How to use Adobe Firefly for AI video, images, and upscaling in one workflow

Most creative workflows still look like a relay race: generate an image in one tool, hand it off to an upscaler, then drag it into a video app, then export somewhere else to translate or clean up audio. Every handoff is a chance to lose quality, waste time, or just give up halfway through.

Adobe Firefly quietly collapsed that whole relay into a single browser tab. You can type a sentence and get a video clip back that looks camera-shot, generate a still image with light that behaves the way real light does, and blow that image up to print resolution — all without opening a second app or waiting on a render farm.

This guide walks through exactly how that workflow works, what each tool is actually good at, and where the physics of AI generation still gets in the way.

What is Adobe Firefly, and why does it matter now?

Adobe Firefly is Adobe's generative AI suite for images, video, audio, and design assets, built directly into the browser and connected to the rest of Creative Cloud. Instead of being a single model, it's become a hub where you pick the engine that fits the job.

Adobe frames its own goal plainly: the aim with Adobe Firefly is to give you one place where you can choose the right industry-leading model to generate assets for your project, then get exactly what you want by editing those assets with the creative tools and controls Firefly offers.

That "choose the right model" part matters more than it sounds. As of mid-2026, Adobe's live Generate Video guide listed Kling 2.5 Turbo, Kling 3.0, Kling 3.0 Omni, Ray2, Ray3, Ray3 HDR, Ray3.14, Ray3.14 HDR, Runway Gen-4.5, Sora 2, Veo 2, Veo 3.1, and Veo 3.1 Fast as options alongside Adobe's own native model. That's not a gimmick — different models genuinely behave differently on motion, realism, and consistency, and Firefly lets you A/B them without switching platforms.

How do you generate a video from a single prompt?

Getting started is straightforward. Sign in at firefly.adobe.com, and from there:

Sign in to Adobe Firefly with your Adobe ID or create a free account if you don't already have one, then select Generate Video from the Video tab in the left hand menu.
From the dropdown menu, choose the Adobe Firefly video model or a partner model such as Google Veo, or Runway.
Type your prompt — be specific about camera movement, lighting, and subject.
Generate, review, and either keep the clip or regenerate with a tweaked prompt.

Adobe's own Generate Video documentation breaks down what each partner model is actually optimized for. If you want fluid, photorealistic camera motion, one model is known for fluid photorealistic motion and smooth camera movement, bringing stills to life with cinematic motion and natural transitions and consistent subject continuity across frames. If you need dialogue-driven scenes, another option lets you generate production-ready, multi-shot video with native audio, building variations across shots and syncing native dialogue generation to character lip movements. And if precision matters more than speed, there's a model built for advanced control and production-quality output, with motion direction, character consistency, and refined cinematic control across shots with realistic physics and pinpoint precision.

The practical takeaway: don't marry your first result. Since each generation only takes about a minute, fire off two variants with slightly different phrasing and keep whichever one nails the framing and motion you actually needed.

What if your video needs to be in another language?

This is where Firefly's newer tools start to feel less like a novelty and more like an actual production pipeline. Translate Video takes a finished clip and rebuilds the spoken audio in a different language while preserving the original speaker's voice characteristics — no separate dubbing studio required. Pair that with Enhanced Speech, which cleans up audio recorded in an untreated room, and you've got two tools that quietly solve problems that used to require a whole post-production pass.

How does AI image generation compare to video?

Still images are where Firefly's realism claims get tested hardest, because viewers stare at a still far longer than a two-second video frame. The standout cases tend to be anything involving complex light physics — water, glass, reflections — where the model has to correctly bend and scatter light rather than just paint a plausible texture over the surface.

If a generated image is 90% right but has one problem area (a hand, a background detail, a texture that looks off), don't scrap the whole generation. Firefly's Edit tab lets you make targeted regenerations within the existing image instead of rerolling the entire prompt from scratch — a much faster way to iterate toward a usable final image.

Firefly's image side draws from the same "choose your model" philosophy as video: the promotion includes all image models — including Firefly's commercially safe image model and industry-leading partner models like Black Forest Labs' FLUX.2, Google's Nano Banana, and OpenAI's GPT Image. That range means you can pick a model trained for photorealism versus one better suited to illustration or stylized work, all from the same interface.

How do you upscale an image for print without losing quality?

This is the step most creators skip until a client asks for a large-format print from a file that was never shot at that resolution — and then panic. Firefly's Generative Upscale tool exists specifically for that moment.

Here's the actual step-by-step, straight from Adobe's Upscale Images documentation:

Generate an image and go to the Edit tab, or upload an image and select Generative upscale from the left toolbar.
Select a model from the dropdown — your options are Firefly Upscaler, Topaz Bloom, or Topaz Gigapixel.
From the Scale menu, select 2x or 4x to define how much you want the image to be scaled.
If you're using Topaz Bloom, move the Creativity slider to the left to stay closer to the original image style, or move it to the right to introduce more detail and enhance creative elements during generation.
If you're using Topaz Gigapixel and working with faces, select Face recovery to preserve face details and expressions in blurry or low-resolution faces.
Select Upscale to generate the upscaled image.

Each model has a distinct personality. According to third-party testing of Firefly's model options, the Firefly Upscaler is built for well-balanced, natural outputs on everyday tasks, Topaz Bloom is suited for adding extra detail and a more artistic output, and Topaz Gigapixel is the top choice for achieving maximum sharpness and preserving detail. If you're prepping a hero shot for print, Gigapixel with face recovery turned on is usually the safest starting point.

Why does upscaling sometimes still look soft?

Here's the honest limitation: generative upscaling can't invent detail that was never captured in the first place. If the source image has background text that's too blurry to read, no upscaler — Firefly's or anyone else's — can reconstruct real letterforms from nothing. Same goes for heavily compressed source files, which tend to retain a faint smeared quality in flat color areas even after upscaling. This isn't a Firefly-specific flaw; it's a hard physical limit every upscaling model runs into. The practical rule: if the source file is too degraded to read clearly at 100% zoom, don't expect the upscaler to save it — reshoot or re-source the original instead.

For context on scale, Adobe's broader upscaling work in Photoshop can push files even further — Generative Upscale can increase image resolution by up to eight megapixels, offering scaling options of 2x, 3x, or 4x — which is useful to know if Firefly's web upscaler doesn't get you all the way to your target print size.

What tools round out the workflow?

Beyond generation and upscaling, a couple of newer additions are worth building into your process:

Remove Background from Video — Firefly can now automatically identify the subjects in a clip and remove the background throughout the clip, which used to require rotoscoping software or a green screen.
Topaz Astra for video — for footage that needs a resolution bump rather than a full regeneration, Topaz Astra in Firefly Boards lets you push footage to 1080p or 4K, boosting low-resolution clips for platforms like YouTube or restoring older, low-quality footage.
Auto reframe — resizing a single generated video into multiple aspect ratios for different social platforms without manually re-cropping each one.

Worth checking before any commercial deployment: Adobe explicitly notes that for partner models, Adobe tells users to review the specific model's terms before commercial use and says it cannot verify the partner's training data or whether outputs may contain third-party IP. If you're using a partner model like Veo or Runway inside Firefly for client work, don't skip that terms check.

What's the smartest way to actually use this stack?

Treat Firefly less like a single tool and more like a small production team living in one tab. Generate two takes of anything before committing to one. Push stills through the model that matches your subject — Gigapixel for faces and product shots, Bloom for anything you want to feel a little more painterly. And don't fight the tool on physics it can't solve; a badly compressed or under-resolved source image will always cap what any upscaler can deliver.

The bigger shift here isn't any single feature — it's that video generation, image generation, upscaling, translation, and audio cleanup now sit inside the same interface instead of five different subscriptions. That's less about any one output looking impressive and more about how much friction disappears when the entire loop, from typed prompt to finished asset, never asks you to leave the browser.