Brand UGC Lookbook Video Production Workflow with Sonnet 5.5 and Seedance 2.5
A complete workflow for creating commercial UGC lookbook videos using Claude Sonnet 5.5, GPT Image 2.5 reference sheets, and Seedance 2.5 video synthesis.
AI creator Abdul Shakoor (@abxxai) has shared a full-stack AI workflow for producing high-fidelity brand lookbooks and UGC (User Generated Content) fashion try-on videos without physical model shoots.

Image source: @abxxai on X
This pipeline utilizes Claude Sonnet 5.5 as the primary workflow architect to design the entire scene sequence and prompt package, GPT Image 2.5 to generate five multi-angle visual reference sheets (character, location, garment, start frame, and brand label), and Seedance 2.5 to synthesize realistic vertical smartphone video with multi-reference binding. Audio voiceover is generated via Fish Audio.
1. Workflow Architecture: Orchestration with Sonnet 5.5
Rather than manually crafting prompts step by step, the creator provides Claude Sonnet 5.5 with the target concept and a few reference screenshots to generate the complete production blueprint:
- Scene Progression and Camera Direction: Maps the transition flow from a handheld opening selfie to tripod-mounted try-on footage.
- Cross-Modal Prompt Generation: Produces tailored prompt specifications for image reference plates (character, environment, garment, label) and the final Seedance video.
- Consistency Verification and Error Recovery: Establishes explicit rules for adjusting prompts if a video generation drifts.
2. Generating 5 Key Visual Reference Sheets in GPT Image 2.5
To lock visual identity in Seedance 2.5, five distinct high-resolution reference sheets are generated using GPT Image 2.5.
Character Reference Sheet
Locks facial features (freckles, eye color, balayage hair) and styling across a 4-panel grid (close-up at 18° FOV, 3/4 view at 29° FOV, front full-body at 47° FOV, and back full-body at 47° FOV).
SCENE CONTEXT
A photoreal character reference sheet of one young woman, built from the attached first-frame selfie, for identity lock in a UGC try-on video.
SUBJECT
A young woman in her mid-twenties, slim build. Face: fair skin with a warm golden undertone and a soft capillary flush across the cheeks and nose, dense natural freckles scattered across the nose bridge, both cheeks and the upper lip area, hazel-green almond eyes with a warm golden ring around the pupil, thick natural brows with a fairly straight line and a soft arch at the tail, straight narrow nose with a slight sun-lit sheen on the bridge, full lips in a glossy rose-nude with a defined cupid's bow, dewy bare skin with no visible foundation. Hair: long layered light-brown hair with caramel and honey-blonde balayage streaks, a clean center part, slightly piecey loose waves, soft face-framing strands falling past the collarbones. Hands: short soft almond nails in a sheer nude, two thin stacked silver band rings on the ring finger. Outfit: an oversized burgundy chunky-knit funnel-neck sweater in a fuzzy mohair-blend yarn with a soft halo of fibres, cream relaxed wide-leg trousers, bare feet. 100% matches the reference.
LAYOUT
Four panels side by side, 100% consistent across all 4 shots: face close-up at 18° FOV looking to the lens, head and shoulders three-quarter left at 29° FOV, full body front at 47° FOV, full body back at 47° FOV.
OPTICS
Natural perspective, deep focus, high micro-contrast, fine skin texture preserved.
CAMERA
Eye level, subject centred in each panel, straight rectilinear framing.
BACKGROUND
Seamless neutral light-grey cyclorama, soft contact shadow at the feet.
LIGHTING
Even soft studio softbox key from the front-upper, gentle fill, WB 5200K, freckles and skin texture fully readable.
MATERIAL
Pore-level realism: vellus hair along the jaw and hairline, asymmetric freckles, capillary flush, wet living eyes with catch-lights. Fuzzy knit halo and individual yarn loops readable on the sweater.
STYLE
Photoreal editorial studio photography, 8K, no 3D render.
POSITIVE LOCKS
Single subject, no extra people, no text overlays, no watermark. Face, freckles, eye colour, brows, lips and hair identical in all four panels. Eyes stay natural, no eye glow.
Start Frame Selfie
Specifies phone camera optics and raking afternoon sunlight directly in the prompt text, achieving authentic smartphone lens characteristics without post-processing filters.
A close-up straight-on smartphone selfie of the same young woman, taken at arm's length with a wide-angle phone lens, placed in the right foreground of a sunlit contemporary living room. She has hazel-green eyes, sun-kissed light-tan skin with faint freckles across the nose and cheeks, and long straight balayage hair with darker brown roots and honey-gold highlights, one loose strand framing the left side of her forehead. She rests her cheek gently on her left palm, two minimalist silver rings on her fingers, glossy pink lips, calm introspective gaze into the lens. An oversized chunky deep-burgundy cable-knit sweater with ribbed cuffs fills the lower half of the frame. Direct mid-afternoon window sunlight rakes across her face, crisp shadows and warm highlights. Behind her, softly blurred: a beige pillow on a cream couch, a small bright window, a large arc of sunlight falling on the off-white wall. Palette of 50% creamy off-white and warm greige, 35% rich burgundy wine, 15% golden beige and honey. Moderate shallow depth of field, sharp detail on eyes and knit, slight wide-angle curve in horizontal lines, natural skin texture with visible pores, cozy quiet mood.
Location Reference Sheet
Generates an empty 2-panel environment plate (84° FOV wide and 47° FOV medium view) to lock the architectural layout, furniture placement, and directional golden-hour lighting.
SCENE CONTEXT
A photoreal location reference plate of the sunlit apartment living room seen in the attached first-frame selfie, empty, late afternoon.
LOCATION
A calm minimal apartment living room. A low deep sofa in cream-beige textured bouclé runs across the midground, with a square tan cushion with a soft satin sheen at the camera-left end and a second cream cushion beside it. Behind it a smooth warm off-white plaster wall. High on the camera-left wall, a deep-set square window with a thick white reveal frames a white modern apartment building with rows of windows, a bare tree and blue sky. Pale natural oak floorboards with a soft matte finish. A slim black phone tripod stands empty in the foreground, about two metres in front of the sofa. Nothing else in the room, uncluttered and airy.
LAYOUT
Two panels: wide establishing view at 84° FOV from chest height facing the sofa, and a medium view at 47° FOV of the sofa and wall.
OPTICS
Deep focus, zero distortion, straight rectilinear edges.
LIGHTING
Low golden late-afternoon sun from camera-right front, entering at a steep raking angle, throwing a hard bright rectangle of warm sunlight across the wall behind the sofa and a softer patch across the sofa cushions, deep soft warm shadow on the camera-left wall. Fine dust visible in the beam, faint atmospheric haze. WB locked 4600K, warm amber sun against cooler ambient shadow, soft highlight roll-off.
COLOR GRADE
Honey gold in the sun patch on the plaster, cream and oat in the bouclé, tan sheen on the cushion, cool pale blue in the window sky, color tied to source and surface, not a flat list.
STYLE
Photoreal interior photography, 8K, natural daylight rendition, fine grain.
POSITIVE LOCKS
No people, no text overlays, no watermark. Room layout, window position and sun direction 100% consistent across both panels.
Garment Reference Sheet (Cloth Sheet)
Employs a ghost-mannequin style to detail rib-knit construction, stripe patterns (navy and cream), and lapel collar geometry across 3 panels.
SCENE CONTEXT
A photoreal garment reference sheet of one knit vest, ghost-mannequin style, for a UGC try-on video.
SUBJECT
A cropped fitted sleeveless knit vest in fine vertical rib-knit. Wide horizontal stripes alternating deep navy and warm cream, starting and ending in navy, with a navy ribbed hem band. A solid navy notched lapel collar folds over into a deep V neckline, with a navy rib binding around the neck and the armholes. High cut armholes, close body that follows the bust and nips in at the waist, hem ending at the high waist. Soft matte cotton-blend surface, no print, no logo on the outside. A small off-white woven brand label is stitched into the inside back neck. 100% matches the reference.
LAYOUT
Three panels: front ghost-mannequin at 47° FOV, back ghost-mannequin at 47° FOV, macro detail at 12° FOV of the collar fold and rib texture.
BACKGROUND
Seamless neutral light-grey cyclorama.
LIGHTING
Even soft studio softbox key from the front-upper, gentle raking fill that lifts each rib in relief, WB 5500K.
MATERIAL
Individual rib columns and knit loops readable, crisp clean stripe edges, soft collar roll.
STYLE
Photoreal editorial e-commerce product photography, 8K.
POSITIVE LOCKS
No model, no hanger, no text overlays, no watermark. Stripe count, stripe widths and collar shape 100% consistent across all panels.
Brand Logo Macro Sheet
Renders the woven neck label (ZARA) with true stitch detail and serif typography in 2 macro panels (12° FOV and 18° FOV).
SCENE CONTEXT
A photoreal macro reference of one woven brand label stitched into the inside back neck of a navy and cream striped rib-knit vest.
SUBJECT
A small square woven satin label in off-white with a black high-contrast serif wordmark reading ZARA, clean folded edges, fine visible stitching along the top edge, one lower corner curling slightly away from the fabric. It sits on the navy rib binding of the inside back neck, with the cream and navy stripes of the vest visible soft around it. 100% matches the reference. Take only the label from the reference, never the black leather or dark satin lining around it.
LAYOUT
Two panels: straight-on macro at 12° FOV, and an angled macro at 18° FOV with fingers folding the neck open.
BACKGROUND
The vest fabric itself, softly out of focus beyond the label.
LIGHTING
Soft warm window light from the upper right, WB 4800K, off-white label bright without clipping.
STYLE
Photoreal macro product photography, 8K, true woven thread texture.
POSITIVE LOCKS
Wordmark straight, sharp and fully legible in both panels. No extra text, no watermark.
3. Seedance 2.5 Video Synthesis with @Reference Binding
The five reference assets are bound into Seedance 2.5 using @Image tags alongside explicit action sequencing and spatial blocking instructions:
SCENE CONTEXT
Real, unedited vertical smartphone footage of a young woman filming a fashion try-on alone in her sunlit minimal apartment at golden hour, posted as an organic creator video. She opens with a close handheld selfie on the sofa and calmly tells the lens she just tried Claude Sonnet 5.5. She stands, clips the phone onto a low black tripod, picks up a navy and cream striped rib-knit vest, shows it front and back, steps close and holds the inside neck label to the lens, pulls her burgundy sweater off over a nude seamless tank, slips the vest on, does an outfit check, then walks back up, lifts the phone off the tripod and softly reveals in a close selfie that Sonnet 5.5 wrote the whole video. Every frame must be indistinguishable from real phone footage.
ACTIVE REFERENCES
@[Image 1](image_1) CHARACTER. Face, dense natural freckles across the nose bridge and both cheeks, hazel-green eyes with a warm golden ring around the pupil, thick straight natural brows, glossy rose-nude lips with a defined cupid's bow, straight narrow nose, center-parted light-brown hair with caramel balayage and soft face-framing strands, slim build, oversized burgundy chunky mohair-blend funnel-neck sweater, cream wide-leg trousers, bare feet. 100% matches the reference.
@[Image 2](image_2) ENVIRONMENT. Long cream bouclé modular sofa with a tan silk cushion at camera-left and a cream lumbar cushion, warm off-white plaster wall, deep-set square window with a thick white reveal high at camera-left framing a white apartment building, a bare tree and blue sky, sun from camera-right casting a four-pane window-grid of light across the wall, pale oak plank floor with long streaks of sun, slim black tripod. 100% matches the reference.
@[Image 3](image_3) BRAND LABEL, LABEL ONLY. Off-white woven square label with the black serif ZARA wordmark, fine top stitching, one lower corner softly curled. Take only the label, stitched into the inside back neck of the vest. 100% matches the reference.
@[Image 4](image_4) FIRST FRAME. The exact opening selfie: framing, slight high angle, hand at cheek with the silver rings, sun on the camera-right side of her face, window at upper camera-left, curved patch of sun on the wall behind her. 100% matches the reference.
@[Image 5](image_5) GARMENT, THE VEST. Cropped fitted sleeveless rib-knit vest in wide horizontal stripes, navy at the shoulders, then cream, navy, cream, and a navy ribbed hem band, solid navy notched lapel collar folding into a deep V, high armholes, small plain label at the inside back neck. 100% matches the reference.
LOCATION MAP
Foreground: the slim black tripod about two metres from the sofa, and the woman moving between the tripod and the sofa on the oak floor. Midground: the long cream bouclé sofa, the tan silk cushion at camera-left, the folded striped vest waiting on the seat cushion beside her. Background: the off-white plaster wall carrying the warm window-grid of sunlight, the deep-set window at upper camera-left.
FIRST FRAME / BLOCKING
Non-empty opening frame: identical to @[Image 4](image_4). Close handheld selfie at arm's length from a slight high angle, she sits sideways on the sofa in the burgundy sweater, cheek resting on her fingertips, lips softly closed, golden sun raking across the camera-right side of her face, window at upper camera-left. She is already in motion, a slow breath lifting her chest as her eyes settle on the lens.
4. Audio Generation and Final Delivery
Voiceover and spoken dialogue are synthesized using Fish Audio and integrated into the final edit.
- By consolidating image generation (GPT Image 2.5), video synthesis (Seedance 2.5), and voiceover (Fish Audio) into an integrated pipeline, creators can produce high-converting fashion UGC assets rapidly without extensive manual iteration.
Original source
- Abdul Shakoor on X (@abxxai): Sonnet 5.5 and Seedance 2.5 Brand UGC Workflow