Fal lora training

Skill shahshrey/fal-portrait-lora-skills/skills/fal-lora-training

Validates, uploads, submits, monitors, and persists fal portrait LoRA training runs using fal-ai/flux-lora-portrait-trainer. Use when the user asks to train, fine-tune, start, submit, or monitor a portrait/person LoRA from a prepared ZIP; or mentions images_data_url, trigger_phrase, portrait trainer, training steps, learning rate, LoRA weights, or a fal training request.From its SKILL.md

Install

npx -y skills add shahshrey/fal-portrait-lora-skills --skill fal-lora-training

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, 727 tokens by cl100k_base, as published. Nobody here has run it

fal Portrait LoRA Training

Train a prepared dataset with fal-ai/flux-lora-portrait-trainer.

Defaults

Setting	Default
Endpoint	`fal-ai/flux-lora-portrait-trainer`
Steps	`2500`
Learning rate	`0.00009`
Multiresolution	`true`
Subject crop	`true`
Create masks	`false`
Captions	Matching `.txt`, beginning with `[trigger]`

Paid-submission rule

Training is a paid external action.

If the current user request explicitly says to start, submit, run, or train, proceed without asking twice.
Otherwise, show endpoint and settings and ask for confirmation.
Pass --yes only after one of those conditions is met.

Workflow

Resolve the project-local skill first:

SKILL_DIR=".cursor/skills/fal-lora-training"
[ -d "$SKILL_DIR" ] || SKILL_DIR="$HOME/.cursor/skills/fal-lora-training"
[ -d "$SKILL_DIR" ] || SKILL_DIR="skills/fal-lora-training"

1. Validate the archive

.venv/bin/python "$SKILL_DIR/scripts/validate_dataset.py" \
  "/path/to/dataset.zip" \
  --require-placeholder

Do not submit unless validation passes. The archive must be flat, contain at least ten images, pair every image with a same-stem .txt, and use [trigger] captions.

2. Confirm settings

Ask if missing:

Dataset ZIP path
Trigger phrase
Output directory for request/result JSON

Use defaults unless the user specifies different portrait-training values. An endpoint override is allowed only when it accepts the same input schema. See references/portrait-trainer-schema.md.

3. Submit and monitor

Run as a background command because training is long-running:

.venv/bin/python "$SKILL_DIR/scripts/train_portrait_lora.py" \
  --archive "/path/to/dataset.zip" \
  --trigger-phrase "ohwx man" \
  --output-dir "./portrait_training_run" \
  --env ".env" \
  --yes

The script:

Revalidates the ZIP.
Uploads it with fal_client.upload_file.
Submits images_data_url to the portrait endpoint.
Writes training_request.json immediately.
Waits for completion.
Writes training_result.json containing LoRA and config URLs.

After backgrounding, read the command output once to confirm it reached SUBMITTED request_id=.... Rely on completion notification rather than frequent polling.

4. Deliver

Report:

Request ID
Endpoint and settings
Dataset image count
Trigger phrase
LoRA weights URL
Config URL

Authentication

Requires fal-client and either FAL_KEY or FAL_API_KEY. The script maps FAL_API_KEY to the client-compatible FAL_KEY variable without printing it.

Scripts

Script	Purpose
`scripts/validate_dataset.py`	Validate flat image/caption ZIP contract
`scripts/train_portrait_lora.py`	Upload, submit, monitor, persist results

What ships with it: 3 files

9.3 KB alongside SKILL.md, 2 of them executable

references/

portrait-trainer-schema.md1.2 KB

scripts/

train_portrait_lora.pyruns4.3 KB
validate_dataset.pyruns3.9 KB