Umkhiqizi we-AI Depth Map
Ukusetshenziswa kwezohwebo kuvunyelwe
Amamodeli angu-588
Awukho umbala wamanzi
Akudingeki ukubhalisa
Imodeli:
+ GPT-5, Claude, Gemini
Yenza i-map yobubanzi kusuka kunoma iyiphi isithombe. Isetshenziswa njenge-ControlNet conditioning input, for paralax / 3D effects, or as a starting point for AR/VR scene reconstruction. Powered by Marigold (newer, sharper) or fal Depth (faster).
Okukhethwa kukho okuthuthukisiwe
Umphumela
Amathokheni ayaphela.
Thola Amathokheni Amaningi
❤️ Uyayithanda i-Free.ai? Tshela abangane bakho!
Bhalisela ukuze uthole isixhumanisi sokuncoma futhi uzuze amathokheni angu-30,000 ngomngani ngamunye.
Ufuna okwengeziwe?
Bhalisela mahhala: amathokheni angu-30,000 ngosuku
Bhalisela Mahhala
Kucutshungulwa isicelo sakho...
Yenza i-map yobubanzi kusuka kunoma iyiphi isithombe nge-AI. I-Marigold emahhala (eqinile, ehlala ngokwayo) noma ububanzi obusheshayo obuphezulu - ikhishwa i-ControlNet-elungele ububanzi PNG ye-paralax, AR/VR, 3D scene reconstruction.
Indlela Yokusebenzisa Umkhiqizi we-AI Depth Map
1
Faka okufakayo kwakho
Thayipha umbhalo, layisha ifayela, noma chaza okufunayo. Akudingeki i-akhawunti.
2
Chofoza ukwenza
I-AI yethu icubungula isicelo sakho ngemizuzwana isebenzisa amamodeli aphambili avulekile.
3
Landa futhi wabelane
Landa, kopisha, noma wabelane ngomphumela wakho. Mahhala ekusetshenzisweni komuntu siqu nokwebhizinisi.
Sebenzisa leli thuluzi nge-API
Yenza le thuluzi ngokuzenzakalelayo ngekhodi yakho. Iphoyinti le-REST elihambisana ne-OpenAI, ukuqinisekisa nge-Bearer-token, akudingeki i-SDK eyengeziwe. Izindleko zama-token zifana nokusebenzisa iwebhu.
curl -X POST https://api.free.ai/v1/image/generate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"prompt": "A sunset over mountains", "model": "sdxl"}'Umkhiqizi we-AI Depth Map - FAQ
Layisha phezulu noma iyiphi isithombe, uthole i-map yobubanzi emuva - isithombe esimnyama lapho i-pixel ngayinye ikhowudisa khona ukuthi ibanga liphi isikhundla esisuka kwikhamera. Sebenzisa njenge-ControlNet input ye-image generation ehlobene nobubanzi, njenge-depth pass ye-paralax / 3D effects ku-After Effects noma ku-Blender, noma njenge-start point ye-AR / VR scene reconstruction. Amamodeli amabili: i-Marigold emahhala (eqinile, ehamba kancane) noma ububanzi obusheshayo obuphezulu (iminingwane encane, ihamba ngokushesha ku-inputs enkulu).
Yebo. I-Marigold isebenza kuma-A100 GPUs ethu ahlala e-A100 - isuka epulazini lakho le-token ngosuku (6,000/day anonymous, 30,000/day signed up). I-map yobubanzi ngayinye i-token ~1,500. I-premium fast-depth ibiza i-baseline efanayo kodwa ingashintsha kahle kuma-4K + ama-inputs.
Marigold (iphutha) inikeza okuqinile, okulungile kakhulu ubunzulu - i-fine-tuned kusuka ku-Stable Diffusion, ithatha ubunzulu obuncane emithini / izinwele / ama-folds endwangu. Ipremiyamu ishesha kakhulu ku >2K ukungeniswa futhi ithembekile kakhulu ku-extreme wide-angle shots. Ukwenza i-ControlNet conditioning noma 3D reconstruction, iphutha ku-Marigold. Ukwenza ama-images amawaka, i-premium scales better.
I-512×512 RGB PNG ene-Spectral colormap - ekhanyayo / eluhlaza okwesibhakabhaka = ezungeze, emnyama / eluhlaza okwesibhakabhaka = enyakatho. Amaxabiso e-intshisekelo emnyama ajwayelekile 0-1 ngaphesheya kwesithombe. Uma ufuna intshisekelo ye-16-bit grayscale (i-ControlNet noma i-Blender ye-maphu yokushintshana), sebenzisa i-API futhi udlulise `format=16bit_png`.
Yebo — i-Marigold output iyifomethi ejwayelekile yokulawula ububanzi-ukungalingani kwe-ControlNet. Sebenzisa ku-Stable Diffusion / FLUX nge-depth ControlNet ukukhombisa kabusha ikhamera eyaziyo: shintsha isiqephu, ukukhangisa, isitayela ngenkathi ugcina isakhiwo se-3D. Free.ai chat ingakuveza wena — sithi "umayelana nobubanzi, bese ubonisa kabusha nge-FLUX ngesitayela se-Studio Ghibli".
I-Marigold iyi-monocular evulekile evulekile ebhekene nomthombo ovulekile (CVPR 2024 iphepha elingcono). Kuphakathi ~3-5% yeqiniso lendawo ezindaweni ezingaphakathi (NYU Depth v2), ngaphakathi 5-8% ngaphandle (KITTI). Ubukhulu be-metric (ama-meter e-real-world), hlanganisa nekhamera ehlobene. Ubukhulu obufanayo (lokhu kuncane kunalokho), ukunemba kulungile.
JPG, PNG, WebP kuze kube yi-10MB. I-Marigold isebenza ku-768×768 ngaphakathi kungakhathaliseki ubukhulu be-input futhi iphinde isampule emuva. I-output ihlala i-512×512 RGB. Ubukhulu obuphezulu be-resolution, sebenzisa ukhetho olusheshayo oluphezulu olulinganiselwa ngokuqondile.
Yebo - ama-maps abanzi yimingeniso ejwayelekile ye-After Effects 2.5D paralax (i-plugin ye-Camera Mapper), i-DaVinci Resolve ebhekene nobubanzi obusekelwe ekuqhekekeni, kanye ne-Blender eguqula imodeli. Yenza ububanzi, ngenisa njenge-alpha-matte sequence, i-keyboard ihamba ngekhamera yakho. Abaphothule abaningi basebenzisa lokhu ukukhangisa i-bokeh ye-portrait-mode ku-smartphones ngaphandle kwe-cameras ezimbili.
I-Marigold (2024) isebenza kahle kakhulu kune-MiDaS v3 (2022) ne-Intel DPT kunoma yisiphi isilinganisi sasekhaya, ikakhulukazi kuma-details angcono (izinwele, ama-foliage, ukubukeka okucacile). I-APIs ekhokhelwa ngokujulile (Replicate, fal) ikhokhisa i- $ 0.005-0.02 ngesithombe - sisebenzisa isisindo esifanayo se-Marigold mahhala. Ukuhluka kwekhwalithi: i-zero.
Yebo - UTHENGA ku /v1/image/depth-map/ ngesithombe esiningi. Noma sebenzisa /batch/ ukususa ifolda bese uthola i-ZIP emuva. Kubalulekile ukuphatha isithombe sevidiyo esigcwele se-frame sequence ukuze kube nemiphumela eyaziwayo yobubanzi.
Yebo - POST ingxenye eminingi `isithombe` ku /v1/isithombe/isithombe-sibalo/ ku api.free.ai. Ubufakazi bomuntu ophethe ngesinqumo sakho se-API. 10,000 amathokheni/inyanga emahhala, ama-levels akhokhelwa phezulu. Ibuyisela i-JSON nge-`image_url` ku-PNG encane. /api/ inesibonelo esincane.
Ukubopha kwi-Photoshop njengendawo yomthombo we-mask, i-ControlNet izici zokubuyisela emuva /isithombe/ukulungisa/, i-map yokushintshana kwi-Blender, i-paralax animator kwi-After Effects, noma i-2.5D dolly-zoom video creator (LeiaPix-style). I-chain ne-/isithombe/ukulungisa/ ukuthuthukisa ububanzi-ukulungisa kwithuluzi elinye.
Bhalisela mahhala: amathokheni angu-30,000 ngosuku
Dala I-akhawunti YamahhalaAkudingeki ikhadi lesikweletu
Ungalilinganisa kanjani leli thuluzi?