Merge pull request #63 from some1galaxy/develop

Promote High Quality Video Generation with a full-width banner (#62)

Merge pull request #59 from some1galaxy/develop

Merge pull request #58 from some1galaxy/fix/lower-default-guidance-scale

Lower default guidance scale from 2.5 to 1.2

Merge pull request #57 from some1galaxy/fix/enable-negative-prompt-cfg

Fix negative prompt having no effect at default guidance scale

Merge pull request #56 from some1galaxy/develop

Promote High Quality Video Generation space instead of face swap (#55)

Split HF_TOKEN into per-environment secrets (#54)

Sync develop into main (#52)

Merge main into develop, resolving PR #52 sync conflict

Tune default gpu_duration down: fast 60->30s, high_detail 75->50s (#51)

Remove _COLD_START_BUFFER_S: gpu_duration is now the full declared duration (#49)

Fix stale log wording: text_encoder offload-skip no longer says .to() is unsupported (#47)

Fix GPU task aborted: stop passing pipe as a @spaces.GPU argument (#45)

Spike: load int8 text_encoder at module level instead of per-request (#43)

Thread pipe as an explicit parameter instead of a global in inference.py (#41)

Remove the bf16 text_encoder fallback: int8 is now a hard requirement (#39)

Construct pipe explicitly in inference.py instead of at model_loading import (#38)

Introduce a two-stage develop/main deploy flow (#27) (#35)

Drop text_encoder's unused lm_head to free ~1GB (~0.5GB int8) (#34)

Introduce a two-stage develop/main deploy flow (#27)

Extract _InferTimer into a generic Timer in timing.py (#32)

Rename dev-en's repo variable from HF_SPACE_ID to HF_SPACE_ID_DEV_EN (#30)

Centralize print_*() logging functions in logging_utils.py (#29)

Set up develop/main two-stage deploy: retarget dev-en to develop, add prod-en workflow

Extract print() calls into named print_*() functions (#23)

Clean up experimental logging from the module-level .to('cuda') investigation (#25)

Use CPU-only torch/torchvision wheels for the mypy CI job (#22)

Fix OOM: stop double-allocating text_encoder (bf16 + int8) on GPU (#24)

Experiment: load transformer/pipeline via module-level .to('cuda') (#21)

Enforce strict typing on Python code with mypy (#20)

Split app.py into cohesive modules (#17)

Merge pull request #18 from some1galaxy/ci/deploy-dev-en-on-merge

Drop hardcoded Space id from the error message

Fail loudly instead of defaulting HF_SPACE_ID

Make the target HF Space configurable via a repo variable

Add GitHub Actions workflow to deploy to dev-en Space on merge to main

Add copy-prompt button and improve mobile touch targets/layout

Convert example images 2.jpg, 3.jpeg, 4.jpg, 5.jpg to actual Git LFS objects

Add notice that the Space may end without prior notice

Reduce default GPU duration for High Detail mode from 60s to 30s

Pad declared ZeroGPU duration with a cold-start buffer

Sync visible GPU duration slider default to Fast mode's 15s

Fix step-done log ordering vs text_encoder offload/pre-VAE-decode

Reduce default GPU duration for Fast mode from 30s to 15s

Add VAE decode logging: pre-decode mem snapshot, tiling prediction, peak reset

Raise VAE tile threshold to HIGH_DETAIL.max_dim (2048px) to skip tiling

Bump bitsandbytes to 0.49.2 for CUDA 13.0 support