ControlNet — 12 kutonhora marudzi muimwe chirongwa
Upload a reference image, pick a conditioning type, write a prompt. The AI keeps your reference's structure (lines, pose, depth, etc.) and renders new content in any style. Backed by ControlNet-Union SDXL ProMax — Apache 2.0, fully commercial-use friendly.
Chikamu
ControlNet inoita sei basa rayo
ControlNet inokutendera kuti uite kuumbwa kwechifananidzo nechimiro chemufananidzo wemufananidzo, kwete kuvimba nebhokisi rekuti "text prompt" chete. Preprocessor inoverenga mufananidzo wako uye inotora imwe chete yekuzvibata-mufananidzo yemufananidzo — mativi ayo, mapikicha ekumashure, skeleton yemunhu, nezvimwe. Iyo diffusion model inovharwa kune iyo signal, uye bhodhi rinovaudza nezve style, colors, lighting, uye subject. Iyo result inochengeta iyo exact composition yawakaisa asi inoonekwa sechinhu chakafanana nechinhu chazvino.
Iyi chirongwa inotsigirwa ne ControlNet-UnionSDXL ProMax (Apache 2.0) — imwe chete model iyo inonzwisiswa nezvose 12 zvinodiwa zvinyorwa pazasi, saka unogona kuchinja pakati pavo kubva kune mumwe picker pasina kurodha imwe network nguva dzose.
The 12 kutonhora mhando
Canny
Chiedza chekuongorora. Inonyanya kushandiswa pakuchengeta mazita ezvinyorwa uye kubvisa mavara.
Kudzika
3D dept map. Inochengeta spatial layout — chii chiri pedyo uye chii chiri kure.
Kubva Kwakanaka
OpenPose skeleton yemuviri. Inovhara pfungwa yechifananidzo uye nzvimbo dzemativi.
Scribble
Loose-hand-drawn doodles yakashandurwa kuita yakati wandei art.
Kuparadzanisa
Mapeji enzvimbo anozivikanwa nevara. Gadzirisa chero nzvimbo yemufananidzo kune chikwata.
_Zvakajairika
Surface-normal map. Inochengeta 3D yepasi pevhu uye mabhureki.
Kuumbwa kwemitsara
Fine line kuwedzera - yakanakisisa ye ink, manga, uye kuratidzwa.
Soft-edge
Gentle boundary detection iyo inotevera mafomu zvakasununguka kupfuura Canny.
MLSD
Yakagadzirwa yechinyakare, yekunze, uye yezvigadzirwa zvidimbu.
Tile
Detail-kuchengeta kutonhora kweupscaling uye seamless texture basa.
Inpaint
Mask-aware conditioning kuti igadzirise chete chikamu cheiyo image.
Repaint / outpaint
Kuwedzera canvas kana repaint nzvimbo pamwe kutevedzera zvakaita nharaunda.
Mamwe matanho
- Upload a reference image — a photo, a sketch, a screenshot, chero chinhu nechimiro iwe uchida kuchengeta.
- Choose conditioning type iyo inosangana nezvaunoita (pose for a figure, depth for a scene, canny or lineart for clean outlines).
- Kuwedzera kudzora simba kuti utevedzere chirevo zvakasimba, kuderedza kudzora simba kuti utevedzere chirevo zvakasimba, kuderedza kudzora simba kuti utevedzere chirevo zvakasimba, kuderedza kudzora simba kuti utevedzere chirevo zvakasimba.
ControlNet — 12 kutonhora marudzi muimwe chirongwa — FAQ
A imwe chirongwa chinopa zvese 12 kugadzirira marudzi kubva ControlNet-Union SDXL ProMax mufananidzo — canny, pose, denderedzwa, scribble, lineart, anime-lineart, MLSD, HED, soft-edge, normal, segmentation, uye tile. Pick a kugadzirira rudzi, kudonha a reference vhidhiyo, nyora a prompt, uye SDXL rinopa nyowani vhidhiyo iyo inotevera muviri wako reference.
img2img inopinzazve zvinyorwa zvakananga - mavara, mativi, uye mativi ekuumbwa pamwe chete nechikumbiro. ControlNet inobvisa mavara uye inochengeta chete chimiro chakasarudzwa (mativi, pose skeleton, depth map, nezvimwewo). Izvi zvinokutendera kuti udzoreredze zvakanyorwa uye ugare nechimiro chakasimba. Kudzora kwechimiro kuri kuoma kupfuura kweimg2img.
Canny / lineart yekuchenesa kugadzira mazita. Anime-lineart yekunyora mazita ane hunhu hweanime. Scribble / soft-edge / HED yekugadzira zvinyorwa zvakaoma. Pose kuti uite kopi yenzvimbo yemuviri kubva pafoto. Depth kuti uchengetedze geometry yemufananidzo / 3D layout. MLSD kuti uchengetedze mazita akasviba (chivakwa / mukati). Normal kuti uchengetedze kutenderera kwenzvimbo uye huwandu. Segmentation kuti uchengetedze nzvimbo. Tile kuti ugadzirise kana kusimudzira zviyero zveiyo yazvino vhidhiyo.
ControlNet-Union SDXL ProMax (xinsir, Apache 2.0) inoisa zvese 12 zvinodiwa zvema network muimwe 2.5 GB yakakura. Mamwe maapplication akarodha pasi imwe ~2.5 GB yakakura yemhando imwe neimwe — kuchinja pakati pecanny nepose kwakareva kupisa-kutanga.
Yeah. ~1,200 tokens per render (1,000 base SDXL + 20% ControlNet conditioning surcharge). Vashandisi vakanyoresa vanowana 30,000 ma Tokens emahara pazuva — pamusoro pe25 conditioned renders pazuva pasina mutengo. Anonymus: 2,500 Tokens/zuva (~2 renders).
Yeah — the Control strength slider (default 0.7) inoratidza kuti sei output inotevera yako reference. 1.0 = strict (output inoonekwa seyakadzokera-kuratidzwa kweiyo reference). 0.4 = loose (prompt ine zvakawanda zvekuzvitonga). Lower it for creative variation, raise it when fidelity matters.
SDXL standard ratios — 768×1024 portrait, 1024×768 landscape, 1024×1024 square — all work. Larger outputs consume more VRAM and tokens; the H200 supports up to 1024×1024 comfortably.
Referenzimages zvinogadziriswazve, kukonzeresa kunobuda, uyezve referenzfile rinogadziriswa. Kunyangwe chete kubvunza + final render inoramba pa /account/?tab=history. Hapana kushandiswa kwekudzidzisa. /privacy/ kune yakazara mitemo.
ControlNet-Union SDXL ProMax inoburitswa pasi peApache 2.0 — yakasununguka, kusanganisira kushandiswa kwekutengesa. SDXL base ndeye OpenRAIL++. Imwe neimwe inobvumidza kushandiswa kwekutengesa; mapikicha ako akagadzirwa ndeako kuti uwane mari yekushandisa pasina mari yemubhadharo.
Same model, same quality, same conditioning signals. ComfyUI uye A1111 zvinoda yemuno GPU ine 12+ GB VRAM plus setup. Tichazvishandisa pane yakabatanidzwa nharaunda ine yakazara yemahara pool — hapana kuisa, hapana GPU inodiwa.
Kutanga kufona kunorodha pasi muviri weUnion (~ 2.5 GB) muGPU cache uye kunopisa SDXL pipeline. Kutarisira 30-60 masekondi pakutanga kwekutanga chikumbiro mushure mekuisa kana LRU kubvisa.
Yeah — POST multipart to /v1/image/generate/ with model=sdxl (or model=controlnet-union-sdxl-promax), prompt, control_image (file), control_type=<one of: canny, pose, depth, scribble, lineart, anime-lineart, mlsd, hed, soft-edge, normal, segmentation, tile>, optional control_strength (0.1-1.5). Bearer auth, 10K free tokens/month. /api/ has curl examples.