Artificial Intelligence (AI) is expanding from a standalone image generator into a creative studio for graphics, audio, video, and simple games. This affects creators, small businesses, educators, and anyone producing media without wanting a separate specialist tool for every step. New features from Adobe, OpenAI, and Meta show which parts of a production can already be created directly inside an AI tool.
What can AI already handle?
Adobe is adding tools to Firefly that cover several components of video production. According to the report on Firefly’s new features, Generate Music creates royalty-free tracks for videos, Generate Speech turns scripts into voice recordings, and Generate Sound Effects produces audio for individual scenes. All three features are broadly available, although the source does not specify supported languages, regions, or prices.
This allows more of a typical production process to remain within one platform. For a training video, for example, you could turn a written script into speech, generate background music, and add an appropriate sound to an action shown on screen. Adobe also offers free daily generations through the Firefly AI Assistant and says Create Storyboard and Create Brand Kit are already among its most-used features.
A storyboard is a visual plan of individual scenes, while a brand kit collects design rules such as colors and recurring brand elements. Firefly also provides access to external models including Gemini Omni Flash, Runway Aleph 2.0, and Kling 3.0. That makes the platform a control center for different models, but it does not guarantee that their output will share the same style or quality.
Adobe describes its audio tools as commercially safe to use. This is a provider claim and has not been independently verified in the supplied sources. The term “royalty-free” also does not answer every possible question about using a generated track in a specific client project.
How are images and video changing?
For images, work is also shifting from editing after the fact to directly generating production-ready components. OpenAI’s GPT-Image-2 can create PNG files with transparent backgrounds through a preview feature. Transparency means the subject can be placed over another image, slide, or layout without a visible rectangular background.
In its description of transparent image output, OpenAI lists four uses: product images for online stores, presentation diagrams, icons and stickers, and artwork for merchandise. A small online store could generate a cutout product visual for several different backgrounds. Someone preparing a presentation could insert a generated diagram as a graphic element, but would still need to verify its numbers manually.
According to OpenAI, generating transparency directly should preserve difficult details such as clear glass or thin fibers better than removing a background afterward. However, the feature is available through an application programming interface, a technical connection that lets other software communicate with the model. The source says setup requires Python, additional software libraries, and an access key, so it is not yet a particularly easy entry point for everyday users.
In video, the development already goes beyond isolated effects. Several prominent YouTube filmmakers demonstrated Higgsfield and its Seedance 2.5 feature, including footage of a real person placed inside science-fiction settings. Another video combined its creator’s professional history with a fictional trip through Mauritania assisted by a robot.
The coverage of the Higgsfield videos also describes the generated imagery as distinctly uncanny. The demonstrations therefore show both sides of the technology: AI can combine real footage with invented worlds, but visible inconsistencies remain. If you need a convincing final product, selection, editing, and quality control are still your responsibility.
Where does the creative studio fall short?
The clearest limitations appear in more complex interactive projects. A so-called Gauntlet Loop for Claude Opus 5 splits a game idea into smaller tasks: a “Builder” creates each component, while a “Critic” compares it with a reference and requests revisions. Prompting means writing instructions for an AI system.
According to its creator, this method produced a playable Call of Duty imitation containing 55,000 lines of code from one initial prompt. A dedicated website collected 47 browser games created with the loop. However, a test of ten Claude-generated games produced disappointing results: many closely followed established references rather than introducing original game ideas.
The tested Kart Royale visibly borrowed its basic concept from Mario Kart, offering eight drivers and a single course. One game ended after three to five minutes, and the review criticized issues including poor balancing, meaning that rules and difficulty were not calibrated fairly. AI created a playable product, but showed little understanding of what makes a game engaging and fair over time.
Meta is taking a more accessible approach with Pocket. The experimental app lets people create small games through text instructions and publish them to a scrolling feed. These games can respond to touch and the tilt of a phone, play sounds or clips from songs, and use photos from the camera roll or access the camera.
Other users can save Pocket games, repost them, or remix them into new creations. According to TechCrunch’s report on Pocket, the app is rolling out to all users in the United States after a test in Brazil. Availability in Switzerland is not mentioned.
Pros and Cons of the AI creative studio
Pros:
- Fewer tool changes – Images, speech, music, and sound effects can increasingly be created within the same production environment.
- Faster drafts – Storyboards, cutout product visuals, and simple playable ideas can become visible and testable earlier.
- Lower barriers to entry – Apps such as Pocket turn text instructions into interactive results without requiring you to write code yourself.
- Easier variation – Generated media and games can be adapted, recombined, or prepared for different backgrounds.
Cons:
- Inconsistent quality – Uncanny video imagery, inaccurate diagrams, and poorly balanced games still require correction.
- Limited originality – The tested games relied heavily on familiar references and offered little creative novelty.
- Unclear commercial interests – Enthusiastic demonstrations may be paid partnerships without initially appearing to be labeled as such.
- Restricted access – Some features require technical setup, while others, such as Pocket, are only available in certain countries.
The dispute surrounding the Higgsfield videos also shows that the output is not the only issue. Fans suspected paid promotion after other creators apparently shared partnership offers from public relations firms working for Higgsfield. The videos were not labeled as advertisements; after the report was published, a Higgsfield representative said the creators had been compensated through a negotiated combination of money and platform credits.
Some criticism also focused on comparing generative AI with an earlier, more affordable video camera. The objection was that people still had to learn how to use a physical camera, while generative AI is trained on human-made material without crediting its creators. The supplied sources do not provide a final legal assessment, but they make the cultural conflict visible.
What does this mean for your work?
If you are just starting, a clearly limited media component is the most practical first step. You could generate a temporary voice track and one sound effect for an internal explainer video, or test a transparent product visual for an online store. Compare the result with your script or the original product, and check numbers, pronunciation, visual details, and usability before publishing it.
For advanced users, the greater benefit lies in a multistep production. You can begin with a storyboard, select image and video variations, and then add music, speech, and scene-specific sounds. More automation does not mean less editorial work: style inconsistencies, incorrect data, and weak game mechanics become easier to miss as the system produces material more quickly.
Specific questions remain unanswered for Switzerland. The sources do not state which languages Adobe’s new speech feature supports, nor do they provide Swiss pricing or special privacy terms. Pocket is currently described specifically as a U.S. rollout, while no Switzerland-specific limitations are given for the other tools; an absence of restrictions in the reports is not confirmation of full availability.
AI now handles a surprising number of individual production steps and increasingly connects them inside one interface. It is most convincing for drafts, variations, and clearly defined media components, but less reliable with exact information, original game design, and consistent artistic quality. The unresolved risk is not that the systems produce nothing useful, but that a quickly generated and plausible-looking result is treated as finished work without proper review.
Sources
- Adobe Firefly generiert jetzt Musik, Sprache und Soundeffekte für Videos – Unknown, 2026-08-20
- Major YouTube creators are facing backlash for accepting AI money – Unknown, 2026-08-21
- Gamifizierter KI-Slop: Wir haben von Claude erstellte Videospiele getestet, damit ihr es nicht müsst – Unknown, 2026-08-21
- OpenAIs Bildmodell GPT-Image-2 kann jetzt Bilder ohne Hintergrund direkt generieren – Unknown, 2026-08-21
- Meta brings Pocket, an app that lets you vibe-code and share games, to US users – Unknown, 2026-08-20
