Adobe launches Firefly Graph beta to make creative workflows reusable

|Author: QUASA Editorial Team|5 min read| 45
Adobe launches Firefly Graph beta to make creative workflows reusable

Adobe introduced Firefly Graph in beta at Adobe MAX in Miami Beach on November 10, 2026. Adobe’s Graph product page identifies the release as a public beta: creators with an Adobe ID can connect AI models, assets and Creative Cloud operations as nodes, run the sequence and save it for reuse. The use cases include character studies, shot lists and applying brand guidance across campaign assets.

Creative Bloq’s MAX coverage describes Graph as a way to chain tools and models, save the workflow and run it again within a broader Firefly environment that brings Adobe and third-party models together. For a creator, the lasting output can be the production sequence as well as the image or video it produces: change an input, rerun the connected steps and review the new result.

What Graph’s connected nodes automate

Each node carries out an action, and a connection passes its output into the next node. An input might be an existing asset or a text prompt; subsequent nodes can generate content, select a result, mask part of an image, edit it or composite it. An output node receives the resulting asset. Graph runs the connected sequence and displays what each stage produced, so a creator can locate the step that needs adjustment.

This structure is useful when the same kind of asset needs several versions. The creator can replace a source image, change a prompt or adjust a node while retaining the connections and decisions that still apply. A repeat run uses the revised input and the saved sequence; it does not guarantee that the new asset will match the previous one in every creative detail.

Graph also allows models from Adobe Firefly, Google and OpenAI on the same canvas, including comparisons of outputs from a shared input. A built-in AI Assistant can assemble or refine a graph from natural-language instructions, while users can connect nodes themselves. The visual process does not require code, although a multistep graph still asks its creator to decide which operations and inputs belong in the sequence.

A starter workflow from source asset to reusable output

A character study develops into storyboard frames, illustrating a reusable Firefly Graph workflow.

Consider a hypothetical campaign image that needs related variations. A creator could begin with an approved source image and a text prompt specifying the intended treatment, then connect those inputs to an image-generation node. After inspecting the generated result, the creator could feed a selected version into an editing operation and connect that to an output node. This is an illustrative sequence based on Graph’s documented capabilities, not an account of a completed production test.

The review point matters because a connected operation will carry forward whatever its preceding node produces. If the generated image changes the subject or composition in an unwanted way, editing downstream may preserve the mistake. The creator can revise the prompt or input, run that part again and inspect the intermediate result before accepting the exported asset.

Once the sequence produces a usable asset, the graph itself can be saved. For a related campaign variation, the creator could substitute a new source image or prompt while keeping the relevant generation, editing and output connections. If the new asset needs a different crop or adjustment, that particular node can change without rebuilding the whole chain. A shared graph can also give another creator the established sequence to run or refine, while the finished asset remains subject to review.

Review and credit costs remain part of the workflow

A creator selects a campaign variation after reviewing outputs from a Firefly Graph workflow.

Graph automates the handoff between connected actions, but the creative decision about an output remains human. In the illustrative campaign, the subject, composition and effect of the edit need inspection before a variation can be treated as approved. Intermediate results help show whether a problem arose during generation or a later operation, making a targeted revision possible.

Model comparison has the same boundary. Several models can produce alternatives from the same input, but a saved graph cannot determine whether a result fits a particular brief or brand on its own. Reusing the process can reduce the need to rebuild connections; it does not remove the need to judge each new output.

Generative-model nodes require credits, while some other nodes do not. The amount varies by model and node and is displayed on the node before generation. That matters when a creator reruns variations or compares models: the credit requirement is attached to the generative action, not to every connection in the graph.

Graph’s public beta beside Adobe’s other MAX releases

Graph’s public beta has a different access status and purpose from Firefly Builder’s private beta. Builder is a design-to-code prototyping tool for interactive websites, presentations and apps using natural-language instructions, with access through a waitlist. Graph instead connects creative tools and models into a repeatable production sequence that can be refined and shared. Always-on agents for recurring creative work were presented as a preview, another distinct status.

Edit Text, Retouch, Relight and Quick Actions were generally available in Firefly at the MAX announcement. Their release does not make Graph a generally available product; it remains a public beta. A creator can access Graph through Firefly with an Adobe ID and begin with a blank canvas, a starter workflow or the built-in assistant. That access makes it possible to build and share a repeatable process now, while approval of each resulting asset remains a separate decision.

Also read:

Share:

Subscribe to our newsletter

Get the latest Web3, AI, and crypto news delivered straight to your inbox.

0