Keep your brand's AI characters consistent across campaigns

Build me an app that manages my brand's recurring cast of AI characters, so the same face shows up consistently across every campaign instead of drifting each time someone generates a new image. The main surface is a roster of character cards. Each card is one character and holds its trained Higgsfield Soul ID, a thumbnail, its look notes, its preferred style, and a count of the shots and shoots produced from it. Clicking a card opens that character's detail page with its full gallery.

Store the roster in the app's own storage: for each character keep a name, the Soul ID returned by Higgsfield, the reference image URLs used to train it, free-text look notes (hair, wardrobe rules, framing preferences, anything the team must not let drift), and a preferred Soul style id. For each shoot keep the character it belongs to, the theme the user typed, the time it ran, and its status. For each shot keep the image URL, the exact prompt used, the shoot it belongs to, its status, and whether it has been archived to Drive or queued to social.

The "Train a new character" form takes a name, look notes, and a set of reference photos uploaded from my machine. Reference images have to be public URLs before Higgsfield can use them, so the handler first calls Generate Upload URL for each file, PUTs the bytes to the returned presigned URL, then calls Create Soul ID (Custom Character Reference) with those URLs to train a reusable character reference. Training is asynchronous, so poll Get Request Status until the request resolves and show the card in a training state until it is ready. Populate the preferred-style picker from List Soul Styles, since styles are chosen by id. Also give me an "Import existing" option that pulls Soul IDs already on my account with List Soul IDs (Custom Character References) so I can bring characters I trained elsewhere into the roster.

On any character I hit "Run a shoot" and type a theme, for example autumn campaign, gym wear, or office headshots. That kicks off a background agent scoped to that character. The agent reads the character's look notes and preferred style, then writes a varied set of shot prompts around the theme, varying framing, setting, wardrobe, and lighting while holding the character's described look fixed. It generates each one with Generate Image with Soul (Text-to-Image), passing that character's Soul ID as the custom reference and its preferred style id, so the likeness holds. Generation is asynchronous: each submit returns a request id, so the agent polls Get Request Status until each one reaches completed, failed, or nsfw, and writes the finished images back into that character's gallery grouped as a single shoot tagged with the theme. Shots that fail or get blocked are recorded with that status rather than silently dropped, and the rest of the shoot still lands. Show shots streaming into the gallery as they finish instead of waiting for the whole batch.

The gallery on a character page is grouped by shoot, newest first, with the theme as the group heading. Every individual shot must always show which shoot and which theme it came from, including when I open it full size or view a filtered feed of the character's best work. Keep the character's look notes and preferred style pinned in a panel right next to the gallery, editable in place, because consistency is the entire point of this app and those notes are what the next shoot reads from.

From any shot in the gallery I can promote a favourite into motion. Give me a "Make it move" action that offers Generate Video with DoP (Image-to-Video, Motion Presets), with the motion preset chosen by id from List Motions, or Generate Video with Kling 2.1 Pro (Image-to-Video) for a longer clip. Both take the shot's image URL as input, both are asynchronous, so poll Get Request Status the same way and attach the finished video back to that shot in the gallery so the clip stays connected to the shoot and theme it came from.

Approved assets get archived with Google Drive. An "Archive" action on a shot or on a whole shoot uploads the image or video with Upload File (Multipart), filing it into a folder named for the character and the shoot theme, creating that folder with Create Folder the first time it is needed. Mark archived assets in the gallery so I can see at a glance what has been filed and what has not.

A "Queue to social" action on a shot schedules it through Buffer. Populate a channel picker from List Channels, then call Create Post with the shot's image or video attached, letting me pick between adding it to the queue, sharing now, or scheduling for a specific date and time, and let me edit the caption before it goes. Record on the shot that it has been queued, to which channel, and when, so the gallery doubles as a record of what has already gone out.

Rules to hold throughout: never generate a shot for a character without passing that character's Soul ID, since an unanchored generation defeats the purpose. Treat the roster as the home screen and the character gallery as the surface people live in. Handle the asynchronous nature of every Higgsfield generation gracefully with visible pending states, and treat a blocked result as an ordinary outcome to display rather than an error that breaks the page.