Two announcements from August 2026 show the same development from different directions: media assets that previously required post-processing are increasingly being created fully finished already at the generation stage.OpenAI has unlocked a preview function for transparent backgrounds in the programming interface for its image model GPT-Image-2.Adobe is making the audio tools in Firefly available to all users.

Alpha channel instead of cut-outs

Until now, cutting out was a separate work step: an image is generated, then a tool removes the background and creates a transparency mask.With GPT-Image-2, according to OpenAI, the alpha channel – the additional image information that defines which pixels are transparent – is created directly during image generation.This is activated via a single parameter in the API call.

The difference sounds technical but has practical consequences.Subsequent background removal works with an estimate: a second model tries to detect where the subject ends.With hair, smoke, glass or soft edges, this regularly leads to fraying or colour fringes.If, on the other hand, the transparency is created during the generation process, this guessing step is eliminated.

Beyond the description as a preview function, no reliable information is available on the quality, limits and pricing of the function.Anyone who wants to use it productively should take the preview status seriously: functions at this stage can change or fluctuate in availability.

Firefly now also covers the audio track

In parallel, Adobe has made three audio functions generally available in Firefly.Generate Music creates music tracks, Generate Speech voice recordings and Generate Sound Effects individual sounds.All three are geared towards use in video projects, and the results are described as royalty-free.

In addition, Adobe is integrating external models into Firefly, including Gemini, Runway and Kling.Firefly is thus developing less into a single model than into an interface in which several providers are available side by side.For teams, this means: switching tools between providers is no longer necessarily tied to switching the working environment.

The common denominator: fewer intermediate steps

Both announcements concern production steps that regularly cost time in web and campaign projects without creating conceptual value.A product image for a landing page must be cut out, otherwise it will not fit on coloured surfaces.An explainer video needs an audio track, otherwise it looks unfinished.Until now, both were either manual work or a question of stock material and licence research.

When these steps move into the generation process, the effort shifts.It does not disappear: instead of cutting out and searching for audio, the focus is on formulating good prompts, selecting among several results and quality control.The difference is that this work is closer to the design decision and contains less routine.

What this means for the asset pipeline

The API availability of GPT-Image-2 is interesting above all for automated workflows.Anyone who does not create image assets individually in an interface but programmatically – for example for a product catalogue with many similar representations – can set the transparency parameter directly in the call.A downstream processing step with a separate cut-out tool is thus eliminated.

For integration into a content management system such as TYPO3, this is relevant because cut-out images are more flexible in output.A PNG or WebP with an alpha channel can be used against different background colours without having to maintain a separate image for each variant.This reduces the number of assets in file management and thus the maintenance effort.

One thing to note: transparent formats are generally larger than comparable JPEGs.Anyone delivering many cut-out images should keep an eye on image compression and delivery in multiple resolutions – otherwise the production effort saved will be used up again by longer loading times.

Licence questions remain a management task

Adobe describes the audio content generated in Firefly as royalty-free.This is an essential point for companies because music rights regularly lead to queries in video and campaign projects.At the same time, the integration of external models such as Gemini, Runway or Kling changes the starting position: what applies to Adobe's own models does not automatically have to apply in the same way to connected third-party models.

For practice, a simple rule is therefore advisable: document which model was used to generate an asset.In larger projects with several participants, this information can otherwise no longer be reconstructed after a few months – and that is exactly when it is needed.

Assessment

Taken on their own, the two announcements are not a revolution.They describe detailed improvements to tools that many teams use anyway.In sum, however, it is becoming apparent that generative tools are moving from pure idea generation towards production-ready output: a result that can be adopted into a layout or a video edit without an intermediate step.

For web and software projects, this means above all re-examining the asset workflow: which post-processing steps are still necessary, which can be shifted into generation and where is design control still needed?Anyone who answers these questions properly once will save time on every following campaign.

At the same time, those responsible should document the origin, licence status and technical properties of generated assets from the outset – this can only be caught up on later with considerable effort.

Sources