Создаем видео, где мы танцуем на парковке.

Нужен аккаунт в Syntx.ai (https://syntx.ai/welcome/mEQ2WKuT)
*Подписка PRO и выше.
Нейросети: Grok Imagine 2.0 или Gpt Image 2 для фото, Seedance 2.5 для видео.
Нужно: 3 фото героев ролика, + видео-исходник (ссылка)

Create a photorealistic character reference sheet from the uploaded person photos: three vertical panels, clean white or light-gray background, clear vertical dividers.

RIGHT: Shoulder-cropped frontal portrait, direct gaze, neutral expression, soft even light. Preserve identity: facial proportions, eyes, nose, lips, brows, ears, skin tone and texture, hair color and shape, moles and freckles. Highly detailed, natural realism.

LEFT: Front view, framed from the neck/shoulder line down to the feet; head and face outside the crop. Standing upright, arms at sides, feet shoulder-width apart, wearing the person's everyday clothing. Preserve build, proportions, and outfit details.

MIDDLE: Full-body BACK view, including the head and hair from behind. Same person, build, proportions, and outfit.

Keep identity, clothing, lighting, and photographic style consistent across panels. Match body scale in the left and middle panels. High resolution, realistic anatomy and fabric.

*С остальными фотками тоже самое.

Для начала скачиваем видео-исходник: ССЫЛКА

[Generation Task — New Video, Not Source Editing] Generate a NEW photorealistic video using @video1 as a reference for scene layout, action choreography, emotional performances, camera movement, and overall pacing. Do NOT edit, repaint, or face-swap the original video frames. Do NOT place reference faces onto the original actors' heads or bodies. Build three complete, coherent people from the photographic references and render them naturally within a newly generated scene. Target duration: approximately 20 seconds. Aspect ratio: 9:00 — use the standard vertical 9:16 presentation. [Reference Roles and Priority] @image1 = complete appearance of PERSON_A. @image2 = complete appearance of PERSON_B. @image3 = complete appearance of these PERSON_C. @video1 = reference for: - Location and spatial relationships. - Sequence of events. - Human actions and gestures. - Facial expressions and emotional changes. - Camera positions, handheld movements, turns, and zooms. - Approximate action timing and overall visual rhythm. - Evening lighting and atmosphere. @video1 is NOT a reference for: - Character identities. - Head shapes, facial proportions, or hairstyles. - Body types or gender presentation. - Clothing, hats, or wearable accessories. - Captions, lettering, emojis, logos, or other graphics. If the source actors' anatomy conflicts with the photographic references, prioritize the photographic references. Adapt the choreography rather than distorting the reference person to fit the source actor. [Complete Character Construction] Construct each person from their assigned photograph as a unified head and body, not a face mask. Preserve the reference person's: - Overall head shape and cranial proportions. - Face width and length, forehead, cheekbones, jawline, and chin. - Eye shape and spacing, nose shape, mouth shape, and ears. - Apparent age, gender presentation, and skin tone. - Hairline, hairstyle, hair color, and facial hair if present. - Visible neck proportions, shoulder width, and physique. - Clothing, garment fit, colors, materials, and layers. - Headwear and wearable accessories, where present. - Footwear, where visible. Do not inherit the original actors' skull shapes, jawlines, body silhouettes, or characteristic facial features. Maintain natural head-to-body proportions. Do not stretch, compress, or resize reference facial features to match an original actor's silhouette. For details not visible in a photograph, use a restrained, plausible continuation consistent with the visible reference. Do not invent distinctive accessories. Do not copy a photograph's background, pose, lighting, or frozen expression. [Character Assignments] PERSON_A, from @image1: The opening performer on the parking lot, corresponding to the source person originally wearing white. PERSON_B, from @image2: The camera operator, visible during BOTH selfie sections. PERSON_C, from @image3: The person who exits the grey car and walks forward. These assignments never change. No identity blending or exchanged outfits. [Environment and Visual Style] Recreate the recognizable setting of @video1: an outdoor roadside parking area near a service station, asphalt ground, trucks and parked cars, a grey car beside a dark SUV, service-station structures, and evening sky. Match the general spatial arrangement so the camera turns and character actions remain coherent. Render a fresh, plausible environment rather than copying source frames. Use plain, unlettered surfaces wherever the source contains signs, captions, logos, or readable text. Casual handheld smartphone footage. Natural evening exposure, realistic skin and fabric, ordinary phone-camera sharpness, mild noise, motion blur during rapid turns, and believable automatic exposure adjustment. No polished commercial look or new cinematic camera moves. [Performance Transfer] Use @video1 to understand the actions and emotions, then have the reference people perform those actions themselves. Transfer the meaning and timing of gestures, head turns, smiles, expressive mouth movements, and reactions. Preserve each reference person's own facial anatomy during every expression. Do not trace the source facial landmarks onto a differently shaped reference face. Do not use the original actors' heads or bodies as fixed templates. Adapt steps, arm reach, balance, and car clearance naturally to the replacement people's physiques. Maintain believable contact with the ground and objects. [Timeline — Approximate Source-Inspired Timing] [00:00–00:06] PERSON_A — OPENING PERFORMANCE Film PERSON_A, built entirely from @image1, on the parking lot. Reproduce the source opening performance: expressive singing-like or shouting-like mouth movements, animated arm gestures, rhythmic body movements, steps, head turns, and emotional intensity. PERSON_A wears the clothing and accessories from @image1, not the source performer's white outfit unless the photograph independently shows the same outfit. Follow the source camera distance, angle, framing, and handheld movement. Keep PERSON_A's characteristic head shape and facial proportions recognizable throughout the performance. [00:06–00:08] PERSON_B — FIRST SELFIE Follow the source camera turn into a close selfie view of the operator. The operator is PERSON_B from @image2, with their own reference face, skull shape, hairstyle, physique, outfit, and accessories. Reproduce the source smile, expressive mouth movements, head motion, and emotional energy without inheriting the original operator's facial anatomy. Use believable close-range phone-lens perspective. Do not add the original bucket hat or black T-shirt unless they are present in @image2. [00:08–00:17] CAMERA TURN AND PERSON_C EXITING THE CAR Reproduce the source's rapid turn toward the parked vehicles and its change in framing or zoom. Show the grey car and nearby dark SUV in a coherent parking layout. PERSON_C, built entirely from @image3, exits the grey car, turns, smiles, and walks forward in the same general sequence and rhythm as the source. Use PERSON_C's reference clothing, headwear, accessories, physique, hairstyle, and facial structure. Adapt the exit naturally to PERSON_C's body dimensions. Show believable footing and hand contact; the person and clothing must clear the door and roof without clipping. Keep the reference identity stable as the face changes angle and moves closer. [00:17–00:20] PERSON_B — FINAL SELFIE Follow the source camera movement back to the operator. Show exactly the same PERSON_B from the first selfie: identical head shape, facial identity, hair, outfit, and accessories. Reproduce the source's final expressive reaction, smile, mouth movements, and handheld framing. End with the same general pacing as the reference. [Population] Only the three designated people may appear. Do not recreate source background pedestrians, bystanders, passengers, or crowds. Keep background walkways and parking spaces unoccupied by additional people. No extra people in windows, reflections, or vehicle interiors. [Strict Text-Free Rendering] Generate clean imagery with NO text at any point. No captions, subtitles, title cards, speech bubbles, emojis, stickers, watermarks, interface elements, readable signs, license-plate characters, or clothing lettering. The source video's text and graphic overlays are excluded from the reference interpretation. They must never appear in the generated scene. Do not create a panel, blurred strip, blank caption box, or other substitute where source text appeared. Render the actual scene across the entire frame. Preserve reference garment colors, materials, and non-text patterns, but omit lettering and typographic logos. [Continuity and Quality] Keep all three identities separate and consistent. Preserve reference head shapes through profiles, smiles, open-mouth expressions, close-ups, and fast camera turns. Natural skin texture and expressions; no pasted-face appearance. No original actors' facial features, source-body silhouettes, identity drift, doubled features, floating hair, distorted hands, or changing accessories. Follow the source camera choreography and approximate pacing without forcing pixel-level alignment with the original frames. [Audio] Generate only natural location ambience and nonverbal human reactions consistent with the scene. No new intelligible dialogue, generated song, or background music. Leave the clip suitable for synchronization with the original soundtrack during post-production. [Final Priority] 1: Faithful, complete character appearances from @image1, @image2, and @image3. 2: Natural anatomy, coherent bodies, and stable identities. 3: Source-inspired actions, emotions, camera choreography, location, and pacing. 4: Clean text-free imagery with no additional people. A newly generated scene performed by the reference people — NOT the original actors with replacement faces.

ПРОМПТ ДЛЯ ВИДЕО (РЕЖИМ - EDIT):

[Video Editing Task] Edit @video1 using @image1, @image2, and @image3 as COMPLETE CHARACTER APPEARANCE references. Fully replace the three designated people, including their identities, hair, physiques, clothing, headwear, footwear, and wearable accessories. Do not perform face-only replacements. Preserve the exact source duration, approximately 20 seconds, vertical 9:16 aspect ratio, original shot sequence, action timing, camera movements, zooms, transitions, and original audio. [Reference Mapping] PERSON_A = the complete person and outfit from @image1. Replace the opening performer originally wearing a white polo shirt, light jeans, and white sneakers. PERSON_B = the complete person and outfit from @image2. Replace the camera operator in BOTH selfie appearances: the middle selfie section and the final selfie section. PERSON_C = the complete person and outfit from @image3. Replace the person who exits the grey car and walks forward after the camera swings toward the parked vehicles. Keep these assignments consistent throughout the entire video. Never exchange identities, outfits, hairstyles, or accessories between the three references. [Full Appearance Replacement — Highest Priority] For each designated character, use their assigned photograph as the source for: - Recognizable facial identity and facial structure. - Apparent age and gender presentation. - Skin tone. - Visible physique and body proportions. - Hair color, hairline, hairstyle, and hair length. - Facial hair, if present. - All visible clothing, including garment cut, fit, layers, colors, patterns, and materials. - Headwear, glasses, jewelry, watches, and other visible wearable accessories. - Footwear, if visible in the reference. Do NOT retain the original video's clothing, hairstyles, hats, or accessories unless the same items are also present in the assigned reference photograph. If the reference person has no headwear, remove the source character's headwear. If the reference includes headwear, reproduce it. Apply the same rule to glasses, jewelry, and other wearable accessories. Where a clothing or body area is not visible in the reference photograph, use a restrained, plausible continuation consistent with the visible appearance. Do not claim to reproduce unseen details. Do not copy the reference photograph's background, pose, lighting, or static expression. Do not introduce handheld objects from the photographs that would interfere with the original actions. Reference appearance determines WHO the person is and WHAT they wear. The source video determines WHAT they do, HOW they react, and HOW the camera moves. [Movement Transfer] Transfer the original movements to the replacement bodies while preserving the exact action timing, screen positions, movement paths, and interactions. Adapt the motion naturally to each reference person's physique. Preserve foot contact with the ground, hand contact with objects, balance, and plausible joint movement. Clothing and accessories must move naturally with the replacement body. Do not preserve the outline of an original garment beneath a different reference outfit. [Approximately 00:00–00:06 — PERSON_A] Fully replace the opening performer with the person from @image1, including their complete reference outfit and accessories. Remove the original white polo, jeans, and sneakers unless those items are actually present in @image1. Preserve every original step, expressive movement, arm gesture, body lean, head turn, gaze direction, and facial reaction. Transfer the source singing or shouting mouth movements and emotional performance to the replacement face. Do not hold the expression from the reference photograph. Preserve the parking lot, trucks, vehicles, evening sky, camera distance, and framing. [Approximately 00:06–00:08 — PERSON_B, First Selfie] Fully replace the operator with the person from @image2. Replace the original black T-shirt and bucket hat with the actual clothing and headwear from @image2. If @image2 contains no hat, show the reference hairstyle without a hat. Preserve the exact camera rotation toward the operator, handheld perspective, lens distortion, face-to-camera distance, smile, mouth movements, and head motion. Keep the operator's appearance consistent during all partially visible and motion-blurred frames. [Approximately 00:08–00:17 — PERSON_C, Car Exit] Preserve the original rapid camera swing and zoom toward the parked vehicles. Fully replace the person exiting the grey car with the person from @image3, including their reference age, gender presentation, physique, hairstyle, outfit, headwear, and wearable accessories. Do not retain the source person's jacket, lower garments, or hairstyle unless they match @image3. Preserve the exact car-exit choreography, stepping motion, turns, smile, walking direction, and timing. Adapt the replacement body and garments naturally to the doorway and seat. Preserve all source contact points and occlusions involving the car door, roof, window frame, hands, and body. No passing through solid surfaces or clothing clipping through the car. Keep the cars, parking layout, and camera movement unchanged. [Approximately 00:17–End — PERSON_B, Final Selfie] Replace the operator again with the SAME complete appearance from @image2. The face, hairstyle, headwear, clothing, and accessories must match the first selfie section exactly. No return of the original bucket hat, T-shirt, or source identity. Preserve the final expressions, mouth movements, camera shake, framing, and ending. [Timing] The time ranges above are approximate navigation guides only. Follow the actual source transitions and action boundaries exactly. Do not retime the video to fit rounded timestamps. [Remove Other People] Keep only the three designated characters, each appearing where their corresponding source character originally appears. Remove all incidental background people throughout the clip, including the passersby in the final selfie section. Do not remove a designated character if they remain visible in the background of another character's shot. Reconstruct the revealed background naturally. Remove visible reflections and shadows belonging to deleted background people. No ghost figures, blurred silhouettes, or flickering repair patches. [Remove All Text] Remove the existing caption overlay, its backing panel, emojis, and any other added graphics throughout the clip. Reconstruct the scene behind the overlay, including moving clothing or body parts when necessary. Do not hide the overlay with cropping, blur, or another panel. No readable text in the final video. Remove visible lettering from garments, accessories, signs, license plates, and watermarks while preserving their underlying colors, materials, shapes, and non-text patterns. This text-removal requirement is the only exception to exact reference outfit reproduction. Do not generate replacement letters or pseudo-text. [Preserve the Source Video] Keep the locations, vehicles, props, evening atmosphere, and lighting unchanged, except for the required character replacements, background-person removal, and text removal. Preserve all original handheld shake, whip pans, zooms, exposure changes, motion blur, transitions, and playback speed. No added stabilization, camera movements, cuts, slow motion, cropping, or reframing. [Facial and Visual Continuity] Preserve each source character's original expressions, blinking, gaze directions, smiles, and lip movements on the assigned replacement identity. Match the source lighting, shadows, exposure, grain, sharpness, and motion blur. Update body shadows and reflections where needed to match the replacement person and outfit. Maintain stable faces, bodies, garments, and accessories across all frames. Respect all foreground occlusions. No identity mixing, original faces reappearing, wardrobe reversion, duplicated accessories, distorted anatomy, oversized heads, mask edges, or floating clothing. [Audio] Preserve the original soundtrack unchanged and synchronized, including vocals, music, and ambience. Do not change voices to match the replacement appearances. No new speech or additional sounds. [Final Result] The same source video with: 1. The opening performer fully replaced by the person and outfit from @image1. 2. The operator fully replaced by the person and outfit from @image2 in BOTH selfie appearances. 3. The person exiting the grey car fully replaced by the person and outfit from @image3. 4. All incidental people removed. 5. All visible text removed. Complete reference appearances; original source performances, camera work, timing, and sound.