Consulting Higgsfield

Current brand

Recoup/Consulting/Business work defaults to Recoup Sky. Read brand/GUIDE.md and the workspace DESIGN.md; explicit client/artist branding or a fidelity-only migration wins. Stage local resources: node <installed-skill>/brand/materialize.mjs <output-directory>. Use its CSS, actual DM Sans/IBM Plex Mono fonts, exact outlined SVG logo, and brand/finals.json reference IDs. No recurring brand setup question. Save editable source and brand.lock.json with reference IDs/expression beside the output. Marketing owns the authored identity; the bundled snapshot works without another checkout.

Workspace: use the selected project and its AGENTS.md; keep existing entity folders and Reality headings. Business paths below are relative to that project. Bundled resources are relative to this installed skill; sibling capabilities resolve by their installed names. Never search another private checkout for missing inputs. Use workspace identity, audience, pricing, and DESIGN.md fonts. Local _work adapters and evals are optional workspace tools, not bundled dependencies. Check presence and current help first; otherwise use an available connector for the same scoped operation. If neither exists, report that step incomplete. For missing scorers, perform the stated checks and label the result manual/unscored; never invent a numeric score or successful provider action.

Generative AI media for the content flywheel. Where recoup-internal-consulting-graphics designs a still (HTML → PNG, pixel-perfect text) and recoup-internal-consulting-hyperframes-video renders video from HTML, Higgsfield generates pixels from a prompt — photoreal scenes, cinematic video, a face-consistent character, product photography, voiceover — the imagery you can't author by hand.

The Higgsfield CLI is the client (browser auth, bills credits). Command reference, auth, and cost mechanics live in integrations/higgsfield/AGENTS.md — read it; don't restate it here.

Route first — is this even a Higgsfield job?

The power move is the combo: generate the imagery in Higgsfield → add the text + branding on top with recoup-internal-consulting-graphics (HTML overlay). Never bake headline copy into an AI generation — the type won't be on-brand or reliably legible.

Prerequisite

CLI installed + authed — higgsfield account status shows the plan + credits. If it errors with Not authenticated, the user runs higgsfield auth login (browser; can't be automated). Full setup and the command cheat-sheet → integrations/higgsfield/AGENTS.md.

Workflow

1. Confirm the brief + load the brand. What's the asset, where does it go, what should it feel like? Pull look-and-feel from recoup-internal-consulting-tasteful-design (it carries DESIGN.md — palette, mood, the anti-AI-slop bar) and voice from recoup-internal-consulting-copy-writer. Encode brand mood/palette into the prompt; for anything the brand canon doesn't cover, ask — don't invent.
2. Pick the model from the live catalog — never invent a name. higgsfield model list (--video to filter) is the source of truth. Match the model to the job (image vs video vs voice vs product/marketing).
3. Write the prompt. Specific subject, composition, lighting, mood, palette (from DESIGN.md), and aspect ratio. Pass reference images with --image <path|uuid> (a local path auto-uploads). Keep text out of the image.
4. Preview cost — every time (money). higgsfield generate cost <model> --prompt "…", and check higgsfield account status for balance. State the credit cost to the user before any paid run.
5. Generate. higgsfield generate create <model> --prompt "…" [--image <id>] --wait (blocks, prints the result URL). Add --json to parse. Image gen can take a couple minutes; video longer.
6. Land the raw result. Use the exact dated work folder supplied by the calling workflow. Otherwise create work/YYYY-MM-DD-<task-name>/ in the selected workspace; reuse it for the task. Download original provider files there and record their URLs and hashes. Keep raw downloads out of Git; promote only selected delivery assets and compact provenance into the bundle.
7. Finish + brand (usually a handoff). Needs text/logo/framing → pass the generated image to recoup-internal-consulting-graphics as a background and overlay branded type. A clip for a post → recoup-internal-consulting-hyperframes-video can package it with captions/overlays.
8. Promote keepers → content/. Move only what you'll use into the draft's folder (or content/_work/ while staging; content/04-published/ once shipped). The repo isn't the archive — Higgsfield's cloud keeps job history (higgsfield generate list).
9. Publish. Hand the final asset + caption to recoup-internal-consulting-linkedin-publisher (Postbridge). Never auto-send — confirm the final creative + copy first.

Soul characters (a consistent brand persona)

To reuse the same AI person across many posts (a recurring face, a mascot, the owner-as-avatar): train a Soul once with higgsfield soul-id …, then reference it in later generations for a consistent look. Training is an expensive, gated lane — confirm the spend and the reference images with the user before running it. See integrations/higgsfield/AGENTS.md and higgsfield soul-id create --help.

Cost discipline (paid — non-negotiable)

• Preview before every spend (generate cost); after a paid run, cite the actual credit cost — quote the real number from the CLI, don't estimate (evidence discipline).
• Gate the expensive lanes — video + Soul training — behind explicit user go-ahead. Iterate the look on a cheap image model first, then commit to the costly render.
• Never loop generations unattended — each create bills credits.

Don't

• Don't bake headline / marketing text into the AI image — overlay it with recoup-internal-consulting-graphics.
• Don't invent model names — higgsfield model list is the source of truth.
• Don't commit raw media to git — it lives in gitignored _work/; promote only keepers to content/.
• Don't auto-publish — confirm the final creative + caption before recoup-internal-consulting-linkedin-publisher sends.