The README promises a wardrobe app, while the package installs a downloader
The README describes Dress AI Service as a private wardrobe catalog with automatic tagging, outfit recommendations, weather-aware styling, and generated outfit pictures. That would be a useful self-hosted product. The repository does contain FastAPI models and routes for wardrobe items and outfits, so the pitch is not detached from every file. The problem is that the pieces do not form the application the page says you can start.
The clearest conflict sits in setup.py. It names the package coomertool, registers a coomertool command, and describes a downloader for Kemono and Coomer archives. That downloader occupies its own top-level directory beside the wardrobe code. In a checkout with only 32 files and about 1,902 source lines, this is a major identity split rather than incidental sample code.
Missing files keep the FastAPI service from being a usable base
The application tree is incomplete. app/routers/outfits.py imports app.services.outfit_engine and app.services.image_gen, while app/routers/generate.py also imports the image generator. Neither module exists at commit 0f6af90 or on the current main branch. app/main.py mounts app/static, another absent path. These are startup dependencies, so a buyer would have to implement or recover them before evaluating the advertised outfit flow.
The Dockerfile copies missing inputs: scripts/, migrations/, and alembic.ini. Its compose file maps port 8080, mounts data and model directories, and offers optional Ollama and NVIDIA overlays. Those instructions look complete at a glance. The 1.5 MB checkout tells a less flattering story because required application material never made it into the tree.
What happened when we ran it
Our run used commit 0f6af90 in a fresh Debian container with 3 CPUs and 8 GB of RAM. Installation succeeded in 117 seconds, adding 158 packages and leaving 5,880 MB on disk. The build step succeeded in 1 second. Those results show that the dependency resolver and the harness build command completed. They do not show that the FastAPI process served a request, and the supplied lab record does not claim that it did.
There was no test script or target, so our harness skipped tests. The checkout also had 0 CI workflow files and no tests directory, even though the README tells contributors to run pytest tests/ -v --cov=app. Pip-audit reported 228 known vulnerabilities in the installed environment. That figure needs remediation and triage before any deployment, especially because the dependency set includes web handling, image parsing, model runtimes, and cryptography packages.
The implemented feature set is much smaller than the checklist
The tagging service loads CLIP for clothing category guesses and uses K-means for dominant colors. Fabric and pattern recognition are placeholders that always return cotton and solid. Brightness can label an item minimalist or luxury, and color rules add tags such as trendy. That is prototype logic. It cannot support the README's broad claims about automatic fabric, pattern, style, and wardrobe analysis without more work and validation.
The 32-file repository also lacks a visible authentication layer for its photo routes. The checked-in upload endpoint writes the original and a thumbnail, while list and delete operations depend only on an item ID. Configuration defaults to host 0.0.0.0 and a secret value of change-me-in-production. If you repair the missing modules, keep port 8080 off the public internet until you add access control, request limits that are actually enforced, and an image retention policy.
A September 26 push does not establish product maintenance
GitHub recorded the last push on September 26, 2026, and the repository had 1,221 stars when fetched. The recent sequence is dominated by README edits, including the five latest commits, while the latest code-shaped upload was a zip file. The project has no GitHub release. Its single open issue has no replies, and the only recently closed pull request corrected README grammar. That is activity, but it offers little evidence that runtime gaps are being worked through with users.
The lab snapshot gives the health signals sharper edges: 0 CI workflows, no tests directory, 228 known vulnerabilities, and no runnable test target. A small project can be young without being careless, but here the documentation claims a complete platform while the source omits its central engines. Stars do not resolve that conflict. A release containing a complete tree, a verified Docker build, and tests that exercise upload through visualization would materially change this assessment.
Focused virtual try-on projects are better starting points
OpenTryOn is aimed directly at fashion AI APIs and models. FASHN VTON v1.5 and IDM-VTON narrow the job further to virtual try-on research and inference. None gives you the exact wardrobe dashboard promised here for free. Their narrower boundaries are useful because you can evaluate the garment-transfer component before writing catalog, identity, storage, and consent controls around it.
For this repository, the buying decision is unusually clear. Our 117-second install and 1-second build completed, yet 5,880 MB of dependencies led to no test run and an audit count of 228 known vulnerabilities. The source then asks you to supply the missing image generator, outfit engine, static app, and Docker inputs. Start elsewhere unless your actual goal is to document and repair this incomplete scaffold.