Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Cast Library and LoRA training · guaardvark | DecimalAI

ai mlby guaardvark

Cast Library and LoRA training

guaardvark/cast

Source guaardvark/guaardvark ↗License MIT

Build consistent characters, environments and props in Guaardvark's Cast Library and train LoRAs for them locally (reference photos → vision bible → sample plan → approved samples → training).

SkillScore87Maintained

Automatic

AI Engineer pack

Last verified 10/10/2026

What it changes

Passed 20 of 23 test cases with the skill, and 6 of 23 without it.

Without skill26%

With skill87%

Checks it passed that the model alone missed

Includes JSON body with kind set to prop.
Identifies default base URL as http://localhost:5000.
Specifies GET /api/cast-library for listing subjects.

and 17 more in the full report

14 fixed6 passed either way3 still failing0 broke

Pass-rate lift: +61 pp
Tokens: -22%
Agent turns: 0%
Response time: -51%

Model gemini-3.6-flashMeasured 10/10/2026v120 of 23 cases passed with skill

Tokens, turns and time describe resource use, not a dollar price.

Use this skill

Save it into your project's skills folder. Claude Code picks it up in its next session.

curl -fsSL --create-dirs -o .claude/skills/guaardvark-cast/SKILL.md https://app.decimal.ai/s/guaardvark-cast/SKILL.md

Want it pinned and updatable? pip install "decimalai>=0.15", then decimalai skills add guaardvark-cast, and start Claude Code with claude --plugin-dir "$(git rev-parse --show-toplevel)/.agents"

1 · One-time — install the CLI (docs ↗)

pip install "decimalai>=0.15"

2 · Get the skill — writes .agents/skills/guaardvark-cast/ and pins it in .decimal/skills.lock

decimalai skills add guaardvark-cast

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/guaardvark-cast/SKILL.md https://app.decimal.ai/s/guaardvark-cast/SKILL.md

Raw file: /s/guaardvark-cast/SKILL.md ↗ · machine-readable summary at /s/guaardvark-cast.json ↗

Safety checks: 3 of 3 passed

Code scan CleanIntent PassedContent Clean

See it in action

One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.

1 of 14

Prompt

I am writing a curl script to interact with the local cast library server. If `GUAARDVARK_URL` is not set in my environment, what is the default HTTP base URL and port used by the library?

What was checked, without and with the skill
Checked	Without skill	With skill
Identifies default base URL as http://localhost:5000.	Missed	Passed

case-13 · gemini-3.6-flash · 10/10/2026 · answers are capped and redacted

What's inside

The instructions your model reads when the skill loads.

SKILL.md39 linessingle file

Cast Library and LoRA training

Read setup first; training needs the lora_trainer plugin (CUDA, bf16) and an installed train-ready base (Z-Image, SDXL or FLUX family). B=${GUAARDVARK_URL:-http://localhost:5000}.

The pipeline

Create the subject

curl -s -X POST $B/api/cast-library/subjects -H 'Content-Type: application/json' -d '{
  "kind": "character", "name": "Mara", "description": "late 30s, short grey hair, freckles",
  "trigger_word": "mara_v1", "voice_id": null
}'

kind is character, environment or prop. GET $B/api/cast-library lists subjects; the numeric id is what generate_image and batch routes take as subject_ids.

Upload reference images (5 to 20 clear shots, varied angles, same subject): curl -s -X POST $B/api/cast-library/subjects/$ID/upload-refs -F [email protected] -F [email protected] Ask first whether the person consented; do not train on someone who has not.
Vision bible from the refs: POST $B/api/cast-library/subjects/$ID/bible/from-refs (the vision model writes the identity description that every prompt inherits).
Plan and generate training samples: POST .../$ID/plan then POST .../$ID/generate (cancel with .../generate/cancel). Review GET .../$ID/samples; view one with GET .../samples/<sample_id>/image; drop bad ones with DELETE .../samples/<sample_id> or POST .../samples/<sample_id>/regenerate.
Approve samples: POST .../$ID/samples/approve.
Train: POST $B/api/cast-library/subjects/$ID/train with optional {"training_settings": {...}}. A 409 already_training means wait; a train_base_not_ready error names the base model to install first. Cancel: .../train/cancel. Progress: GET $B/api/cast-library/subjects/$ID (status, base model, LoRA path when done).
Use it: generate_image with subject_ids=[ID]; batch image/video routes with subject_ids; the Film Crew casts it; the music-video Director locks it per cut.

Rules

Training is a GPU job of tens of minutes; say so and check inspect_gpu for conflicts first.
The trigger word is applied by the system when subject_ids is passed; the user does not need to type it.
Everything stays local: refs, samples and the LoRA file under data/.

Track record

Benchmark

20 / 23

test cases passed, vs 6 / 23 without

Adoption

Users: 0
Times used: 0

Safety and source details

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v5 — 9/14/2026

View 15 scanner checks

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

guaardvark / guaardvark

.agents/skills/cast/SKILL.md

View on GitHub ↗

License MITImported 9/14/2026Version v1

More skills from the same source and nearby in the registry, with their headline benchmark lift.

Ready to supercharge your AI agents?

See it in action

What's inside

Cast Library and LoRA training

The pipeline

Rules

Track record

Benchmark

Adoption

SkillSafety

Source

Related skills