Releases: daydreamlive/scope
Releases · daydreamlive/scope
v0.1.0-beta.1
This is the first beta release of Scope (on the way to the first stable v0.1.0 release) and introduces:
- A Windows desktop app with a simplified installation process (no need to use the CLI) for Windows users with a local GPU.
- Experimental support for using reference images and control videos (eg. depth, pose, scribble, optical flow, etc.) to guide generation with autoregressive video diffusion pipelines using a modified version of VACE. The current status of these features is tracked in the VACE guide.
What's Changed
- 01 VACE feat: add image control by @ryanontheinside in #256
- 02 VACE feat: Infrastructure by @ryanontheinside in #257
- 03 VACE feat: Longlive + Windows video support by @ryanontheinside in #258
- 04 VACE feat: stream diffusionv2 by @ryanontheinside in #259
- 05 VACE feat: server + frontend by @ryanontheinside in #260
- Initial VACE support fixes by @yondonfu in #269
- Add initial VACE support for RewardForcing by @yondonfu in #270
- Fix VACE model weights download by @leszko in #271
- Fix param change after the Download Dialog by @leszko in #272
- Download and load VACE module instead of full by @yondonfu in #273
- Validate resolution in UI by @leszko in #209
- Update VACE docs and a few misc fixes by @yondonfu in #274
- Bump to v0.1.0-beta.1 by @yondonfu in #275
Full Changelog: v0.1.0-alpha.12...v0.1.0-beta.1
Contributors
leszko, yondonfu, and ryanontheinside
Assets 4
v0.1.0-alpha.12
Contributors
yondonfu
Assets 4
v0.1.0-alpha.11
Contributors
yondonfu
Assets 4
v0.1.0-alpha.10
This is a pre-release for testing the new Windows desktop app.
What's Changed
- Electron App: Fix FPS/Bitrate not visible by @leszko in #246
- Improve log message when the model files are corrupted by @leszko in #249
- Include model download progress in log files by @yondonfu in #251
- Bump to v0.1.0-alpha.10 by @yondonfu in #252
Full Changelog: v0.1.0-alpha.9...v0.1.0-alpha.10
Contributors
leszko and yondonfu
Assets 4
v0.1.0-alpha.9
This is a pre-release for testing the new Windows desktop app.
What's Changed
- Add Electron Desktop App by @leszko in #212
- Build Docker image upon tag push by @leszko in #240
- Update frontend icon by @leszko in #239
- Electron App: Update LogViewer to read logs from ~/.daydream-scope/logs by @leszko in #242
- Bump to v0.1.0-alpha.9 by @yondonfu in #245
Full Changelog: v0.1.0a8...v0.1.0-alpha.9
Contributors
leszko and yondonfu
Assets 4
v0.1.0a8
This release introduces:
- Support for RewardForcing (text-to-video and video-to-video)
- Support for StreamDiffusionV2 text-to-video
- Support for LongLive video-to-video
- Support for Spout I/O (Windows only) for sharing real-time video with other local applications.
- Various bug fixes
What's Changed
- Fix Dockerfile by @yondonfu in #176
- Add quantization controls for streamdiffusionv2 and longlive by @yondonfu in #179
- Fix default cmd in Dockerfile by @yondonfu in #182
- Fix broken docs link for pipelines by @yondonfu in #185
- Remove Windows note for Krea + add note about vcredist on Windows by @yondonfu in #189
- Fix to AttributeError: 'ShapeAsConstantBuffer' dtype error with compiling by @BuffMcBigHuge in #192
- Fix: Warmup changes for krea to maximize startup performance. by @BuffMcBigHuge in #186
- Do not load sage on B200 for attention by @yondonfu in #196
- Add Docker image build "main" after every commit by @leszko in #198
- Update README.md about the main Docker image by @leszko in #199
- Add dynamic note for fp8_e4m3fn by @yondonfu in #201
- Improve LoRA Conversion + Fixes by @ryanontheinside in #171
- Ryanontheinside/feat/independent lora merge strategies by @ryanontheinside in #172
- 01 feat: add shared VAE infrastructure by @ryanontheinside in #203
- 02 feat: schema infrastructure by @ryanontheinside in #204
- 03 feat: migrate pipelines to MultiModePipeline architecture by @ryanontheinside in #205
- Add NoOp block to silence logs by @yondonfu in #215
- Skip latent normalization on encode for StreamDiffusionV2 by @yondonfu in #216
- 04 Universal Input Modes Front End by @ryanontheinside in #214
- Support fetching ICE servers + trickle ICE by @yondonfu in #224
- Enable running scope on non-gpu machines by @leszko in #211
- Add RewardForcing pipeline by @yondonfu in #226
- fix: use input frame rate for v2v consumption instead of production rate by @ryanontheinside in #219
- Default quantization to null when switching from krea by @yondonfu in #230
- Spout Send & Receive by @BuffMcBigHuge in #213
- Update visual layout of Spout Receiver/Sender by @yondonfu in #233
- Add LoRA guide link in UI by @yondonfu in #234
- Special case first generation for RewardForcing V2V by @yondonfu in #235
- Fix mode switch from text to video by @yondonfu in #237
- Spout guide + bump to v0.1.0a8 by @yondonfu in #238
Full Changelog: v0.1.0a7...v0.1.0a8
Contributors
leszko, yondonfu, and 2 other contributors
Assets 2
v0.1.0a7
This release introduces:
- Support for LoRAs which can be used to customize concepts and styles used in generations.
- A new modular pipeline design built on modular diffusers that allows the steps of pipelines to be represented as composable blocks that can be pieced together.
- Experimental support for using Scope as a library (Scope Core) or as a standalone server API (Scope Server) both of which will be accompanied by more documentation in future releases.
What's Changed
- Add offline generation section to pipeline usage docs by @yondonfu in #116
- fix: prevent zero-sized tensor in flex_attention padding removal by @ryanontheinside in #117
- Modularize pipelines by @leszko in #118
- Ryanontheinside/feat/LoRA by @ryanontheinside in #141
- Improve block input/output params by @leszko in #146
- Delete VOD pipeline by @leszko in #147
- Fix CleanKVCacheBlock by @leszko in #150
- Ryanontheinside/fix/perm merge by @ryanontheinside in #166
- Make pipeline status checks async by @yondonfu in #167
- Move uvicorn.access logging behind VERBOSE_LOGGING by @yondonfu in #168
- Refactor into scope.server + scope.core package structure by @yondonfu in #169
- Add Server API by @leszko in #158
- Add LoRA guide by @yondonfu in #174
- Bump to v0.1.0a7 by @yondonfu in #175
Full Changelog: v0.1.0a6...v0.1.0a7
Contributors
leszko, yondonfu, and ryanontheinside
Assets 2
v0.1.0a6
This release includes:
- Support for negative attention bias for Krea Realtime Video which helps mitigate error accumulation and prevent repetitive motion.
- Support for temporal transition (used for interpolating prompts over a time period) settings in timeline exports.
- Sage Attention 2 installation for Krea Realtime Video (see the docs for how different attention backends are used).
- Windows Triton installation which is a step towards getting more stable performance on Windows for Krea Realtime Video.
- A bug report button with the ability to copy logs to a GitHub issue.
- Various frontend bug fixes.
What's Changed
- Add Daydream Project links to usage docs by @yondonfu in #77
- Make CUDA versioning requirement more clear by @hthillman in #81
- Add python3-dev to Dockerfile by @leszko in #86
- fix: skip kernel import if windows by @ryanontheinside in #88
- Refactor PromptInput and TimelinePromptEditor by @leszko in #87
- Fix timeline rewind prompt issue by @leszko in #95
- Fix playhead movement timing issue by @leszko in #96
- Fix negative time bug in timeline by @leszko in #97
- Change pipeline error to a single Toast message by @leszko in #94
- feat(app): no browser flag by @ryanontheinside in #79
- Ryanontheinside/feat/temporal blend timeline support by @ryanontheinside in #76
- Add info about CUDA version for H100 by @leszko in #101
- Updates docs link to the Scope tutorial page by @viborc in #70
- Add Report Bug button with ability to copy logs by @yondonfu in #102
- Project Url -> Docs Url in pipeline card by @yondonfu in #103
- Align UI style for temporal controls in append vs editing mode by @yondonfu in #104
- Fix version in frontend + server by @yondonfu in #106
- feat: Add triton dependency for platform-specific support by @BuffMcBigHuge in #98
- Add sageattention dependency by @leszko in #68
- Ryanontheinside/experiments/attn bias by @ryanontheinside in #112
- Set default cache bias for krea to 0.3 by @yondonfu in #114
- Bump to v0.1.0a6 by @yondonfu in #115
New Contributors
- @hthillman made their first contribution in #81
- @viborc made their first contribution in #70
- @BuffMcBigHuge made their first contribution in #98
Full Changelog: v0.1.0a5...v0.1.0a6
Contributors
leszko, yondonfu, and 4 other contributors
Assets 2
v0.1.0a5
This release introduces:
- Support for Krea Realtime Video on GPUs with >= 32 GB (requires lower resolution and fp8 quantization on 5090) and Linux (Windows support is limited without guarantees right now for this model which will be addressed in a future release)
- Support for selectively downloading models for pipelines in the frontend
- Support for controlling transitions between prompts using temporal interpolation
What's Changed
- streamdiffusionv2 timeline usage examples by @ryanontheinside in #63
- Add Download models dialog by @leszko in #59
- Download streamdiffusionv2 from HF + cleanup model config logging + remove estimated download size by @yondonfu in #64
- Add krea-realtime-video by @yondonfu in #55
- Ryanontheinside/feat/temporal interpolation by @ryanontheinside in #71
- Add kernels + FA3, regional torch.compile to speed up krea-realtime-video by @yondonfu in #74
- Add krea-realtime-video usage docs by @yondonfu in #75
Full Changelog: v0.1.0a4...v0.1.0a5
Contributors
leszko, yondonfu, and ryanontheinside
Assets 2
v0.1.0a4
This release introduces:
- Pausable/resumable generation
- A timeline editor with support for exporting/importing settings and prompt sequences which allows generations to be replayed and modified by different users
- Prompt blending which allows multiple weighted prompts to be combined to steer generations
What's Changed
- Fix input/output video res sync bug + default pipeline list ordering by @yondonfu in #44
- Add pausable generation by @yondonfu in #43
- Ryanontheinside/feat/prompt blending fixes by @ryanontheinside in #51
- Add Timeline by @leszko in #54
- Ryanontheinside/fix/proportional prompt weights by @ryanontheinside in #58
- fix: Adjust visible time range when importing timeline by @yondonfu in #62
- Update longlive docs with timeline examples by @yondonfu in #61
Full Changelog: v0.1.0a3...v0.1.0a4
Contributors
leszko, yondonfu, and ryanontheinside