Commit Graph
1427 Commits
Author SHA1 Message Date
Jukka Seppänen 309491b269 Merge pull request #1908 from jjdejong/patch-1
Implement conditional offloading for diffusion model
2026-02-16 15:35:45 +02:00
kijai 06122f1e9d Rope offset should be disabled by default 2026-02-16 15:32:54 +02:00
kijai 5d36631795 I2V cross_attn fixes 2026-02-16 15:31:20 +02:00
kijai 3d7b49e2df Move string_to_seed import 2026-02-02 17:02:05 +02:00
kijai e091c4a774 version 1.4.7 2026-02-02 01:43:38 +02:00
kijai 60a387579e Update wanvideo_2_1_14B_I2V_SkyReelsV3_TalkingAvatar_example_01.json 2026-02-01 15:16:33 +02:00
kijai 9ae1a4f2c9 Update multitalk_loop.py 2026-02-01 15:15:37 +02:00
kijai f55b7b3d89 Update wanvideo_2_1_14B_I2V_SkyReelsV3_TalkingAvatar_example_01.json 2026-02-01 15:15:33 +02:00
kijai e4e7f413f7 Make compatible with latest ComfyUI version 2026-02-01 15:15:27 +02:00
kijai e21fe20a4d Support SkyReels TalkingAvatar (A2V) 2026-01-31 23:24:29 +02:00
kijai 2c5a04cc63 Init image_cond_mask 2026-01-30 18:59:41 +02:00
kijai d00abe52d7 version 1.4.6 2026-01-28 02:09:50 +02:00
kijai 2952f5d9dd Show T5 loading progress bar 2026-01-26 15:53:48 +02:00
kijai 339e0fec81 Make NAG inplace application optional 2026-01-23 18:35:30 +02:00
kijai 2c2a6e1889 RoPE frequency offset option for storymem
I'm not 100% sure on this, I initially tested this when I noticed the original code doesn't, but it's described in the paper... now I see the original code has added it too so it seems to be the intended way to use it.
2026-01-23 14:31:04 +02:00
kijai 8640bfad52 VRAM optimizations
Minor for almost everything, major for multitalk when using masks
2026-01-22 19:46:52 +02:00
kijai 3b4a711a40 Reduce NAG memory usage 2026-01-22 17:43:34 +02:00
Jean J. de Jong b99eac73da Implement conditional offloading for diffusion model
Add condition to skip offloading if load_device is main_device.
2026-01-21 17:33:58 +01:00
kijai 9f9a8e71c2 Revert " Fix: Fix torch.arange bounds error in context window processing"
This reverts commit 707bbcd72b.
2026-01-17 20:13:43 +02:00
Jukka Seppänen f2e2e2550a Merge pull request #1877 from YFJack/patch-1
Fix: Fix torch.arange bounds error in context window processing
2026-01-15 17:06:31 +02:00
Yifu Wang 707bbcd72b Fix: Fix torch.arange bounds error in context window processing
Fixed "upper bound and lower bound inconsistent with step sign" error
       when using WanVideoContextOptions with pose latents. Changed end index
       from c[-1] to c[-1] + 1 to properly include the last frame in the range.
2026-01-09 11:36:53 +08:00
Jukka Seppänen 855d103ee6 Merge pull request #1868 from vantagewithai/main
longcat avatar GGUF support.
2026-01-08 13:36:43 +02:00
kijai 64191921d4 Squashed commit of the following:
commit fdb23dec7d
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Jan 5 22:11:04 2026 +0200

    Update model.py

commit 07d7d8ca8e
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Jan 5 22:10:02 2026 +0200

    remove prints

commit 01869d4bf5
Merge: 55c6720 bf1d77f
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Jan 5 18:47:48 2026 +0200

    Merge branch 'main' into longvie2

commit 55c672028b
Merge: b551ec9 be41f67
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 29 15:39:43 2025 +0200

    Merge branch 'main' into longvie2

commit b551ec9e31
Merge: 9f019d7 19bcee6
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 29 15:03:53 2025 +0200

    Merge branch 'main' into longvie2

commit 9f019d7dfb
Merge: fc5322f c5d3fb4
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 23:40:25 2025 +0200

    Merge branch 'main' into longvie2

commit fc5322fae4
Merge: 222fc70 e75f814
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 22:04:15 2025 +0200

    Merge branch 'main' into longvie2

commit 222fc70eb7
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 17:18:55 2025 +0200

    Update nodes.py

commit 8509236da1
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 14:20:18 2025 +0200

    init
2026-01-05 22:11:20 +02:00
Vantage with AI 576f065073 Update tensor name condition for multitalk audio proj, to support longcat video avatar GGUF 2026-01-04 21:13:54 +05:30
kijai bf1d77fe15 Update wan_video_vae.py 2025-12-31 15:56:10 +02:00
kijai cced7fefbe Make VAE memory reporting optional to reduce log spam, and other logging updates 2025-12-31 15:46:40 +02:00
kijai 36bb0c73ee Correct infinite talk loop log entry
Just visual bug, no effect on anything else
2025-12-31 14:50:27 +02:00
kijai b982b4ef0c Update nodes_model_loading.py 2025-12-30 02:55:51 +02:00
kijai 3a7100bc39 Add node for UltraVico params 2025-12-30 02:53:38 +02:00
kijai be41f67fae Fix res_multistep steps 2025-12-29 15:39:27 +02:00
kijai 351829a2b6 Fix er_sde steps 2025-12-29 15:36:43 +02:00
kijai 19bcee67ed remove print 2025-12-29 02:50:04 +02:00
kijai 2fe4834178 Adjust ultravico frame_tokens 2025-12-29 02:48:59 +02:00
kijai 486564060f Add node to set attention mode per step and/or blocks 2025-12-29 01:35:32 +02:00
kijai 3730ccf603 Fix s2v 2025-12-28 10:34:13 +02:00
kijai 1eab022bb0 Update nodes.py 2025-12-28 02:35:42 +02:00
kijai 6fd4c6640c Adjust scheduler graph drawing 2025-12-28 01:27:18 +02:00
kijai 7a6efc1456 Add node for SVI 2.0 Pro 2025-12-28 00:35:14 +02:00
kijai e855726f10 Fix enhance-a-video 2025-12-26 23:12:18 +02:00
kijai a896101ec8 I don't know why this suddenly errors 2025-12-26 22:28:04 +02:00
kijai fd818faa08 Fix context window ref latent device 2025-12-26 16:09:01 +02:00
kijai b132a82f7a Fix LongCat-Avatar audio padding when not enough audio provided for given window 2025-12-26 15:57:30 +02:00
kijai 4b709a7a04 Cleanup example_workflows folder some 2025-12-26 15:44:08 +02:00
kijai 220aac2771 Update LongCatAvatar_audio_image_to_video_example_01.json 2025-12-26 13:52:27 +02:00
kijai 027bed8c3d Cleanup 2025-12-26 13:51:05 +02:00
kijai b2c520ca44 Fix s2v 2025-12-26 12:55:13 +02:00
kijai 20942b8fd9 Cleanup: Remove flowedit code, restructure some other code
Due to lack of use and code maintainability
2025-12-26 12:51:22 +02:00
kijai 74f337e06c Adjust StoryMem lora scaling
This was probably too high afterall
2025-12-26 01:41:32 +02:00
kijai 264212dddb Fix uncond variable name when using zero star or fresca 2025-12-26 01:27:48 +02:00
kijai f988d19fdb StoryMem latents are supposed to be encoded one by one 2025-12-25 20:44:37 +02:00