A wave of image editors built rapidly with generative-AI assistance is drawing criticism over performance, file compatibility and the economics of cloning mature creative software. In a technical commentary, Shokunin Studio said it had tracked roughly a dozen projects attempting to reproduce Photoshop and other established applications, alongside numerous forks. The author’s conclusions are based on personal testing and should be read as an assessment, not an independent benchmark.
The central complaint is that visible feature lists have advanced faster than the underlying graphics engines. According to the studio, none of the projects it examined matched Photoshop’s brush performance, with responsiveness degrading on canvases larger than 8,000 by 8,000 pixels. In the most prominent unnamed project, the author also observed delays and brief freezes when using copy-and-paste operations and the undo stack on billboard-scale canvases.
Professional image editing requires more than reproducing familiar panels and tools. The assessment identified incomplete file-format compatibility, lossy editing paths, reduced-resolution previews for filters, inconsistent colour-profile rendering and limited soft-proofing. Font availability can also depend on licensing. These details are particularly important in print, photography and production workflows, where predictable colour, large files and reversible edits matter as much as whether a feature appears in the interface.
The post also challenges claims based on application download size. A smaller binary does not by itself indicate efficient or maintainable code, because frameworks, media assets, models and bundled libraries often account for much of an application’s footprint. Runtime memory, stability and performance under realistic work are more useful measures. An editor that opens quickly but stalls on large documents may still be unsuitable for professional use.
Shokunin Studio further questions the cost case for AI-assisted clones. The author estimated that individual efforts can spend thousands of dollars on model usage and claimed that the largest project had attracted more than 100 contributors generating large numbers of pull requests. Those spending estimates were not independently documented in the supplied evidence, so they should not be treated as audited project costs.
The critique does not show that AI-assisted development cannot produce a competitive editor. It does show why rapid interface replication is different from rebuilding decades of specialised engineering. Brush engines, colour management, nondestructive operations, format fidelity and cross-platform font behaviour are deep systems problems that require testing against demanding workloads.
Open-source creative tools can mature over long periods, as established projects demonstrate, but a burst of generated code is not a substitute for that process. The current crop’s progress will be judged through reproducible benchmarks, compatibility tests and sustained maintenance—not screenshots, contributor counts or the speed at which new features are added.
