LoRA Studio v0.4.6-beta (Phase 11.9.1)

LoRA Studio phase11.9.1 / v0.4.6-beta consolidates Candidate Standard Comparison, Retry Signal Summary, performance profiling, and validation image generation follow-up review automation into the current beta release.

Highlights:

• Added Candidate Standard Comparison for Standard Validation v1 across loss candidate epochs.
• Unified validation image generation flows so missing Embedding / Machine Review can run automatically after generation.
• Improved Epoch cross Matrix weight generation flow and URL normalization.
• Refined Job / Review / Performance / Retry Signal UI wording and Japanese labels.
• Updated app version and release docs to v0.4.6-beta / Phase 11.9.1.

Validation:

• python -m compileall app start_lora_studio.py
• node --check app/static/js/app.js
• python -m pytest -q -p no:cacheprovider: 132 passed
• git diff --check

LoRA Studio v0.4.5-beta (Phase 11.9.0)

LoRA Studio v0.4.5-beta packages Phase 11.9.0.

Highlights:

• Adds Retry Signal Summary as a read-only diagnostic for Training / Review / Weight Calibration results.
• Shows retry_signal_label, confidence, reasons, and recommended next actions on Project, Job, Review Session, and LoRA Profile screens.
• Keeps retry guidance separate from the Recommendation Engine; no Draft Job creation or automatic Run is performed by Retry Signal Summary.
• Includes Phase 11.8.x performance profiling and Machine Review optimization updates.

Validation:

• python -m compileall app start_lora_studio.py
• node --check app/static/js/app.js
• python -m pytest -q -p no:cacheprovider: 130 passed
• git diff --check

LoRA Studio v0.4.4-beta (Phase 11.7.1)

Highlights

• Added the Step Estimator / Target Step Assistant to show expected total steps from image count, repeats, epochs, batch size, and gradient accumulation.
• Added Optimizer Definition / Optimizer Profile / Training Recipe target steps, including optimizer LR meaning and target source tracking.
• Added target-step repeat auto-calculation, with creation snapshots for auto-calculated repeats and target step source.
• Added the Weight Calibration Pipeline flow for adopted LoRA validation, including preflight, pipeline status, Weight Review Matrix links, and profile apply flow.
• Fixed multi-subset repeat calculation so Step Assistant uses bounded, uniform-repeats calculations without hanging.

Validation

• python -m compileall app start_lora_studio.py
• node --check app\static\js\app.js
• python -m pytest -q -p no:cacheprovider
• git diff --check

LoRA Studio v0.4.3 (Phase 11.6.16.6)

Highlights

• Added weight selection controls to validation matrices, including 0.0-2.0 selectable weights with the 1.1-2.0 range hidden by default.
• Added matrix-level actions for refreshing the displayed weights, generating only missing images, and recalculating only missing Embedding / Machine Review results.
• Improved bulk validation generation state handling so queued/running runs stay visible and stale queued records are reconciled.
• Updated Matrix actions so weight selection is the primary context and display refresh, additional generation, and missing review recalculation are same-level actions.

Validation

• python -m compileall app start_lora_studio.py
• node --check app\static\js\app.js
• python -m pytest -q -p no:cacheprovider
• git diff --check

LoRA Studio v0.4.2 (Phase 11.6.16.5)

Highlights

• Refined Review Session detail flow so ungenerated planned/prepared sessions keep generation as the primary path.
• Added clearer guidance from selected LoRA jobs to weight validation while preserving a Review Matrix recheck link.
• Improved safeguards and recovery around review preparation progress, partial generation counts, and retry paths.

Verification

• python -m compileall app start_lora_studio.py
• python -m pytest -q -p no:cacheprovider
• git diff --check

LoRA Studio v0.4.1 (Phase 11.6.16.2)

Release Version

v0.4.1

Internal Milestone

Phase 11.6.16.2

Highlights

• Added 25% / 50% / 75% / 100% image scale controls for Validation Matrix, Epoch Cross Matrix, and Review Matrix.
• Added click-to-open 100% image preview popup for Matrix images.
• Set Matrix default image display to 25% to make wide comparison screens easier to scan.
• Improved multi Validation Run image generation, embedding, and Machine Review assist flow from the Job detail page.
• Improved compact log display for completed Validation Run generation rows.

Notes

• Machine Review Assist is advisory only. Human visual review remains the final decision point.
• Existing generated static Matrix HTML files need to be regenerated to include the new scale controls.
• Dynamic Epoch Cross Matrix views reflect the new display controls immediately.

LoRA Studio v0.4.0 (Phase 11.6.16.1)

Release Version

v0.4.0

Internal Milestone

Phase 11.6.16.1

Highlights

• Added Review Session driven candidate epoch review workflow.
• Added Review Preparation pipeline for candidate epoch image generation, embeddings, Machine Review, and Review Matrix creation.
• Improved Project / Job / Review Session / Validation Run navigation and primary action design.
• Added Active Operation Monitor for running operations and short log tail display.
• Improved Review Matrix navigation and Review Session state-specific actions.
• Added OpenCLIP / transformers CLIP embedding provider groundwork and Machine Review Assist flow improvements.

Notes

• Machine Review Assist is advisory only. Human visual review remains the final decision point.
• Review Session is for pre-selection candidate epoch comparison.
• Validation Run remains the post-selection weight calibration workflow.

LoRA Studio v0.3.0-beta (Phase 11.5)

Release Version: v0.3.0-beta
Internal Milestone: Phase 11.5

LoRA Studio v0.3.0-beta is the Phase 11.5 operations beta release.

Highlights:

• Added the experimental OpenCLIP embedding provider (open_clip).
• Added ViT-B-32 / laion2b_s34b_b79k as the standard OpenCLIP model with 512-dimensional image embeddings.
• Allows OpenCLIP pretrained weight downloads only when allow_model_download=true is explicitly enabled.
• Uses CUDA first with CPU fallback; CPU runs use fp32.
• Keeps real embedding providers from starting while training or validation image generation is running.
• Keeps the mock provider for CI/tests and stores Machine Review Assist results by provider/model.
• Includes Phase 11.1-11.5 Reference Set, Embedding Cache, Machine Review Assist, transformers_clip, and OpenCLIP work.

Notes:

• Machine Review Assist is advisory only. Human review remains the final decision.
• WebUI/reForge API generation, ChatGPT API review, FLUX, and LyCORIS/LoCon are not included in this release.

LoRA Studio v0.2.0-beta (Phase 10.11.1)

Release Version: v0.2.0-beta
Internal Milestone: Phase 10.11.1

LoRA Studio v0.2.0-beta は、Phase 10.11.1 時点の運用ベータリリースです。

主な変更:

• sd-scripts gen_img.py を使ったValidation画像生成を安定化しました。
• Validation生成時のsampler表記をsd-scripts互換値へ正規化しました。
• 生成中ログの簡易表示とAJAX更新を改善しました。
• 生成完了後に画像生成ボタンが復帰しないUI問題を修正しました。
• Validation Run / Matrix / レビューまわりの操作導線を改善しました。
• READMEのOSS向けスクリーンショットと説明を整理しました。
• one-offのローカル移行処理をアプリ本体から分離しました。

注意:

• WebUI / reForge APIによる自動生成は未実装です。
• sd-scripts gen_img.py による標準Validation画像生成には対応しています。
• ChatGPT API連携、AI画像評価、FLUX、LyCORIS/LoConは未実装です。