AI-Built Adobe Clones Are Free. Are They Usable?

By Lazar Milicevic · Published October 12, 2026 · 8 min read

Free, open-source Adobe alternatives built with Claude Opus just landed, and the people building them say they're nowhere near finished. If you run a small shop that ships client files every week, the license fee isn't the real cost. Rework is. Here's how to test one of these tools against a real job without betting your delivery schedule on it.

What actually shipped (and what didn't)

The news is an ambitious set of free apps, not a replacement for anything. Developer Brandon Thomas says he used Anthropic's Claude Opus 5.5 to build "clean-room replacements" of Adobe apps in Rust, with WebAssembly versions that run in a browser. Ars Technica's summary of the story: ambitious, free, and nowhere near finished.

The suite is called ArtCraft, with seven apps:

Naming varies across coverage (some outlets call the Acrobat clone PdfCraft), so check the official repo for current names before you search for downloads.

Licensing is permissive: the apps are free and open source under MIT or Apache 2.0. Thomas says development will be funded by selling token-based access to ArtCraft's existing image model. That AI access is optional, and the editing apps don't depend on it. Treat "fully funded" and "won't be abandoned" as his assurances, not guarantees.

The maturity signals are blunt:

• Thomas first promised 100% feature parity within a month on Reddit. Days later on Hacker News he softened that to 99% parity in "months and not years".
• He also said the apps are in an early alpha state and "not anywhere close to ready".
• The PhotoCraft repository was created on September 30, 2026. The project is days old.

That's the whole story. Not "Adobe replaced." Not "cancel your subscriptions." A fast-moving prototype, honestly labeled.

The gap between a screenshot and a deliverable

A design tool is finished when it survives your Tuesday revision, not when the interface looks convincing. It has to open the files you receive, keep the parts you need editable, export what the client expects, and let someone else open the result.

Picture a small agency graphic. The designer gets an existing file, changes a price, swaps an image, exports the approved format, and sends it for review. Now break it three ways:

1. The import mangles the layout.
2. Editable elements get flattened into pixels.
3. The export differs from what the client approved.

Any one of those turns a $0 license into hours of rework. The same principle holds outside design. An invoicing tool isn't finished when it renders an invoice on screen. It's finished when the right data reaches accounting and someone can audit a correction later.

This is the lens I use on every tool I evaluate, including the ones I build. In my own automation work, the interface is the smaller part of the job. The handoffs are where things go wrong.

What PhotoCraft's own documentation admits

The project is unusually candid, and that candor tells you exactly where to aim your test. PhotoCraft's own README says it's early alpha and not yet a Photoshop replacement for daily professional work. The gaps it lists, per MadRobot:

• AI/generative features
• About twenty missing tools (this count moves as the repo updates, so check it the day you test)
• Typography and pro workflows
• Plug-in support

Typography and pro workflows are the ones that bite agencies. If your client files use specific type treatments or rely on plug-ins in the pipeline, those gaps are a fail before you start.

The most useful number for a handoff test is the round-trip data. In the project's own tests, PhotoCraft re-saves PSDs that render the same for 307 of 309 files in the psd-tools test set and 169 of 170 in another set. Read that carefully:

• These are the project's own results, not independent ones.
• "Renders the same" means the same in PhotoCraft's own pipeline. It doesn't tell you Photoshop will accept the file.
• A PSD re-saved by PhotoCraft is not byte-identical to the source file.

One secondary source claims the roadmap admits PSDs PhotoCraft writes aren't yet reliably accepted by Photoshop. I couldn't confirm that from a primary source, so I'm not asserting it. But it's exactly the question your test should answer, because your client's machine runs the real thing.

Reading the issue tracker like a due-diligence report

Open issues tell you what breaks in practice, and for a project this young the tracker is the most honest document available. Around 8 October 2026, PhotoCraft's tracker had 447 open issues, including a crash opening a 32-layer PSD and PSD files that wouldn't open at all. Star and issue counts differ across sources and change daily, so check the live repo rather than trusting any snapshot, including this one.

Qualitative reports line up with that. An early Hacker News tester who said they'd worked in design and film reported laggy controls, broken text editing and shortcuts that did nothing. FilmCraft's team puts its readiness for real projects at about 50 to 60%.

Two more risk factors before you install anything:

• Platform: the authors name macOS as the best-tested platform, with Windows and Linux little exercised. Run your test on macOS first, or treat other platforms as higher risk.
• Legal posture: "clean-room" is the project's own description, not a legal determination. Ars (via tbreak) notes that a close copy of Adobe's look could raise trade-dress questions. If you deliver client work, read the license and check how the tool handles files before you load client material into it.

I'd also hold off reading too much into readiness percentages. Different outlets cite different roadmap figures, and the roadmap changes by the day.

The pass/fail test (a script you can actually run)

One repeatable task, one real file, one person who normally receives the output. That's the whole test. Don't audit the suite; audit one job.

Write the spec before you touch the tool. I keep it as a small file next to the test copy:

# handoff-test.yaml
task: "Update price + swap hero image on client promo graphic"
source_file: "promo_v7.psd"        # a COPY you're allowed to use
platform: "macOS"                   # best-tested per project authors
must_stay_editable:
  - text layers
  - smart objects
  - layer names/groups
required_output:
  format: "PSD + PNG export"
  receiver: "account manager (opens in Photoshop)"
sign_off: "Dana"
time_budget_minutes: 30            # what this takes in your current tool

Then run it in five steps:

1. Copy the file. Never test on the original. Confirm you're permitted to use the client material at all, and check the project's license and file handling first.
2. Do the edit in the alternative. Change the price, swap the image, export.
3. Hand the export to the real receiver. The person who normally gets the file opens it in their normal tool.
4. Request one ordinary revision. Something mundane: "make the headline two words shorter."
5. Try to go back. Can you return to your original tool without rebuilding the work?

Log three things, nothing more:

cat >> handoff-log.md <<'EOF'
## Run 1 - promo_v7.psd
- Changed unexpectedly: 
- Extra time vs. 30 min budget: 
- Could return to original tool without rebuild: yes/no
EOF

Pass criteria are binary. It passes only if the receiver opens the export cleanly, the editable elements stayed editable, the revision took roughly your normal time, and you could back out. One failure on any of the four is a fail. If it passes, expand to a second file type. If it fails, you learned something for free and delivery never wobbled.

What to use instead while the clones mature

You don't have to choose between paying for Adobe and gambling on an alpha. For PSD handling specifically, Photopea is the established free browser-based comparison point: it opens PSD files with layer structure intact, edits them, and saves them again as PSD. It's a mature tool with a track record, which is the thing the new clones can't claim yet. Run the same handoff test on it as your baseline. If Photopea passes and PhotoCraft fails, that gap is your honest measurement of how early the clone is.

Pricing for these alternatives is inconsistent across the aggregator posts I found, so verify current plans on each vendor's site rather than trusting a comparison article.

Meanwhile, the reason this matters beyond design: most small-business pain isn't a missing app, it's the seams between apps. If you're losing hours a day to email, invoicing, lead follow-up, and reporting because your tools don't talk to each other, a free Photoshop clone fixes none of it. AI-built open-source tools might become useful components later. A component isn't a workflow. Handoffs, permissions, error checks and human approvals still have to be designed by someone.

Where this fits in a real workflow stack

I'm Lazar Milićević, a senior engineer who builds AI automations for solopreneurs and small teams, and the same pass/fail discipline above is how I evaluate any new component before it gets near a client system: one real task, one real receiver, a way to roll back. If a free tool earns a slot in a workflow, great. If it doesn't, the test tells you in an afternoon. More at bizflowai.io.

My take: the most valuable AI-built software won't win by looking like an existing app on day one. It'll win by completing a specific job reliably, including the boring handoffs. These Adobe alternatives are worth watching because they show how fast new options can appear. On the facts reported so far, they're ambitious prototypes, not production replacements. Test one file. Let the file decide.

Want more like this?

Planning an AI automation project or need a second opinion on your architecture?

Connect with me on LinkedIn — Lazar Milićević, senior engineer building practical AI automation.

Visit bizflowai.io for services, case studies, and AI consulting.