read the current terms before relying on this for high-stakes enterprise use, since the specifics of coverage can change between plan tiers.
Every Firefly-generated image carries Content Credentials, the provenance metadata standard developed through the Content Authenticity Initiative, which Adobe co-founded. This metadata tags an image as AI-generated and records which model produced it — important for newsrooms, regulated industries, and any organization with a disclosure policy on synthetic media. Free-tier outputs additionally carry a visible watermark; that watermark drops away starting at the Standard paid tier, though the underlying Content Credentials metadata persists across all tiers.
Adobe’s official Firefly help documentation covers the current interface in detail, including the getting-started overview and a full video playlist walking through each module, both useful references if you get stuck on a specific tool beyond what’s covered here. For credit balances, plan changes, and billing questions, Adobe’s Firefly FAQ page and the Creative Cloud plans page stay current with pricing changes faster than most third-party trackers.
Frequently Asked Questions
Is Adobe Firefly free to use?
Yes, with limits. The free plan includes 25 generative credits per month, enough for roughly 2-3 standard image generations, and outputs carry a visible watermark. Paid plans start at $9.99/month for 2,000 credits.
What happened to Firefly Image 3?
Adobe began removing Firefly Image 3 from Photoshop’s model picker starting with version 27.6 on April 28, 2026, and is fully retiring it from the Firefly service across all apps and plans in August 2026. Firefly Image 4 and 4 Ultra are the current default models.
Can I use Firefly-generated images commercially?
Yes. Firefly is designed for commercial use, and Adobe extends indemnification protections to paid subscribers under its terms of service. Review the current Firefly and Creative Cloud terms directly before relying on this for large-scale commercial or enterprise deployment.
How many credits does one image generation cost?
Standard generation with Firefly Image 4 costs 10 credits per image. The premium Image 4 Ultra model costs 20 credits per image. Video generation runs roughly 10 credits per 5-second clip on the native Firefly Video Model.
Does Firefly work inside Photoshop?
Yes. Photoshop 27.5+ includes Generative Fill and Text to Image directly in the Contextual Task Bar, and 27.5+ also supports sending work to Firefly Boards. You need Photoshop 27.6 or later to access the current Image 4 model, since 27.6 dropped Firefly Image 3 from the picker.
What’s the difference between Style Reference and Structure Reference?
Style Reference extracts color palette, texture, and visual treatment from an uploaded image and applies it to new generations. Structure Reference preserves composition, camera angle, and object placement instead, letting you keep layout consistent while changing subject matter or style.
Can developers access Firefly through an API?
Yes, through Firefly Services. Register a project at the Adobe Developer Console, authenticate via OAuth to get an access token, then call the image generation endpoint. API calls draw from the same generative credit pool as web app usage.
How does Firefly compare to Midjourney or Nano Banana Pro?
Firefly trades some raw aesthetic edge for legal safety and ecosystem integration. Midjourney generally wins on stylized artistic output; Nano Banana via Gemini is the strongest free option; Firefly’s advantage is Content Credentials provenance, Photoshop/Express integration, and Adobe’s commercial indemnification terms.
Related Coverage
- How to Use Nano Banana Pro: 11 Steps, 80 Min [2026]
- How to Use Midjourney: 10 Steps, 75 Min [2026]
- How to Use Seedream 5.0 Pro: 12 Steps, 90 Min [2026]
- How to Use Grok Imagine Image 2.0: 12 Steps, 70 Min [2026]
- How to Use Stable Diffusion 3.5: 11 Steps, 90 Min [2026]
- How to Run FLUX Locally in ComfyUI: 13 Steps, 90 Min [2026]
- Best AI Models 2026: Full Comparison Hub
