5e4ed9c9bb
fix ltx2
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-12 17:12:13 +01:00
280da2c10b
cleanup
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-12 16:14:32 +01:00
1cdb46c455
Merge pull request #4538 from CalamitousFelicitousness/state-cleanup-patch
...
fix(control): ensure state cleanup runs reliably
2026-01-12 16:12:27 +01:00
ca3aecef2d
Revert: btn_generate.click call removal
2026-01-12 14:12:44 +00:00
9f6e34371f
revert: restore state tracking in control/run.py
2026-01-12 14:12:44 +00:00
2fd52ae230
fix(control): ensure state cleanup runs reliably
...
Fix control tab getting stuck in running state
after generation completes.
2026-01-12 14:12:44 +00:00
e4e863fd6d
lint and update diffusers
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-12 12:31:29 +01:00
1f92d7c24c
Merge pull request #4544 from CalamitousFelicitousness/feat/prompt-enhance-update
...
Prompt Enhance Update
2026-01-12 12:00:26 +01:00
86744ebd76
feat(prompt-enhance): add Top-K and Top-P sampler controls
...
Add Top-K and Top-P sampling parameters to match VQA capabilities:
- Top-K: 0-100 (default 0 = disabled)
- Top-P: 0.0-1.0 (default 0 = disabled)
Parameters are only passed to the model when > 0, matching VQA behavior.
2026-01-12 00:17:20 +00:00
e881faa2fb
feat(vlm): align default generation settings
...
Update default settings for both Prompt Enhance and VQA:
- max_tokens: 512 (prompt enhance)
- do_sample: True
- temperature: 0.8
These defaults provide better output quality with more creative
variation while maintaining coherent results.
2026-01-12 00:17:20 +00:00
0659759e90
fix(vqa): improve unload logging consistency
...
Add before/after debug messages when unloading VQA model to match
the pattern used in prompt enhance for better debugging visibility.
2026-01-12 00:17:20 +00:00
de1d520929
fix(prompt-enhance): align with VQA approach and fix multiple issues
...
Thinking mode alignment with VQA:
- Expand is_thinking_model() detection (moondream, mimo patterns)
- Use manual </think> tag approach instead of enable_thinking parameter
- Handle prefill via string concatenation, not assistant message
Bug fixes:
- Fix max_new_tokens (was input_len + tokens, now just tokens)
- Fix vision toggle not disabling vision (wrap image processing in check)
- Fix VL processor error when vision disabled (explicit text= parameter)
- Add error handling when vision enabled but no image provided
Debug logging consistency with VQA:
- Add model_class to entry debug
- Add input_ids_shape to tokenization debug
- Rename raw -> response_before_clean, response -> response_after_clean
UI improvements:
- Make Enhanced prompt textarea scrollable (max_lines=12)
- Remove redundant mode variable from logging
2026-01-12 00:17:20 +00:00
d1a08ec801
fix(prompt-enhance): add Kanvas image support and debug logging
...
Add fallback to ui_control_helpers.input_source when Gradio image
component is empty, which occurs when Kanvas is active (replaces
gr.Image with gr.HTML).
Also standardize debug logging:
- Add debug at enhance() entry, chat template, tokenization
- Change shared.log.trace() calls to use debug_log() for consistency
- All debug output controlled by SD_LLM_DEBUG environment variable
2026-01-12 00:17:20 +00:00
5f1fbbaeca
ui(prompt): add descriptive placeholders to text fields
...
Clarify purpose of empty fields with placeholder text:
- System prompt: explains built-in instructions used when empty
- Prefix/suffix: describe where text is added to result
- Output: indicates where enhanced prompt appears
2026-01-12 00:17:20 +00:00
2b8ecf01e2
docs(i18n): improve prompt enhance hints
...
Improve formatting and clarity of prompt enhance tooltips:
- Add spacing with <br><br> for better readability
- Clarify that prefix/suffix bypass LLM rewriting
2026-01-12 00:17:20 +00:00
7c8f7fd296
feat(prompt): add prefill and thinking mode handling
...
Add prefill text support and proper thinking tag handling:
- Prefill text input to pre-fill start of model response
- Keep Prefill toggle to retain/remove prefill from output
- Keep Thinking Trace toggle to format <think> tags as readable output
- Proper <think> tag cleanup: strip or format as "Reasoning:/Answer:"
- Support for prefill + thinking combination via keep_think_block_open()
Chat template uses continue_final_message=True when prefill provided,
allowing model to continue from the prefill text naturally.
2026-01-12 00:17:20 +00:00
a77173881f
feat(prompt): add vision/reasoning symbols and vision toggle
...
Add visual capability indicators and user control for image input:
- Vision symbol (eye icon) for VL-capable models in dropdown
- Reasoning symbol (lightbulb) for thinking-capable models
- "Use vision" checkbox to control image input for enhancement
- Toggle dims and unchecks when non-VL model selected
- Vision auto-enables when switching to VL model
Also:
- Rename "Do sample" to "Use samplers" for consistency with VQA
- Add tooltips/hints for all prompt enhance UI elements
- Add CSS styling for dimmed vision toggle appearance
2026-01-12 00:17:20 +00:00
c1f9646eaa
Merge pull request #4541 from awsr/save-update
...
Update config saving
2026-01-11 13:05:29 +01:00
418134b820
Update default filename handling
2026-01-10 14:53:07 -08:00
7e947e8150
Update saving
2026-01-10 13:58:13 -08:00
0faabffc14
Simplify options init/save/load
2026-01-10 13:27:38 -08:00
15ba025044
Merge branch 'dev' of https://github.com/vladmandic/sdnext into dev
2026-01-10 20:50:50 +01:00
f5f4c60820
recommend restart on update
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 20:50:48 +01:00
fa2917a6be
cleanup
2026-01-10 20:48:36 +03:00
f4d87d6cdf
fix ltx2 fp64 on alchemist
2026-01-10 20:46:03 +03:00
a6a5b9ff54
ltx image cond
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 12:56:42 +01:00
ffe1e2a861
cleanup
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 11:32:32 +01:00
85332594fc
triton test reduce verbosity
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 10:32:13 +01:00
2cba0d8fe0
explictly disable hires for detailer pass
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 10:26:51 +01:00
b386675c06
ltx-2 prototype
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 09:31:59 +01:00
641ba05d15
add nunchaku-z-image-turbo
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 09:09:45 +01:00
9261b65beb
add Qwen-Image-2512-SDNQ-4bit-dynamic
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 08:47:28 +01:00
ec007a3018
add laplace beta scheduler
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-10 08:41:27 +01:00
3dac59fd45
Merge pull request #4537 from CalamitousFelicitousness/genai-auth-fix
...
Google GenAI Auth Fix
2026-01-10 08:19:54 +01:00
1aab61aa47
Cleanup
2026-01-10 03:08:22 +00:00
3302522fdb
Cleanup
2026-01-10 03:07:44 +00:00
d7e5c97c89
fix lora loader with torch-no-distributed
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-09 14:32:31 +01:00
9fe9d9521c
fix(cloud): support three Google GenAI auth modes, use UI settings only
2026-01-09 00:30:55 +00:00
47dcab3522
update sdnq
2026-01-09 00:34:32 +03:00
3bb6e5b47c
fix kandinsky-5
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 21:08:19 +01:00
81d0d33573
update todo
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 09:36:03 +01:00
49e09e625c
update diffusers
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 09:27:49 +01:00
61905f5d85
update kanvas
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 09:18:50 +01:00
899514c5b8
lint update
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 08:51:35 +01:00
7fe09db97e
cleanup offload calls
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 08:35:25 +01:00
a36cd157ec
force move to device if offload=none
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-08 08:18:41 +01:00
71df7e7ae2
Merge pull request #4534 from ryanmeador/fix_4526
...
Fix Wan missing frames on first generation
2026-01-08 08:09:41 +01:00
c1c4d6e0ee
prevent accelerate from recreating keys such as and so that the VAE decode will progress. resolves #4526
2026-01-07 22:19:11 -05:00
8d03d7c5b2
hidiffusion tracing
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-07 15:27:14 +01:00
1a39b82fea
switch processing class not restoring params
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-07 11:35:13 +01:00
vladmandic
CalamitousFelicitousness
awsr
Disty0
Ryan Meador