Fashion · Lookbook pipeline
WorkflowFashion & apparelFrom product photos to a seasonal lookbook with stills and short films
A clothing brand needs a lookbook for every drop and rarely has the budget for a full shoot each time. These notes cover how product photos become lookbook stills and short clips, explored in Chat and repeated by a workflow.
Valnora & Elric team5 min read
A clothing brand releases a new drop every six weeks. Each drop needs a lookbook: a set of stills that show the garments worn in a place and a light that says what the season is about, plus a few short clips for the lookbook page and social. A full shoot per drop costs more than most small brands can spend.
These notes follow a creative lead who builds lookbooks in Agent Serve from the product photos the team already has. The exploring happens in Chat. The repeatable part becomes a workflow. For how a Valnora & Elric brand runs its own creative this way, see the Aestlo story.
Where does the lookbook start?
It starts with two things the team owns already: clean packshots of every product, and a one-page art direction brief for the season. For an autumn drop, the brief might say: stone courtyard, late-afternoon light, one model walking, muted palette, no logos visible, never change the collar shape, pocket placement or print.
The brief matters more than the model. Every prompt in Chat and every block in the workflow reads it, so the forty-eighth still is held to the same direction as the first.
How does Chat help find the look?
The creative lead opens a new chat and uses + → Media models to pin an image model, such as Nano Banana 2.1, and a video model, such as Seedance 2.5, for that chat. The reasoning model stays separate, so the conversation about the brief does not change which model makes images.
They attach the packshots of two hero products and ask for three stills each, against the brief. They compare, ask for tighter framing, try Seedream 5.0 for a softer light, and settle on what the season looks like. Then they ask for a six-second clip from the best still, used as the first frame, and try Kling v3.0 for a second take.
Each generation estimated above the workspace's cost confirmation threshold, US$1 by default, asks before it runs. Video is where that usually happens. A running generation shows a progress card with a cancel button, so a take that is clearly going wrong can be stopped.
Where do all the files go?
Into the Library, automatically. Stills land under Generated/Images and clips under Generated/Videos, each with its prompt, model and cost. When the founder asks next week which prompt made the courtyard shot, the answer is attached to the file. Nothing lives only in a download folder.
How does the exploration become a workflow?
One look, from product photo to approved clip
Workflow · Stills
- 01Product photosThe Shopify block reads the product and its packshots for the look.
- 02Art directionThe season brief: location, light, casting notes, styling rules and what never to change on a garment.
- 03Lookbook stillsAn Image Generator block, with a model such as Nano Banana 2.1 or Seedream 5.0, uses the packshots as references.
- 04Creative lead picksA Human in the loop block pauses until a person chooses a still or rejects all.
Workflow · Clip and record
- 05Cinematic clipA Video Generator block, with a model such as Seedance 2.5 or Kling v3.0, uses the approved still as the first frame.
- 06LibraryStills and clips are saved under Generated/Images and Generated/Videos with prompt, model and cost.
- 07Lookbook tableOne row per look: product, chosen still, clip, status and who approved it.
Once the look is settled, the creative lead turns it into a workflow that runs once per look:
- Read the product. The Shopify block fetches the product and its packshots.
- Generate stills. An Image Generator block makes three stills against the art direction, with the packshots as reference images.
- Pick one. A Human in the loop block pauses the run. The creative lead compares each still with the packshot, picks one or rejects all three.
- Make the clip. A Video Generator block animates the chosen still into a short clip, with the still as the first frame so the garment stays as approved.
- Record it. The workflow writes a row to a Lookbook table in Tables.
For a drop of 24 looks, the workflow runs over the list in a loop or a Parallel block. There is no single button that generates a whole drop unattended, and the pause at step 3 is deliberate.
What does a finished look record?
What the solution produces
Lookbook table row for one look (illustrative)
Record- Look
- AW-07 · linen overshirt, ecru, with wide-leg trouser
- Stills generated
- 3, with Seedream 5.0, packshots as references
- Chosen still
- Still 2: walking shot, late-afternoon light, stone courtyard
- Rejected
- Still 1: collar shape changed. Still 3: pocket missing.
- Clip
- 6-second slow push-in from the chosen still, Seedance 2.5, first frame locked
- Library
- Generated/Images and Generated/Videos, with prompt, model and cost
- Status
- Approved for the lookbook page · not yet for paid ads
The row says which still was chosen, why the others were rejected, which models made what, and where each file lives. The rejection notes are the most useful part over time: if one model keeps changing collars, the team knows which to avoid for structured shirts.
What keeps quality and cost in check?
Controls and governance
Approvals before action
Relays never send email or book meetings without an approval, under any autonomy setting. An approval covers the exact payload the reviewer saw and expires if nobody decides.
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
More on how Agent Serve handles access, data and audit on the security page.
Media Generation is switched on only for the creative team's access policy, with default image and video models set, so nobody reaches for the priciest video model by habit. The creative lead approves every still and clip before it appears anywhere. Approved for the lookbook page does not mean approved for paid ads; that is a separate decision, covered in short product video ads from one approved still.
Questions teams ask
Will the garment look exactly like the real product?
Not always. Image models can change a collar, a pocket or a print. That is why every still is checked against the packshot by a person, and why the art direction lists what must never change. Rejecting all three stills is a normal outcome.
Can it generate the whole drop in one go?
The workflow runs per look. For a drop of 24 looks, run it over the list in a loop or Parallel block, with the Human in the loop step pausing each look until someone decides.
What you need
- Clean packshots for every product in the drop, front and back, in Shopify or the Library
- A one-page art direction brief per season: location, light, casting, styling and garment rules
- Media Generation switched on in your access policy, with default image and video models set
- A cost confirmation threshold the creative lead is comfortable with
- A creative lead who approves every still and every clip before it is used
- Rights checked for any model likeness, location reference or music you add later