Skip to content

Claude/friendly johnson nfbrru - #1728

Open
lotus0510 wants to merge 5 commits into
Nerogar:masterfrom
lotus0510:claude/friendly-johnson-nfbrru
Open

lotus0510 wants to merge 5 commits into
Nerogar:masterfrom
lotus0510:claude/friendly-johnson-nfbrru

Conversation

@lotus0510

Copy link
Copy Markdown

Summary

Test plan

  • pre-commit run --all-files passes
  • Launched the affected UI or script and exercised the change
  • Tested with at least one real preset / config when relevant (note which: ____)

AI assistance

  • No AI involvement
  • AI-assisted — I have read every line in this diff and can defend each change
  • Early AI prototype — opened for discussion, not ready for review

lotus0510 and others added 5 commits October 2, 2026 16:15
- Add modules/util/i18n.py with a t() lookup and persisted language setting
- Add resources/i18n/zh_TW.json translations
- Add scripts/i18n_extract.py to collect translatable strings
- Wrap UI strings in t() across the PySide6 views and components
- Add a language selector to the top bar (applies after restart)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New simple training UI (start-ui-simple.bat / scripts/train_ui_simple.py): pick a
  preset, an image folder and an output name; everything else comes from the preset.
  Shows image/caption counts, estimated total steps, optional preview prompt,
  restore defaults, and an "advanced mode" export to the full UI.
- Anima LoRA starts from the model card values (rank 32, alpha 32, lr 2e-5) in the
  simple UI; the built-in preset is ~10x weaker.
- Wheel scrolling no longer changes combo boxes / spin boxes in the simple UI.
- Show Hugging Face download progress in the status bar while the model loads.
- Anima loader accepts ComfyUI-layout checkpoints (model.diffusion_model.*) as the
  transformer override and loads the checkpoint's own LLM adapter.
- Translations for the new UI and trainer status messages; README section on the fork.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The progress bar now shows step/total and percentage, and a new line
below the status shows the step within the epoch, speed (s/it or it/s),
elapsed time, current and smoothed loss, and the estimated finish time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018jKB4UUBvrBsQjEBw9kkYt
Preview images sampled during training now appear in a panel above the
progress bar, with epoch/step/time captions, buttons to browse earlier
previews of the run, a 'generate preview now' button and a shortcut to
the samples folder.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018jKB4UUBvrBsQjEBw9kkYt
Prints the stored training settings, how much the LoRA changed each layer
and caption statistics of the training images (missing trigger word,
character traits in every caption).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018jKB4UUBvrBsQjEBw9kkYt
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants