Operations & Training

See training tasks, model status and failures in one local console.

Review training data, start supported local model jobs, follow live output, stop active tasks, and inspect which quality-control model is currently available.

Request beta access Explore Selection

Local task runner Style and hand-quality models Hand detector Reviewable logs and rollback status

Setvyn — Operations ConsoleLOCAL

Style QC trainingCompleted

Hand quality trainingRunning

Hand detector trainingQueued

[10:14:02] Dataset validation complete

[10:14:05] Embeddings ready

[10:14:08] Evaluating candidate model▋

1 running2 tasks queued/done

Datasets Style QC Hand quality Hand detector Tasks & logs

Local runtime visibility

Keep datasets, model state, tasks, and output in the same operational view.

Local quality-control training produces several moving parts: labelled samples, embeddings, candidate models, validation reports, background tasks, and failure output. The Operations Console brings their current state together before the user starts or changes a model.

Check readiness first

Review labelled examples, embeddings, pending work, and available model files before training.

Follow the active job

See queued, running, completed, failed, or stopped tasks with recent output.

Keep recovery visible

Review quality reports and use supported rollback actions for hand models when a previous validated model is available.

01 · Style QC

Train a local scoring model from reviewed style labels.

The Style QC section summarizes locally reviewed examples across Bad, Okay, Good, and Top labels, together with available embeddings, pending work, and active or latest model status.

Bad, Okay, Good, and Top label totals
Embedding count and pending work
Active and latest style-model status
Start the supported local Style QC training job

Style QC datasetDEMO DATA

42Bad

118Okay

205Good

64Top

Embeddings429

Pending18

ModelActive

Distribution values are demo data. The model does not automatically understand artistic quality, replace human review, or improve without correctly labelled and representative training data.

02 · Hand quality

Review the balance behind the local hand-quality model.

The Hand Quality section separates visible good hands, occluded good hands, bad examples, ungradable samples, embeddings, reviewed images, and manual annotations so the dataset balance can be checked before training.

Visible-good, occluded-good, bad, ungradable, and unlabeled context
Good and bad embedding totals
Reviewed-image and manual-annotation counts
Good-to-bad dataset balance guidance
Active, latest, and previous model status
Candidate quality report with available validation metrics
Supported rollback to a previous hand-quality model

Hand quality reportDEMO

Visible good164

Occluded57

Bad98

Good : bad balanceAcceptable

0.86Balanced accuracy ·Demo

0.82Bad recall ·Demo

0.88Good recall ·Demo

Roll back previous model Start training

Quality metrics describe evaluation on the available local validation data—demo values here, not Setvyn benchmarks. They do not guarantee accuracy on every future image, pose, style, or dataset.

03 · Hand detector

Inspect detection samples before training the local box model.

The detector section summarizes sample images, annotated hand boxes, negative images, detector-only examples, invalid records, active model status, and the latest available quality report.

Positive and negative image counts
Annotated box totals
Detector-only and invalid-sample visibility
Active and previous detector model status
Start the supported hand-detector training job
Supported rollback when a previous detector model exists

Detector datasetDEMO

box 01 box 02 abstract sample · not a real image

Positive images312

Negative images96

Annotated boxes548

Invalid records7

The diagram is abstract—no real training images or hand crops are shown. Setvyn does not claim medical-grade detection, universal anatomy correction, automatic image repair, perfect hand localization, or reliable performance outside the reviewed dataset.

04 · Monitor

See what is queued, running, completed, failed, or stopped.

Supported training actions start local background tasks. The console refreshes their state, keeps recent tasks available for selection, and shows the latest standard and error output for the selected job.

Queued, running, completed, failed, and stopped states
Task title and start time
Recent standard-output and error-output tail
Stop control for an active task
Duplicate-job protection for the same training kind

Task · Hand quality trainingRunning

Started 10:13 · elapsed 00:04:12

[10:14:02] Dataset validation complete

[10:14:05] Embeddings ready

[10:14:08] Evaluating candidate model

[10:14:11] Writing validation report▋

stdout stderr · 0 lines Duplicate job blocked

Stopping a process is an operational action and may leave an incomplete candidate or temporary data that requires review. This is not transactional cancellation and does not guarantee automatic cleanup of every external process artifact. Log text shown is illustrative—no real command prompt, arguments, PIDs, or paths.

Model safety

Make the active model and available recovery state visible.

Where a quality report is available, Setvyn shows whether the latest candidate was promoted or rejected, the recorded reason, and supported comparison metrics. Hand-quality and hand-detector sections can expose a previous model for rollback.

1Train candidate

2Evaluate report

3Promote or reject

4Keep active model visible

Roll back supported hand model

Not every candidate is promoted automatically; promotion follows the report and the user's decision.

Rollback is exposed for the hand-quality model and hand detector—not the style model in the current console.

Rollback needs an existing previous model and no conflicting job; it restores that model state, not dataset edits or all downstream results.

Local by design

Keep training data, models, tasks, and logs inside the local runtime.

The Operations Console reads local dataset statistics, model files, task state, and recent logs, then launches the packaged local training scripts available to the installed edition. Training requirements and runtime support depend on the local system, dependencies, datasets, and configured Setvyn package.

Local operational data

Labels, annotations, embeddings, dataset statistics, model state, task metadata, and log files.

User responsibility

Review training material, remove private or unlicensed data, monitor resource usage, inspect failures, and validate model behavior before relying on it.

This does not imply cloud training, anonymous telemetry, automatic backups, cross-platform availability, isolated execution, guaranteed GPU support, or complete protection from malicious training data.

Connect reviewed data to the quality-control workflow.

Choose the operational controls available to your workflow.

Operations Console and training capabilities vary by Setvyn edition, installed components, local scripts, datasets, and runtime configuration.

Compare Setvyn editions

FAQ

Operations and training FAQ

What models can I train from the current console?

The current visible workflow supports local Style QC, Hand Quality, and Hand Detector training when the required capability, scripts, data, and dependencies are available.

Does Setvyn train a foundation image model?

No. This page describes focused local quality-control and hand-detection models, not training a general image-generation foundation model.

Can I watch a training job while it runs?

Yes. The console shows task status and recent standard and error output for the selected supported training task.

Can I stop an active job?

The console provides a stop action for active managed tasks. The resulting files and model state should still be reviewed after an interrupted run.

Can I restore a previous model?

A supported rollback is available for the hand-quality model and hand detector when a previous model exists and no conflicting job is running. A style-model rollback is not promised by the current console.

Do validation metrics guarantee future accuracy?

No. They describe performance on the available evaluation data and can change with the dataset, labels, model, and image domain.

Does the Operations Console currently expose Reddit scan controls?

The current public description focuses on the training and managed-task interface that users can access. Reddit scan controls are not marketed as part of this visible page.

Understand the local training state

Review the data, task, report, and active model before the next run.

Bring focused model training and operational visibility into one local Setvyn console without relying on disconnected scripts and log files.

Request beta access Explore all features