Video darslik · Seedance 2.5
Siz videoni bir marta, bitta kameraga olasiz. Keyin sun'iy intellekt o'sha sahnani boshqa joydan «suratga oladi»: yon tomondan, orqadan, boshning tepasidan, hatto stoldagi chelak yoki stakan ichidan. Qahramonning harakati, yuzi va stoldagi narsalar o'zgarmaydi — faqat kamera ko'chadi. Quyida biz o'zimiz qilgan ishning to'liq yo'li: ketma-ket, tayyor promptlar bilan.
Haqiqatda faqat shu kamera bor: qahramon uning obyektiviga gapiradi. Qolganlarini AI chizadi.
AI bilmaydigan odam uchun ham hammasi shu: video yuklaysiz, matn yozasiz, natijani montajda kesasiz.
| Tushuncha | Ma'nosi | Nega muhim |
|---|---|---|
@video1, @image1 | Promptda yuklangan faylga ishora. Raqam yuklash tartibi bilan beriladi. | Har bir @ nima uchun ekani yozilmasa, model o'zi tanlaydi. |
| Lock (qulf) | «Bu o'zgarmaydi» degan qat'iy qoida: kamera, qahramonning kadrdagi o'lchami, stoldagi narsalar, vaqt. | Qulflanmagan narsani model o'zicha o'zgartiradi. |
| Motion reference | Video faqat harakat, vaqt va ovoz manbai. Uning kamera joyi va kompozitsiyasi olinmaydi. | Rakurs almashtirishning asosi: harakat eski, ko'rinish yangi. |
| Edit va reference rejimi | Edit asl videoni tahrirlaydi. Reference videoni noldan yaratadi. | Kichik tuzatish — edit. Yangi rakurs yoki yangi odam — reference. |
| Ekran koordinatasi | «Ekranning chap tomoni» — tomoshabin ko'rgandek, qahramonning chap-o'ngi emas. | Kamera orqaga yoki tepaga o'tganda chap va o'ng almashadi. Ekran koordinatasi chalkashlikni yo'qotadi. |
Avval qaysi joyni boshqa rakursga almashtirishni aniqlang. Uchta narsani toping:
Keyin videoni kadrma-kadr ko'rib chiqing. Taxmin qilmang, hammasini o'lchang:
Yangi rakursni AI chizadi, lekin ovozni siz originaldan olasiz. Qahramon gapirayotgan paytda lablari ovozga tushmaydi. Jim turgan joyida esa hech qanday moslik talab qilinmaydi — shuning uchun vrezkani pauzaga qo'ying. Orqadan yoki chelak ichidan olingan rakursda og'iz ko'rinmaydi, u yerda bu muammo umuman yo'q.
Referens — bu AI qarab oladigan namuna. Ularsiz model boshqa odamni va boshqa joyni o'ylab topadi.
@video1 deb ataladi.@image1.@image2.Kadrlarni videodan olasiz: montaj dasturida kerakli joyda to'xtatib, «export frame» yoki «kadrni saqlash» tugmasini bosasiz. Fayllarni promptdagi tartibda yuklang.
Stoldagi narsalar faqat videodan olinadi. BTS rasmda narsalar boshqa joyda turgan bo'lishi mumkin — bizda stakan karuselda edi, videoda esa stolda. Hakam doim video.
Referens rasm o'z rakursini ham «o'rgatadi»: model undagi kamera joyini nusxalab qo'yishi mumkin. Suratda shtativ, kamera yoki suratga olayotgan odamlar bo'lsa, AI ularni kadrga chizib qo'yadi. Shuning uchun ikki yo'l bor: atrofni rasm o'rniga matn bilan tasvirlang, yoki rasmni bersangiz, promptda uning kamera joyi va texnikasini aniq taqiqlang.
Har bir yangi joydan kadrga boshqa narsa kiradi va boshqa narsa chiqib ketadi. Promptni yozishdan oldin shuni aniq tasavvur qiling: kamera haqiqatan o'sha yerda tursa, nima ko'rinardi.
| Rakurs | Kamera joyi | Kadrda nima bor | Asosiy e'tibor |
|---|---|---|---|
| Chap 90° | Stol balandligida, 4 m yonda | Profil, fonda yo'l va daraxtlar | Kamera yonda qoladi, orqaga o'tmaydi |
| O'ng 90° | Qarama-qarshi yonda, 4 m | Narigi profil, terrasa, o'rindiqlar | Fudtrak kadrga tushmasligi yoziladi |
| Orqadan | Fudtrak tomi, ~3 m balandlik, 4–5 m orqada, 30° pastga | Qahramonning orqasi, stol ustidagi narsalar orqa tomondan, oldida maysa va ko'cha | Qo'l harakati orqadan qanday ko'rinishi alohida yoziladi |
| Tepadan | Boshdan ~2 m yuqori, tik pastga | Boshning tepasi, stol usti, plitalar, maysa | Kadr tepasi = qahramonning orqasi, pasti = stol: shunda tepaga qaragan yuz to'g'ri ko'rinadi |
| Chelak ichidan | Chelak tubi, tik tepaga | Chelak devori, daraxt shoxlari, egilganda yuz | Kamera chelak bilan birga qimirlaydi |
| Stakan ichidan | Stakan tubi, tik tepaga | Oq ichki devor, shoxlar, shisha uchi, suv oqimi | Qahramonning yuzi ko'rinmaydi — faqat shisha uchi va suv |
Prompt — bloklarga bo'lingan texnik topshiriq. Har bir blok bitta savolga javob beradi:
| Blok | Nimani hal qiladi |
|---|---|
| FORMAT | Davomiylik, nisbat, umumiy ko'rinish |
| REFERENCES | Qaysi fayl nima uchun ishlatiladi; @video1 ning kamera joyi olinmasligi |
| SUBJECT | Qahramon va stoldagi har bir narsa — yangi burchakdan qanday ko'rinishi bilan |
| CAMERA | Kamera qayerda turadi: balandlik metrda, masofa, burchak gradusda; kadrning tepasi va pasti qayoqqa qaraydi |
| SCALE | Qahramon kadrda qancha joy egallaydi va qayerda turadi |
| GAZE | Ko'zlar qayerga qaraydi |
| ACTION | Soniyalab nima sodir bo'ladi — 1-qadamdagi jadvaldan |
| BACKGROUND | Orqa fonda nima bor — BTS rasmdan, yaqindan uzoqqa qarab |
| LIGHTING | Yorug'lik va rang originalga mos kelishi |
| AUDIO | Asl ovoz tegilmaydi yoki faqat fon shovqini |
| CONSTRAINTS | Nima bo'lmasligi kerak; barcha qulflar bir joyda takrorlanadi |
Promptni ingliz tilida yozing — model uni aniqroq tushunadi. Uzunligini 2000–3000 belgida saqlang: bizda 8000 belgilik prompt «failed» bergan. Quyidagi uchta prompt tayyor: nusxa oling va faqat qahramon bilan fonni o'zingiznikiga moslang.
FORMAT Single continuous shot, 16:9, 5 seconds, photoreal documentary look. REFERENCES @video1 is the reference for the man's identity and face, his wardrobe, the table, the lighting and the color. Ignore its camera position and framing. @image1 is a photo of the same place, used only as reference for the surroundings. Do not reproduce its camera position or framing, and do not include the tripod, the camera or any crew. SUBJECT The same young man from @video1: pink t-shirt, dark hair, black wristwatch, a ring, a small lavalier mic clipped to his shirt. He sits at the small round glass table. On the table, exactly as in @video1: the dark laptop open in front of him, a branded paper bucket, a black cup and a water bottle. CAMERA AND GEOMETRY Pure side view, exactly 90 degrees from the direction his body faces. The camera stands level with the table, 4 meters away, at his seated eye level, holding still with a very slight handheld drift. In frame: he is seen in full profile, his shoulder line running across the frame, his nose and chin pointing toward the edge of the frame, his ear and cheek facing the camera. Shallow depth of field: his face sharp, the background soft. GAZE A person is standing exactly where the camera is, and he is talking to that person. His eyes meet the lens directly: the pupils sit in the middle of his eyes, centered on the lens, not turned toward the corners. 0.0s-1.0s - he is in profile, looking ahead at the laptop. 1.0s-1.6s - he turns his head toward the camera and his eyes land on the lens. 1.6s-end - he holds eye contact with the lens while he talks, his head in three-quarter view, with only small natural blinks and head movements. ACTION He talks in that direction, relaxed and friendly, one hand resting on the table beside the laptop, the other making small gestures near his chest, a short nod and a light smile. BACKGROUND Behind him: the paved terrace and grass, then the road with parked cars and tall green trees above, soft and bright in the sun. LIGHTING AND LOOK Bright natural daylight from above through foliage, dappled highlights on his shoulder and the table, warm summer color, the same camera grain, softness and exposure as @video1. AUDIO Quiet outdoor ambience only: distant street sounds, soft wind. No dialogue. CONSTRAINTS - The camera is beside him, never behind him: the back of his head is never toward the lens, his face stays visible in every frame. - The open dark laptop stays on the table in front of him in every frame. - The man's identity, face, hair, t-shirt, wristwatch, ring and lavalier mic stay exactly as in @video1. - No camera, no tripod, no lights, no crew and no filming equipment anywhere. - Photoreal, one continuous shot, no on-screen text.
FORMAT Single continuous shot, 16:9, 5 seconds, photoreal documentary look. REFERENCES @video1 is the reference for the man's identity and face, his wardrobe, the table, the lighting and the color. Ignore its camera position and framing. SUBJECT The same young man from @video1: pink t-shirt, dark hair, black wristwatch, a ring, a small lavalier mic clipped to his shirt, sitting at the small round glass table with the dark laptop open in front of him, a branded paper bucket, a black cup and a water bottle. CAMERA AND GEOMETRY Pure side view, exactly 90 degrees from the direction his body faces, from the opposite side. The camera stands level with the table, 4 meters away, at his seated eye level, holding still with a very slight handheld drift. In frame: he is seen in full profile in the right half of the frame, his nose and chin pointing toward the left edge, his ear and cheek facing the camera, the whole outline of his profile clearly readable against the background. GAZE He looks ahead in the direction his body faces, toward the left edge of the frame, talking to a person standing there. Twice during the shot he glances toward the camera, his pupils landing centered on the lens for about half a second each time, then returns to the person he is talking to. ACTION He talks, relaxed, one hand resting on the table beside the laptop, the other rising briefly in a small gesture and settling back, a light shrug of the shoulder, then he leans back slightly in the chair. BACKGROUND Behind him: the open paved terrace with white garden chairs, grass, tall green trees and, further away, a road with parked cars, all soft and bright. LIGHTING AND LOOK Bright natural daylight from above through foliage, dappled highlights on his shoulder and the table, warm summer color, the same camera grain, softness and exposure as @video1. AUDIO Quiet outdoor ambience only: distant street sounds, soft wind. No dialogue. CONSTRAINTS - The camera is beside him, never behind him: the back of his head is never toward the lens, his face stays visible in profile in every frame. - No food truck, no red vehicle and no food-truck signage anywhere in frame. - The open dark laptop stays on the table in front of him in every frame. - The man's identity, face, hair, t-shirt, wristwatch, ring and lavalier mic stay exactly as in @video1. - No camera, no tripod, no lights, no crew and no filming equipment anywhere. - Photoreal, one continuous shot, no on-screen text.
Top-down overhead shot. The same 8.9-second moment as @video1, filmed from a static camera 2 meters directly above the young man's head, lens pointing straight down. Only the camera position changes. His face, expressions, mouth movements, hand and head movements with their exact timing, his clothes, the table and objects, the light and the audio are 100% identical to @video1. CAMERA: completely static top-down view for the whole shot. Never moves, tilts, rotates, zooms or cuts. Never a front or side view. The top of the frame is behind him (his chair); the bottom of the frame is in front of him (the table). When he looks up, his face appears upright and looks straight into the lens. FRAME: his head and shoulders in the centre, his head about one-fifth of the frame height. The table in the lower-middle part. Ground visible all around. TABLE, 1:1 from @video1, seen from above: grey-teal square wooden top. Dark laptop in front of his chest, lid open toward him. Paper chicken bucket at the far left corner, seen as a round open top. Black paper cup to the right of the laptop, seen as a small round opening. A clear plastic water bottle standing upright at the far right corner: from above we see its round blue screw cap in the centre with the clear rounded shoulders around it - it is a bottle, not a cup. Nothing added, nothing removed, nothing moves. GROUND: small square grey concrete paving stones in mixed grey tones under the table and his light beige wooden cross-back chair, a few dry leaves on them. Beyond the table, green lawn. Dappled leaf shadows from a large tree above. No people, no tripod, no truck. MOTION, exactly @video1: 0.0-1.5s: facing forward, we see the top of his head; he talks, hands gesturing in front of his chest. 1.6-1.8s: he tilts his head back and looks straight up into the lens. 1.8-2.9s: looking into the lens, calm, smiles at 2.8s. 3.0-4.1s: raises his left arm toward the lens, open palm waving, laughing. 4.2-4.6s: points his left finger up, then brings it to his nose. 4.7-5.1s: lowers his left arm. 5.2-6.2s: palms pressed together at his chest, looking up. 6.3-8.9s: fingers clasped at his chest, looking up into the lens, talking. No other movements. Watch on his left wrist appears on screen-right; ring and bracelet appear on screen-left. LIGHT: the same bright summer sun as @video1, soft dappled light. AUDIO: original audio of @video1, unchanged. Photoreal, real skin texture, real fabric, fine natural grain. 16:9, 1080p, 30fps, 8.9 seconds, one continuous shot.
Tepadan promptining birinchi variantida suv shishasi faqat «ko'k qopqoqli» deb yozilgan edi — tepadan u shunchaki doira, model uni ikkinchi stakanga aylantirib qo'ydi. Yuqoridagi versiyada shishaning tepadan ko'rinishi to'liq tasvirlangan.
Kerak bo'lgan qismdan uzunroq plan buyuring: montajda o'zingizga yoqqan bo'lagini olasiz. Tayyor bo'lgach, original va natijani bir xil soniyalarda yonma-yon qo'yib, shu ro'yxat bo'yicha tekshiring:
Yoqqan variantlarni saqlab boring — keyingi urinish yomonroq chiqishi mumkin.
Natijada bitta narsa xato bo'lsa, butun videoni qaytadan yasamang. Ikki yo'l bor:
| Vazifa | Rejim | Natija bizda |
|---|---|---|
| Bitta stakanni suv shishasiga almashtirish | Edit | Ishladi |
| Kamerani boshqa joyga ko'chirish | Reference | Ishladi |
| Videodagi odamning yuzini boshqa odamga almashtirish | Edit | Uch urinishda ham yuz o'zgarmadi |
| Xuddi shu yuz almashtirish | Reference yoki face swap vositasi | Ishlaydigan yo'l |
Bir xil natija ikki marta takrorlansa, ayb promptda emas, rejimda. To'xtang va rejimni almashtiring. Har bir edit passi sifatni biroz pasaytiradi — bir sohaga ketma-ket ko'p pass qilmang.
Qahramon bilan u gapirayotgan odam orasida xayoliy chiziq bor. Yon kamerani shu chiziqning bir tomonida qoldiring. Agar narigi tomonga o'tsangiz, vrezkada u qarama-qarshi tomonga qaraydi va tomoshabin adashadi. Orqadan, tepadan va idish ichidan olingan rakurslar esa alohida «effekt kadri» sifatida ishlatiladi — ularni asosiy plan orasiga qisqa urg'u qilib qo'ying.
| Muammo | Yechim |
|---|---|
| Kadr «baliq ko'zi» bo'lib ketdi | Obyektiv haqidagi gapni yumshating: normal field of view, like a 28 mm lens, straight lines stay straight, no distortion, hamda no fisheye, no circular vignette qo'shing. |
| Kamera orqaga o'tib ketdi | facing away from the camera kabi iboralarni olib tashlang. Buning o'rniga kadrda nima ko'rinishini yozing: yelka chizig'i, burun qaysi tomonga qaragani. |
| Bosh burildi, lekin ko'z obyektivga tushmadi | GAZE blokida pupil so'zini ishlating va qarashni oxirida emas, oldinroq boshlab, uzoq ushlab turing. Kamerani 3 metr o'rniga 4 metrga suring. |
| Kadrga noutbuk yoki stol chizilib qoldi | Agar kamera ekran ichida bo'lishi kerak bo'lsa, promptda «noutbuk» so'zini umuman ishlatmang va taqiqni yozing: no laptop, no desk, no table anywhere in the frame. |
| Stoldagi narsa boshqasiga aylandi | Tepadan yoki pastdan har bir narsaning shaklini yozing: round blue cap with clear shoulders around it — a bottle, not a cup. Bitta so'z bilan atalgan narsa yonidagisiga o'xshatib chiziladi. |
| Kadrda ko'rinmasligi kerak bo'lgan narsa chiqdi | Kamera haqiqatan o'sha joyda tursa nima ko'rinishini tekshiring. Stakan ichidan qahramonning yuzi ko'rinmaydi — uni promptga yozmang. |
| Hodisa noto'g'ri vaqtda boshlandi | Vaqtni siyrak kadrlardan emas, kattalashtirilgan va zich kadrlardan o'lchang. Bizda suv boshlanishi 1,2 deb yozilgan, aslida 1,7 soniya edi. |
| Kamera harakati yo'qoldi | Kamera harakatini alohida blokka chiqaring va soniyalar bilan yozing, keyin CONSTRAINTS ichida takrorlang. |
| Fon boshqacha chiqdi | Har bir rakurs uchun fonni alohida yozing va o'sha tomonda bo'lmasligi kerak bo'lgan narsani taqiqlang. |
| Yuz o'zgarib ketdi | Referenslar sonini oshiring va CONSTRAINTS ichida yuzni qayta biriktiring: the same real person as in @video1, same face, clean-shaven. |
| Generatsiya «failed» berdi | Promptni 3000 belgigacha qisqartiring. Haqiqiy odam yuzi aniq ko'ringan referens rasm ham moderatsiyadan o'tmasligi mumkin. |
Topshirish: asl video, natija, prompt va xato haqida ikki jumla — nima xato chiqdi va qaysi qator sabab bo'ldi.