Kombin AI Generation
Upload & Submit
Person photo + 1-2 items with labels.
PersonRequired
Item 1Required
Item 2Optional
Result
Waiting for a submission
Status: idle
No output yet. The pipeline will show the result here.
Attempts0
Validation-
Submit an outfit to see the latest validation notes.
Worked Examples
3 real runs through the pipeline — inputs, output, validation result, and commentary.
Graphic Tee — Studio Try-on
Classic
Person
Person input
→
Output
Output
PassedScore: 100/100No changes needed; the virtual try-on result is accurate.
[TASK] Photorealistic virtual try-on: replace the specified garment(s) on the person using the provided item reference image(s) as the exact source of truth.
[SCENE — DO NOT CHANGE] Do not change the background, floor/ground, ambient shadows, or any non-garment element.
[REPLACEMENT] Image [2] = Item A → Type: t-shirt | Color: white | Pattern: graphic
  Graphic/notes: 'Aelfric Eden CHERRIES', cherry artwork.
No style direction was provided, so the system defaulted to Classic Mode. The strict scene-lock rules preserved the original neutral background and lighting exactly. The vision analysis correctly identified the Aelfric Eden cherry graphic and reproduced it with full fidelity — text, colors, and artwork all match the reference. Validated on the first attempt with a perfect score.
Cable-Knit Sweater — London Street
Hybrid
Person
Person input
→
Output
Output
Style: “londra sokağı akşamüstü”
PassedScore: 100/100No changes needed; the generated output meets all criteria. [Weighted score: 100/100, threshold: 65]
[STYLE & ENVIRONMENT] Street fashion try-on scene, model wearing the product naturally, cinematic London street afternoon background, soft golden hour lighting, warm light reflecting off buildings, relaxed atmosphere, urban lifestyle aesthetic, shallow depth of field, photorealistic, 85mm lens look.
[PHOTOGRAPHIC & ENVIRONMENTAL REALISM — CRITICAL] The person must appear PHYSICALLY INSIDE the scene, not composited on top of it.
The user provided the style direction 'londra sokağı akşamüstü' (London street, late afternoon). The system accepted this as a valid scene descriptor and the Vision LLM expanded it into a detailed cinematic prompt. Hybrid Mode was activated: the background and lighting were fully re-rendered to match a golden-hour London street while the person's identity and the sweater's cable-knit texture were preserved. Validated on the first attempt with a perfect score.