Synergised Consulting
Technical asset

Seven AI-Rebuilt Adobe Clones Are Free. Usable? Barely, and That Is the Point

Synergised Consulting Ltd

5 min read
Title card: Seven AI-Rebuilt Adobe Clones Are Free, with the line Technical Asset: usable today? The measured answer.

Last Updated: 2026-10-08

Seven clean-room Rust reimplementations of Adobe's flagship applications were published free and open source at the end of September, built largely with Anthropic's Claude Opus 5.5 coding agents. They are real, they run, and the project's own parity documents say none of them is ready to replace the original for daily professional work. The usable part today is not the software. It is the method: for the first time, a competing product shipped with its own measured gap visible to everyone, and that is exactly the evidence standard missing from most software a small business already pays for.

Developer Brandon Thomas released the seven apps under the ArtCraft banner, covering Photoshop, Illustrator, Premiere, Lightroom, After Effects, InDesign and Acrobat Pro. According to Ars Technica [1], Thomas has about a decade of experience with large-scale Rust projects and has coded exclusively with AI tools since February. The apps are free under MIT and Apache licences, and the project is funded by selling token-based access to an existing visual AI model rather than by charging for the apps themselves. The repositories went public between 30 September and 1 October, and PhotoCraft, the Photoshop equivalent, had drawn thousands of GitHub stars within a week [2].

What the Project's Own Numbers Actually Say

The honest measure of this project is not its ambition but its own published parity documents, and those documents are more sceptical than the headlines. According to OraRouter's review of the repositories [2], PhotoCraft has all 625 of Photoshop's menu items wired to live commands, which the project reports as 100% coverage. The same document estimates real functional parity on mainstream workflows at roughly 86%, falling to about 75% once Adobe's AI-powered features are counted. Pixel fidelity is lower still: the project's own composite oracle matched Photoshop's actual output on 115 of 170 test files, about 68%.

Coverage Is Not Capability

The gap between those numbers is the whole story for anyone choosing software. A menu item being wired means the command exists and runs; it does not mean the output matches Photoshop's. The project's own materials concede that several commands are faithful but approximate: Camera Raw ships without masking or calibration, Vanishing Point arrives without its dialogue, and selection tools that Adobe drives with trained models are heuristic guesses. OraRouter's analysis [2] also records the pace, 398 menu items wired in a single day, as evidence of how fast agent-built breadth moves compared with depth. The project's own status banner carries the same verdict in plainer words for anyone who scrolls past the celebration: early alpha.

Rebuilding what software does, without copying its code, has repeatedly been found legally acceptable, and these projects sit inside that tradition. According to Ars Technica [1], clean-room reverse engineering, replicating functionality from observation and public specifications rather than decompiling proprietary code, is a long-standing practice that courts have generally treated as lawful. Thomas's repositories state that they worked from public documents such as Adobe's published PSD file-format specification and from observing behaviour, not from Adobe's code.

The open risks sit elsewhere. Trade dress is the first: if an app imitates the appearance of Adobe's interface too closely, that copying of look and feel, as distinct from function, is where a legal challenge becomes credible [1]. The second is contract, analysed in detail by Licenseware [3]: anti-reverse-engineering terms in a vendor's licence bind only those who accepted the licence, and a team that worked purely from public specifications and observed behaviour may be outside that net, but the first court cases on exactly this question are still to come. Trademark law stops a clone using Adobe's names, which these apps do not; it does not stop the same features existing [3].

Can a Small Business Use Any of This Today?

Not as a replacement, and the project says so itself. The practical test for this software today is not "is it as good as Photoshop" but "does it open my files and complete my specific tasks correctly", and the 68% pixel-fidelity figure [2] is exactly the kind of number that makes that test necessary before relying on it for anything client-facing. In discovery workshops, the workflow list is rarely handed over by the owner: it gets built in conversation, walking a normal week task by task, and the repetitive screen-based work separates itself from the tasks that genuinely need a person. The same walk-through is the right test for any tool, including a free one: open the files your business actually lives in, run the operations you actually perform daily, and judge the output, not the menu.

There is a real cost angle worth watching rather than acting on. Adobe subscription spend is a line item many founder-operated businesses carry without measurement, and credible free competition in creative software is new. But switching tooling mid-operation carries its own cost in retraining and file-compatibility risk, and the project's own numbers do not yet clear the bar that switch would require. The correct present-tense action is to know your own usage, so that if the parity numbers reach genuinely usable levels, the decision takes an afternoon instead of a research project.

What This Signals Beyond Creative Software

The transferable fact is that a codebase can now arrive with a measured self-assessment attached, and the measuring is the part worth copying. Every business runs on software nobody verified: the booking system, the accounting package, the industry-specific platform that one former employee chose. PetaPixel's coverage [4] makes the distinction precisely: 100% feature parity is not necessarily 100% functionality or performance parity, and the difference between those claims is where buying decisions go wrong. A task-level verification checklist, written before a renewal date and counting what the tool actually completes, converts a vendor's claim into your evidence.

Sources

  1. [1] Ars Technica, "Software is over": Bold AI developer takes aim at Adobe with open source clones,
  2. [2] OraRouter, Claude Opus 5.5 Built 7 Adobe Clones in Rust: Reality Check,
  3. [3] Licenseware, When AI Can Rebuild Any App: Licensing, IP and Vendor Lock-In,
  4. [4] PetaPixel, Someone Rebuilt Free, Open-Source Versions of Photoshop, Premiere, and Lightroom Using AI,
Keep reading

Related reading

Title card: Testing An AI Agent Before It Goes Live, First 100 Days Operations

How Do You Test An AI Agent Before It Touches A Live Business Workflow?

Test an agent with an eval set of 20-50 tasks from real work, sandbox the tools, run it in shadow mode beside your current process, graduate autonomy on evidence, and turn every failure into a permanent regression test.

Read this
Title card: An Agent Has No Memory, with the supporting line What to store, what to expire, how to test it, on Synergised brand tokens

Your AI Agent Doesn't Have a Memory. It Has a Context Window -- And That Is the Problem

Agent memory fails quietly: context degrades answers and one poisoned write persists. What to store, what to expire, and how to test the store.

Read this
Title card: a 1.3B parser reads messy scans while a separate 2B checker model decides which extracted pages need human review

A 1.3B Model Reads Your Scans. The Interesting Part Is the 2B Model Checking Its Work

OpenBMB released a 1.3B document parser alongside a separate 2B checker that scores OCR output and flags what a human must review, and the split is what makes document automation safe near financial records.

Read this