ControlNet — 12 ngā momo āhuatanga i roto i tētahi utauta

Whakataki i tētahi whakaahua tohutoro, kōwhiri i tētahi momo āhuatanga, tuhi i tētahi whakahau. Ka pupuri te AI i tōna hanganga tohutoro (rārangi, tūnga, hōhonu, ērā atu mea) ā, ka whakarārangi i ngā ihirangi hōu i roto i tētahi kāhua. I tautokona e ControlNet-Union SDXL ProMax — Apache 2.0, he mātauranga-whakahaere.

Canny / paerangi mō te mahi raina mārama. He tūnga mō te tūnga tinana. Rāwhiti mō te hoahoa 3D. Scribble / te pito mārō mō ngā tātuhi. MLSD mō ngā hoahoa. Pāhono / wehenga / tāhae mō ngā rerenga mahi hōhonu.
Ka tangohia te āhuatanga mai i tēnei — ka tangohia ngā tae, ko te tohu hanganga anake (e ai ki tō tātau momo i kōwhiria) e tiakina.
Ko te kaipāpāho 0.7 He uaua
~1,200 tokens (SDXL × 1.2 ControlNet)
Whakamutunga

He pēhea te mahi a ControlNet

E āhei ana a ControlNet ki te whakahaere i te whakawhanaketanga o te whakaahua me te hanganga o tētahi whakaahua tohutoro i te mea e whakawhirinaki ana ki te whakahautuhi kupu anake. Ka pānui tētahi kaikōrero ki tōna tohutoro, ā, ka tangohia tētahi tohu whakawhanake kotahi — ōna taitapa, ōna mahere hōhonu, te āhua o te tangata, me ētahi atu mea. Kātahi ka katia te tauira whakawhānuitanga ki taua tohu i te wā e whakatau ana te whakahauhau i te kāhua, ngā tae, te mārama, me te kaupapa. Ka pupuri te hua i te hanganga tika i whakawāteatia ai e koe, engari e āhua nei he mea hou katoa.

E tautokona ana tēnei utauta e ControlNet-UnionSDXL ProMax (Apache 2.0) — he tauira kotahi e mōhio ana ki ngā momo āhuatanga 12 katoa i raro nei, kia whakawhitia ai koe i waenganui i a rātou mai i tētahi kōwhiringa kotahi me te kore te whakataki i tētahi whatunga rerekē i ia wā. He tino ngāwari te whakamahinga hokohoko: pupuri, hoko, whakarerekē rānei i ngā mea e waihanga ana.

Ko nga momo whakaritenga 12

Canny
Ko te kite i te pito mārō. Ko te pai rawa mō te pupuri i ngā aratau mārō me te rārangi mārō.
Rāwhiti
3D te mahere hōhonu. E pupuri ana i te whakaritenga mokowā — he aha te tata, he aha te tawhiti.
Whakahau
Ko te āhua o te tinana OpenPose. Ka kati i te tūnga me ngā tūnga o te āhua.
Scribble
I huri nga tātuhi tātuhi-ringa ki roto i te toi oti.
Wehenga
He mahere rohe whakahua-rangi. Whakawhiwhia ia rohe o te tirohanga ki tētahi karaehe.
Tūturu
He mahere mata-tōna. E pupuri ana i te anga mata 3D me ngā tūpāpaku.
Karika paerangi
Ko te whakawāteatanga raina mārō—he tino pai mō te tā, te manga, me te whakaahuatanga.
Te pito mārō
He pūkete taitapa māmā e whai ana i ngā hanga i te āhua o Canny.
MLSD
Ko ngā wāhanga raina hāngai. I hangaia mō ngā hoahoa, ngā āhua ā-roto, me ngā tāruatanga hua.
Tile
He āhuatanga whakamātautau mō te whakawhānuitanga me te mahi āhuahira kāore i te āhua.
Pa_ki
Ko te āhua o te mārama-mārama hei whakahōu anō anake i tētahi wāhanga o te whakaahua.
Whakapūtā / whakatūtū
Whakaroatia tētahi kanohi, rohe whakairo rānei i te whakawhiwhinga ki te hanganga huri noa.

E toru nga hipanga

  1. Whakataki i tētahi whakaahua tohutoro — he whakaahua, he tātuhi, he whakaahua, he mea rānei me te hanganga e hiahiatia ana e koe kia pupuri.
  2. Ka kōwhiria te momo āhuatanga e ōrite ana ki te mea e hiahia ana koe (pūnaha mō tētahi tā, hōhonu mō tētahi tirohanga, mārama, paerangi rānei mō ngā tātai mārama).
  3. E tuhi ana i tētahi whakahau e whakaahua ana i te āhua e hiahiatia ana e koe, e waihanga ana hoki. E whakanui ana i te kaha o te whakahaere hei whai i te tohutoro, ka whakaitihia mō te wāteatanga ā-hau.

ControlNet — 12 ngā momo āhuatanga i roto i tētahi utauta — FAQ

He utauta kotahi e whakaatu ana i ngā momo āhuatanga katoa e 12 mai i te tauira ControlNet-Union SDXL ProMax — canny, pose, depth, scribble, lineart, anime-lineart, MLSD, HED, soft-edge, normal, segmentation, me te tile. Ka kōwhiria he momo āhuatanga, ka whakawātea i tētahi whakaahua tohutoro, ka tuhituhi i tētahi whakahau, ā, ka whakawātea e SDXL i tētahi whakaahua hou e whai ana i te hanganga o tō tātou tohutoro.

Ka whakairo anō te img2img i runga i te tāuru — ngā tae, ngā pito, me te āhuahanga e whakakotahi ana ki te whakahau. Ka whakawātea te ControlNet i ngā tae, ā, ka pupuri anake i te tohu hanganga i kōwhiria (ngā raina, te āhua o te āhua, te mahere hōhonu, ērā atu mea). Mā tēnei e taea ai e koe te whakawhiti i ngā ihirangi i te wā e pupuri ana i te hanganga ki te pōhatu mārō. He tino kaha ake te whakahaere hanganga i te img2img.

Canny / paerangi mō te tāuru rārangi mārama. Anime-lineart mō te tāuru rārangi āhua anime. Scribble / Soft-edge / HED mō ngā tātuhi me ngā tātuhi. He āhua hei tārua i tētahi tūnga tinana mai i tētahi whakaahua. Rāwhiti ki te pupuri i te āhuahanga mata / 3D. MLSD ki te pupuri i ngā raina hāngai (hanganga āhuahanga / roto). Tūturu ki te pupuri i te aronga mata me te rōrahi. Wehenga ki te pupuri i ngā rohe. Tile ki te whakamārama, ki te whakawhānui rānei i ngā rerekētanga o tētahi whakaahua tīariari.

ControlNet-Union SDXL ProMax (xinsir, Apache 2.0) e whakatū ana i ngā whatunga āhuahira 12 katoa ki roto i tētahi taumahatanga 2.5 GB kotahi. Ko ngā whakawhānuitanga tahito i whakawāteatia ai i tētahi taumahatanga ~2.5 GB ia momo — ko te whakawhitinga i waenganui i te pū me te āhua o te tīmatanga mātao. Ko te tauira āhuahira kotahi e whakarewa ana, ā, ka noho wera, nō reira ko ia momo āhuahira he tuarua i muri i te whakarongo tuatahi.

He. ~1,200 ngā tohu i ia rā (1,000 pūtake SDXL + 20% ControlNet whakatūtū). Ka whiwhi ngā kaiwhakaari whakatūtū i ngā tohu wātea 30,000 i ia rā — tata ki ngā whakatūtū ā-wāhanga 25 i ia rā, kāore i te utu. Kāore i te mōhiotia: 2,500 ngā tohu/ rā (~2 ngā whakatūtū).

He — ko te whakarerekētanga kaha o te Mana (0.7 tōmua) e whakatau ana i te āhua o te huaputa e whai ana i tōna tohutoro. 1.0 = tino āhua (hei whakarerekētanga o tōna tohutoro). 0.4 = whakawātea (he nui ake te wātea o te whakahauhau). Whakaiti i tēnei mō te rerekētanga auau, whakatū i tēnei ina whai take te pono.

512×512 te pūmau. Ko ngā ōwehenga paerewa SDXL — 768×1024 te whakaahua, 1024×768 te whenua, 1024×1024 tapawhā — he mahi katoa. He nui ake te whakapaunga o ngā huaputa nui ake i te VRAM me ngā tohu; E tautoko ana te H200 tae atu ki te 1024×1024.

Ka tuwhera ngā whakaahua tohutoro i te wā kotahi, ka tangohia te āhuatanga, kātahi ka tangohia te pūkete tohutoro. Ko te whakahau + te whakarārangi whakamutunga anake e noho ana i runga i te /kōrero/?tab=hitori. Kāore i whakamahia mō te whakaakoranga. /privacy/ mō te kaupapahere katoa.

ControlNet-Union SDXL ProMax kua tukua i raro i te Apache 2.0 — he tino whakaaetanga, tae atu ki te whakamahinga hokohoko. Ko te pūtake SDXL ko OpenRAIL++. E whakaaetia ana e rāua ngā whakamahinga hokohoko; ko ngā whakaahua i hangaia e koe kia whakamahia ai te hokohoko, kāore i te utu.

He tauira ōrite, he āhuatanga ōrite, he tohu āhuahira ōrite. E hiahiatia ana e ComfyUI me te A1111 he GPU paetata me te 12+ GB VRAM me te whakaritenga. Ka whakahaeretia e tātau i runga i tētahi hanganga tiritiri me tētahi pūrere wātea - kāore he whakatūnga, kāore he GPU e hiahiatia ana.

Ka whakahuatia e te whakarongo tuatahi te taumaha o te Union (~2.5 GB) ki roto i te pūtahi GPU, ā, ka wera te pūhui SDXL. E tūmanako ana 30-60 sekone i runga i te tono tuatahi i muri i te whakawhānui, i te whakawāteatanga rānei o LRU. Ko ngā whakarongo whai ake i raro i te taumahatanga pūnoa e hoki ana i roto i te 4-7 sekone.

He — POST te wāhanga maha ki /v1/image/generate/ me te tauira=sdxl (tauira=controlnet-union-sdxl-promax rānei), whakahau, whakahaere_iranga (reo), whakahaere_āhua= <kotahi o: canny, pose, depth, scribble, lineart, anime-lineart, mlsd, hed, soft-edge, normal, segmentation, tile>, mana mana ā-pāpāho (0.1-1.5). Whakatika, 10K ngā tohu wātea/whā. He tauira kōaro te /api/.

E hiahia ana ki te Free.ai? Whakapāpāho ki ōna hoa!

Whakawhiwhia tēnei pātū