Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/OSideMedia/higgsfield-ai-prompt-skillnpx agentmods add skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soulWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soul)<a href="https://agentmods.dev/skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soul"><img src="https://agentmods.dev/badge/skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soul/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soul"><img src="https://agentmods.dev/badge/skills/osidemedia/higgsfield-ai-prompt-skill/higgsfield-soul.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Memory Poisoning · line 235 Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.Fix: Protect agent memory and state from modification by untrusted content. Use read-only memory for critical instructions and validate all state changes.
- medium YARA Match · line 711 YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).Fix: Remove offensive tool references and exploit code. Legitimate agent skills should not contain penetration testing tools, exploit frameworks, or reconnaissance utilities.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00086 | $0.12990 |
| Opus 5.5 | $0.00034 | $0.05196 |
| Sonnet 5.5 | $0.00017 | $0.02598 |
| Haiku 4.5 | $0.00009 | $0.01299 |
Grade A, and why
higgsfield-soul scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 1,048 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Higgsfield Soul ID — Character Consistency
QUICK FACTS
Generated-checked block (scripts/build_index.py verifies anchors). Read the linked sections for full context — these lines are routing aids, not the rules themselves.
- Hard rule: every Soul ID prompt splits into Identity Block (static descriptors only) + Motion Block (temporal/camera only) →
- Don't re-describe the face or core features — only describe what differs from the base character →
- Reference image rules: front or 3/4 angle, even lighting, neutral-to-slight expression, no blur, solo subject →
- Reference generators: Soul 2.0 (fashion-forward), Nano Banana Pro (max sharpness), Seedream 4.5 (style range) →
- Character sheet angles: front face, 3/4, side profile, optional full body + optional embedded prop sheets →
- Two-image floor per character: one clear face + one full body, on neutral grey — never a single image; the grey law's shade and mechanism live once in
../../templates/ad-asset-prep.md→ - The Untouched Base: face pass in close-up, then a looks pass; the base close-up is never run through a model again — changes go on with masks, and a new state is a new named asset →
- Reference plates: biological realism ON, photographic capture behaviour OFF (no key direction, cast shadow, bokeh, flare); the background is a field, not a room; lean prompts once a reference carries identity →
- Prefer the single-prompt 3×2 six-panel sheet (one 16:9 generation) — identity locks better than multi-step assembly →
- Outfit change without identity drift: split-panel sheet — ghost-mannequin outfit LEFT + "face matches input 100%" close-up RIGHT →
- Crowds: a single-character reference makes a clone army — build a multi-character lineup and declare it a "VARIETY reference" →
- Character Anchor Block = 10 per-shot attributes (identity, screen position, depth layer, frame occupancy, orientation, pose, gaze, contact points, state lock, expression) →
- One Soul ID sheet PER character state — 5 transformation stages = 5 distinct sheets; the prompt names the stage →
- Micro-expression presets: 9 core + 10 extended →
- Cinema Studio: identity goes in the @ Element definition, motion goes in the prompt field →
- Soul Cast is Business/Team plans only; 8 parameter categories incl. 12 Archetypes, Budget $10M–$500M, Era 1900s–2020s →
- Soul Cast specs: up to 4K (Character/Location modes) / 2K (General); batch 1 or 10; 0.125 credits per image →
- Use 2–3 reference shots (frontal, 3/4, side); if features drift, use the character sheet as @Image1 →
- Soul Cinema is the default CS 3.0/3.5 image-mode model — single-step, ~0.125 credits/image →
- Studio-feeling output is an intermediate, not a final — re-pass through Soul Cinema with grade, directional lighting, lens character; the result is a derived look frame, never the identity base or a reference plate →
- Skip Soul ID for single shots or when you want maximum creative variation →
- Plasticky face in wide shots: crop the face from a closer-shot panel and replace it in post →
- Casting a REAL person: the sheet alone returns a lookalike — erase the head on the body panels and paste their actual photo into the portrait panel →
- Run the same sheet prompt through 2–3 image models and compare before locking; the deciding axis changes per character →
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed · +47 lines 42853377387e
- 1mo ago First seen · 1,001 lines · 86 tokens per session scan A 6cb4acb739f7
higgsfield-soul is a skill published in the GitHub repository OSideMedia/higgsfield-ai-prompt-skill (800 stars, last pushed 2d ago), licensed MIT. It adds 86 tokens to every session and 12,990 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
seedance-shotlist-director
Generate a director's shotlist as an editable HTML production board for Seedance 2.0. Use whenever the user provides a script, scene breakdown, story idea, or treatment to turn into a numbered shotlist with English Seedance prompts — trigger on "make a shotlist", "director’s shotlist", "break this script into…
vox-director
Turn ONE topic into a finished Vox-style paper-collage explainer / ad video, end to end on the Atlas Cloud API + local ffmpeg — script, collage keyframes, motion, voice-over, music, captions, all automated. Use this whenever the user wants a "Vox style" video, a paper/torn-paper collage animation, a "motion collage"…
sketch-correction-worker
Use when doing single-beat sketch quality control and preparing correction labels from a SuperTale project path and episode number. This skill turns one episode's selected sketches plus local script beat data into selectresult.jsonl, selectsummary.json, tasks.jsonl, and labels.jsonl for edit execution. Prefer using…
freezone
Use when the active chat surface is 虾画/Freezone/canvas, or when the user asks about canvas nodes, visual boards, selections, graph edits, canvas actions, layout, or Freezone short-video workflow work.
dramaclaw-workflows
Create, revise, confirm, run, or resume multi-node DramaClaw Freezone canvas workflows. Use for requests involving a workflow, several connected nodes, grouped production stages, storyboards, or text/image/video/audio pipelines; do not use for one standalone canvas operation.
workflows
Use in Freezone/虾画 chat when the user selects a Skill or asks to create, plan, analyze, or expand a dynamic canvas workflow for short drama, ads, ecommerce, product video, MV, text, image, video, or audio production.