Commit Graph
232 Commits
Author SHA1 Message Date
Towsty 7799529788 Extend Prompt Builder with sequence characters, outfits, and scene toggles.
Characters live on the sequence with per-scene on/off and outfit picks; Copy scene exports only the active tab; legacy saves still open.
2026-10-02 08:50:31 -05:00
Towsty 1bf97568bf Add Prompt Builder page for multi-scene prompt sequences.
Name, save, and load ordered scenes on the server; assemble copyable text with a Batch-compatible double blank line between scenes.
2026-10-02 07:15:00 -05:00
Towsty 14dcd234f3 Let finished caption jobs be dismissed from the Queue panel. 2026-09-30 23:18:44 -05:00
Towsty 15ef891b26 Add image-to-text Describe on the bench via exclusive llama.cpp caption jobs. 2026-09-30 22:50:19 -05:00
TowstyandCursor 9481791ac5 Keep fetching library pages after the first 50 so the whole folder is reachable.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-30 00:28:22 -05:00
TowstyandCursor 17bb42bea4 Serve the stage from a 1600px preview instead of the grid thumb.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 20:57:01 -05:00
TowstyandCursor 4ec8ffa8d6 Serve the bench before the catalog and never await ffmpeg on /thumb.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 20:35:28 -05:00
TowstyandCursor a465dd68ec Stop polling the library catalog and keep the stage on thumbs.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 20:15:03 -05:00
TowstyandCursor cfabfd37ee Queue Generate with an immediate local card and stop blocking reload on library or object_info.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 19:46:55 -05:00
TowstyandCursor aa0aa871e4 Add paginated library JSON and dedicated still thumbs that never stream archive masters.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 19:45:22 -05:00
TowstyandCursor 5f6a7aa9bc Rename library actions and always store promptRaw on ingest.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-29 08:14:13 -05:00
TowstyandCursor c4e74628ee Show the Qwen LoRA card with qwen-stem picks and apply them on generate.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-28 22:47:59 -05:00
TowstyandCursor 50319e03a9 Map Photos roles to Qwen image_1/image_2 and inject encode tags silently.
Mention phrases become <imageN> only in the TextEncode string; the inspector never shows Comfy tokens.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 16:01:49 -05:00
TowstyandCursor 58c5458c87 Store the Edit TextEncode prompt on library stills, not a Start still caption.
Edit + Enhance always leads with the keep/<image1> stanza and typed change; PE describe dumps are dropped, and Typed stays promptRaw.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 15:14:20 -05:00
TowstyandCursor a51118d92a Pass through PE refusals to the raw prompt and pin adult PE system prompts.
Enhance no longer fails the studio job when the rewrite refuses or comes back empty; host prompt copies keep the official steps plus the adult appendix after a node re-clone.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 14:46:24 -05:00
TowstyandCursor 99b61e67d7 Keep Enhance + Generate as one queue row with shared progress.
The PE stage now stays on the same jobId so the header and queue badge agree, and jobs behind it do not promote when enhance finishes.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 14:42:22 -05:00
TowstyandCursor 30be4bd5b0 Wire Qwen Edit Start still to image_1 and PE-I2I identity keep.
Edit no longer invents strangers when the prompt omits <image1> or PE drops image tokens; Hero stays Klein-only.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 14:27:36 -05:00
TowstyandCursor ecc4c69a8b Wire Qwen Turbo toggle through Studio 2 and the runner.
Force 6/1 when Turbo is on, skip negatives, keep PE in front, and fail closed if the r128 LoRA is missing.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 12:42:00 -05:00
TowstyandCursor 17cbc99f8f Add Qwen 2.1 Viggle Turbo Comfy graphs.
Use the unmerged ViggleTurboLora path with the official 6-step sigma schedule on the existing Q8 GGUF.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 12:41:59 -05:00
TowstyandCursor 95873744cd Size Qwen 2.1 Generate from the bench aspect table.
Wire EmptyLatentImage to the official ~1 MP grid so 16:9 lands at 1536×864, keep PE wh_ratio advisory, and leave Edit on the encode latent from image_1.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 10:54:18 -05:00
TowstyandCursor 2842855c7b Wire Qwen Enhance prompt toggle through Studio 2 and the runner.
Queue PE first when enabled, fail closed on parse_ok, free VRAM, then run the existing T2I or Edit graph with the rewrite.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 10:25:44 -05:00
TowstyandCursor e0292c8d2a Add Qwen 2.1 PE-only Comfy graphs for prompt enhance.
Keep PE CLIP separate from DiT so the 16 GB card can run rewrite then generate without loading both at once.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-26 10:25:44 -05:00
TowstyandCursor bb27af506e Restore Studio 2 Load hydrate, sticky negatives, and strip actions.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-21 12:23:20 -05:00
TowstyandCursor 52a01e0adc Enable Qwen 2.1 on Edit: wire picker, runner, and source-still validation.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 18:13:07 -05:00
TowstyandCursor 8609ba24a4 Add Qwen 2.1 Image Edit Comfy graph (GGUF + TextEncode + int8 cache).
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 18:10:50 -05:00
TowstyandCursor 776d585345 Fix xAIGen LoRA picker to use isXaigenStudio like the server.
Bench was checking only public.studio === 'xaigen', so when the instance
is branded xAIGen via instanceName the client stripped xaigen-* LoRAs.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 17:34:53 -05:00
TowstyandCursor 608f7d8d47 Fix Qwen 2.1 GGUF load: promote Q8 norms to F32 and wire TextEncode latent.
abenzerps Q8_0 ships 1D RMSNorms as packed Q8 (136 vs 128), which breaks
Comfy rms_rope; tagger now dequantizes small tensors and the graph uses
TextEncodeQwenImage21's 64-ch latent plus AuraFlow shift.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 16:58:00 -05:00
TowstyandCursor 68731916ca Wire Qwen 2.1 as Generate-only engine with 25/1 sampler defaults.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 16:58:00 -05:00
TowstyandCursor 08d507c491 Add studio2_qwen21_t2i Comfy graph for GGUF Qwen Image 2.1 T2I.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 16:57:04 -05:00
TowstyandCursor b194b3cefa Always live-scan Studio 2 LoRAs on load and Refresh like Settings.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-18 12:59:06 -05:00
TowstyandCursor 7221b6a8cf Refresh Studio 2 LoRAs from live Comfy object_info on fresh=1.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-18 12:45:48 -05:00
TowstyandCursor 36d24feb9c Allow Studio 2 MiniMax text-to-video with empty start when locks are off.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-16 14:56:16 -05:00
TowstyandCursor 5694104226 Keep YuE2 jobs live on the bench and wire target length to semantic tokens.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 20:53:07 -05:00
TowstyandCursor 2ebc3066e6 Reject instrumental music edits and force YuE2 on queued track patches.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:20:11 -05:00
TowstyandCursor c868605d8a Drop ACE and YuEGP defaults from studio music queue and library labels.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:19:46 -05:00
TowstyandCursor 1db62e9187 Make Studio 2 the home bench with YuE2 music, LTX, and per-preview hide.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:19:34 -05:00
TowstyandCursor a216afc4ce Remove YuEGP handlers from the Comfy host agent.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:17:59 -05:00
TowstyandCursor cce1faa9b1 Route all music jobs through YuE2 only.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:16:45 -05:00
TowstyandCursor c50bde723f Accept only YuE2 in the music generate API.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 18:16:45 -05:00
TowstyandCursor 62671d05da Allow multi-section YuE2 lyrics and force non-flash attention on Windows.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-14 19:28:43 -05:00
TowstyandCursor 71a61b1765 Add YuE2 to the music engine picker and generate API.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-14 18:46:35 -05:00
TowstyandCursor ec5791c595 Mount YuE2 on the host agent and share GPU ownership.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-14 18:46:35 -05:00
TowstyandCursor 5f68205f01 Route engine yue2 through a standalone job helper.
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-14 18:45:57 -05:00
Towsty 064f1db933 Read production LoRA names in Studio 2 2026-09-12 21:05:11 -05:00
Towsty 401e6c26a2 Remember Studio 2 LoRA stacks for each engine 2026-09-12 20:17:58 -05:00
Towsty 7bb4b16c1e Sample Klein hero identity from image latents 2026-09-12 11:46:56 -05:00
Towsty 3da3656140 Show pending Studio 2 jobs and freeze finished timers 2026-09-12 10:39:33 -05:00
Towsty 3e90d2ff45 Show clear GPU status and move upscale progress into the queue 2026-09-12 10:17:03 -05:00
Towsty a9551911f2 Encode Unicode video titles in playback response headers 2026-09-12 10:17:02 -05:00
TowstyandCursor d8f42ecc68 Tighten needsFit to exact canvas size, guard against empty renders
needsFit only compared aspect ratio, so a same-aspect-but-larger still
(e.g. a 4032x2268 phone photo against a 1344x768 canvas) would skip
fitting and hand MiniMaxH3AddGuide a still that does not match the
job's own latent size. Require an exact width+height match instead;
anything else (including unreadable dimensions) still fits.

Also guard the video save path: an empty Comfy download, an empty
stitched assembly, or a stitched result with no readable duration now
fails the job with a clear error instead of silently saving/attaching
a broken zero-length clip.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-11 23:11:02 -05:00